# Cogneris — Full Documentation > Generated from the tracked public sources listed in llms.txt. Do not edit this file directly. ## Source: products.html [Skip to content](#main) Cogneris platform # One document lifecycle. Seven modules. Collect documents, classify and extract their data, validate exceptions, route human decisions, act in downstream systems, and retain evidence — with each capability labeled by how it is available. Platform modules ## Seven modules. A connected lifecycle. Contract-configured capabilities require tenant enablement or configuration within the agreed scope. Confirm deployment and availability for your workflow with our team. Contract-configured ### Portal Collect customer documents through a tenant-scoped upload experience and connect each submission to its workflow. Requires tenant enablement and a configured workflow under the agreed contract. [See portal workflows](document-automation-portal.html) [Read portal documentation](https://docs.cogneris.ai/manuals/user/portal/overview/) Contract-configured ### Workflow Define ordered processing phases, pause a run for configured manual review, record the decision, and continue the workflow. Requires tenant enablement and configured review phases under the agreed contract. Confidence scores do not automatically select the review path. [See document workflows](email-document-workflow-automation.html) [Read workflow documentation](https://docs.cogneris.ai/docs/developers/mcp-server/) Contract-configured ### Vision Classify documents and extract structured fields with confidence and source references for downstream validation. Requires tenant configuration; confidence and source references depend on the configured extraction output and are not guaranteed for every field. [See document extraction](document-extraction.html) [Read extraction documentation](https://docs.cogneris.ai/docs/api/document-extraction/) Contract-configured ### ECM Organize documents, versions, tags, access rules, search, and lifecycle events in the document record. Requires tenant enablement and configured access and lifecycle policies under the agreed contract. [ECM scope and availability](products.html#module-ecm) Contract-configured ### Sign Create signing envelopes, collect signer events, and validate the resulting signed document through the configured signing flow. Requires tenant enablement and a separately confirmed validation flow under the agreed contract. This contract-configured scope does not establish a public document-validation service or third-party trust-service certification. [Sign scope and availability](products.html#module-sign) Contract-configured ### Templates Manage versioned DOCX templates with named placeholders and render approved template versions to PDF. Requires tenant enablement, a published template version and a configured PDF renderer under the agreed contract. Word add-in availability requires separate deployment confirmation. [Templates scope and availability](products.html#module-templates) Contract-configured ### Governance & Integrations Connect the lifecycle through REST APIs and signed webhooks while retaining contract-specific audit and security evidence. Requires tenant configuration, configured webhook signing and controls agreed in the contract. Client SDKs are published for TypeScript, Python, .NET and Java; no packaged connector catalogue is claimed. [Explore the REST API](api.html) [Read webhook documentation](https://docs.cogneris.ai/docs/api/) [Review trust and security](trust.html) Document lifecycle ## From collection to evidence. Configure the stages your workflow requires, with review phases and downstream actions agreed for your tenant. - CollectPortal, API, or workflow intake. - ClassifyIdentify the document type against tenant configuration. - ExtractReturn structured fields with source evidence. - ValidateApply schema and business-rule checks. - ReviewUse an explicitly configured manual-review phase where required. - DecideApprove, reject, or request correction according to workflow rules. - ActDeliver through REST APIs, signed webhooks, or configured downstream steps. - ProveRetain source references and contract-specific audit evidence. Customer journeys ## One lifecycle, three document journeys. These maps describe contract-configured workflows. Every stage requires tenant enablement and agreed configuration; linked guides describe the intended use case, not confirmation that a capability is deployed for your tenant. ### Accounts payable Contract-configured journey: from supplier invoice intake to a recorded payment decision and an agreed downstream handoff. Business approval remains with your authorized team. - CollectCollect supplier invoices through the tenant-scoped portal in the agreed intake workflow. Requires tenant enablement.[Invoice intake portal](document-automation-portal.html) - ClassifyIdentify invoices and supporting documents against agreed document types. Requires tenant configuration.[Classification configuration](https://docs.cogneris.ai/docs/api/document-classifier/) - ExtractExtract invoice fields for the agreed schema. Requires tenant configuration; source references depend on the configured output.[Invoice extraction](extract-invoices.html) - ValidateDefine the invoice comparisons and reference data needed for matching, subject to confirmation of the agreed checks. Requires tenant configuration.[Invoice matching scope](invoice-matching-automation-api.html) - ReviewPause for configured manual review for the tenant under the agreed contract. Requires tenant configuration. Authorized reviewers record approval, rejection, or a correction request.[Configured review workflow](email-document-workflow-automation.html) - DecideRecord the authorized team's invoice disposition in the workflow. Requires tenant configuration; a recorded decision does not execute a payment.[Accounts payable workflow](accounts-payable-automation-api.html) - ActDeliver the agreed event to a downstream endpoint through signed webhooks, with webhook signing and receiving-system behavior confirmed. Requires tenant configuration.[Webhook handoff](https://docs.cogneris.ai/docs/api/) - ProveAgree which source references, review decisions, and events support the invoice record under contract-specific retention controls. Requires tenant configuration.[Evidence and trust scope](trust.html) ### KYC and KYB onboarding Contract-configured journey: collect identity and company documents, review the agreed checks, and record an onboarding decision. Your authorized team defines acceptance policy. - CollectRequest identity and company documents through a tenant-scoped portal in the agreed onboarding workflow. Requires tenant enablement.[KYC document portal](kyc-document-portal.html) - ClassifyIdentify submitted document types against the onboarding configuration. Requires tenant configuration.[Classification configuration](https://docs.cogneris.ai/docs/api/document-classifier/) - ExtractExtract the identity and company fields agreed for the document set. Requires tenant configuration; field coverage and source references depend on the configured output.[KYC document extraction](extract-kyc-documents.html) - ValidateDefine required fields and cross-document checks for onboarding, subject to confirmation of supported checks. Requires tenant configuration; a document check alone does not establish regulatory compliance.[Document verification scope](document-verification-api.html) - ReviewPause for configured manual review for the tenant under the agreed contract. Requires tenant configuration. Authorized reviewers inspect the document record and record their decision.[Configured review workflow](email-document-workflow-automation.html) - DecideRecord the authorized team's onboarding disposition against its acceptance policy. Requires tenant configuration; business approval is not autonomous.[Banking onboarding context](solutions-banking.html) - ActSend the agreed onboarding event to a receiving service through signed webhooks, with webhook signing and the downstream action confirmed. Requires tenant configuration.[Webhook handoff](https://docs.cogneris.ai/docs/api/) - ProveAgree the document references, decisions, and access controls needed for the onboarding record under contract-specific retention controls. Requires tenant configuration.[Security and evidence controls](security.html) ### Insurance claims Contract-configured journey: assemble claim documents, review the agreed checks, and hand off a recorded disposition. Coverage and settlement decisions remain with your authorized claims team. - CollectCollect the claim packet through the agreed intake workflow after confirming the document submission path. Requires tenant enablement.[Claims document processing](claims-document-processing-api.html) - ClassifyIdentify the document types in the claim packet against agreed categories. Requires tenant configuration.[Classification configuration](https://docs.cogneris.ai/docs/api/document-classifier/) - ExtractExtract the claim fields agreed for the supplied document set. Requires tenant configuration; source references depend on the configured output.[Claims document extraction](extract-insurance-claims.html) - ValidateConfirm the supported checks for the claim packet separately. Validation requires tenant configuration. Fraud detection is not established by the platform evidence and is not promised for this journey. - ReviewPause for configured manual review for the tenant under the agreed contract. Requires tenant configuration. Authorized claims reviewers inspect the record and request corrections or record their decision.[Configured review workflow](email-document-workflow-automation.html) - DecideRecord the authorized claims team's disposition under its coverage and settlement policy. Requires tenant configuration; the workflow does not make an autonomous settlement decision.[Insurance workflow context](solutions-insurance.html) - ActDeliver the agreed claim event to a downstream endpoint through signed webhooks, with webhook signing and the receiving action confirmed. Requires tenant configuration.[Webhook handoff](https://docs.cogneris.ai/docs/api/) - ProveAgree which claim documents, review decisions, and lifecycle events support the record under contract-specific retention controls. Requires tenant configuration.[Evidence and trust scope](trust.html) Supported types ## Classification configured for your workflow. Each one calibrated for the formats, languages, currencies and local rules your customers actually use. Payslip HR Bank statement Credit Tax return Tax Tax invoice Tax Articles of incorporation Legal Driver's license Onboarding National ID Onboarding Proof of address Onboarding Bill / invoice Finance Service invoice Tax Receipt Finance Purchase order Finance Developer experience ## Three lines of code. One clean JSON. API and webhook samples for common development stacks. Signed webhooks, async jobs, and documented REST requests support custom integrations. Start with the [document extraction overview](document-extraction.html), then use the [document extraction API](document-extraction-api.html), parse layout and sections with the [document parsing API](document-parsing-api.html), or use the [OCR API](ocr-api.html) when scans and photos need to become validated data. C# / .NET Python Node.js Go cURL extract.cs ``` // REST API sample — C# using var form = new MultipartFormDataContent(); form.Add(new StreamContent(File.OpenRead("payslip.pdf")), "file", "payslip.pdf"); http.DefaultRequestHeaders.Authorization = new("Bearer", apiKey); var response = await http.PostAsync(extractionUrl, form); response.EnsureSuccessStatusCode(); ``` What ships in the box ## Six capabilities, one orchestration. The full document AI stack runs as one orchestrated pipeline. Every page goes through the same agentic flow — classify, extract, validate, score, audit — and lands as structured JSON ready for your downstream system. For API-first teams, these capabilities map to the [document extraction API](document-extraction-api.html), [document parsing API](document-parsing-api.html), and [OCR API](ocr-api.html). | Capability | What it does | | Ingestion | PDF (digital and scanned), JPEG, PNG, TIFF, DOCX, and email attachments. Multi-page documents stitched automatically. Inbox monitoring, webhook intake, and direct API upload. | | Classification | Tenant-configurable classification and zero-shot extraction return confidence scores and routing decisions. Validate each configured type against representative documents before production use. | | Extraction | ReAct-architected agents pull structured fields with per-field confidence. Cross-field math, date, and entity validation runs in the same pass. Span pointers preserve provenance to the source paragraph. | | Validation | Schema validation, cross-document reconciliation, and configurable business rules. Low-confidence fields are flagged with their confidence score and source citation; a manual review step queues the document for a reviewer. | | Contract-specific checks | Schema rules, reference-data checks, and cross-document consistency can be configured for a contracted workflow. Fraud detection is not established by the current capability evidence; any proposed fraud check requires separate verification before use. | | Audit trail | Configured workflows can retain tamper-evident metadata with model version, prompt and response hashes, reviewer identity, and per-field confidence. Retention is tenant-configurable and contract-specific. | By document type ## Pre-trained extractors for the documents you already process. Extractors return structured JSON with schemas and validation rules. Connect downstream steps through REST APIs and signed webhooks, configured for the contract; a packaged or pre-built connector catalogue is not implied. [ Finance / AP Invoice extraction Vendor, line items, GL-code hints, totals. Fields support customer-configured matching and delivery through APIs or webhooks. ](extract-invoices.html) [ Banking / lending Bank statement extraction Transactions, balances, recurring deposits. Sub-3-second parsing on US and Canadian statements for cash-flow underwriting. ](extract-bank-statements.html) [ Legal / CLM Contract extraction Parties, term, governing law, liability caps, named clauses. Span pointers to source paragraph for legal review. ](extract-contracts.html) [ Onboarding / CIP KYC document extraction Government IDs, proof of address, beneficial-ownership certifications. MRZ parsing and sanctions/PEP screening hooks. ](extract-kyc-documents.html) [ HR / lending Payroll extraction Pay stubs, W-2s, 1099s. Gross, deductions, net pay, and YTD reconciliation for income verification. ](extract-payroll.html) [ Tax / lending Tax return extraction 1040s, K-1s, schedules. AGI, deductions, credits, refund or amount owed — with cross-schedule math validation. ](extract-tax-returns.html) [ T&E / expense Receipt extraction Itemized receipts with merchant, line items, tax, and tip handling. Outputs can be mapped into customer-configured expense workflows through APIs or webhooks. ](extract-receipts.html) [ Logistics Bill of lading OCR Carriers, consignees, shipment IDs, weights, charges, and freight line items for logistics workflows. ](extract-bill-of-lading.html) [ Payments Check OCR API Payer, payee, amount, date, MICR context, and remittance fields for payment operations. ](check-ocr-api.html) [ Insurance ACORD forms extraction Policy, applicant, vehicle, property, coverage, premium, and loss data from insurance packets. ](extract-acord-forms.html) [ Insurance Insurance claim extraction FNOL, police reports, medical bills, repair estimates. Cross-document reconciliation for claim adjudication. ](extract-insurance-claims.html) Integrations ## Lives where your systems already run. REST APIs and signed webhooks support custom integrations with systems of record. Integration scope and field mapping are defined for each customer workflow. - ERP and financeNetSuite, SAP S/4HANA, Oracle Cloud ERP, Workday Financials, QuickBooks Online, Xero, Sage Intacct. - CRM and revenueSalesforce, Salesforce Financial Services Cloud, HubSpot. Customer records, opportunities, and applicants land as structured objects. - Lending and core bankingnCino, Encompass, Mambu, Alloy, Q2, Jack Henry, Fiserv, FIS, Plaid, MX. - CLM and legalIronclad, Agiloft, ContractWorks, Concord. Clauses, parties, and term data flow with span pointers preserved. - HR and payrollWorkday HCM, ADP, The Work Number, Gusto, Justworks. Wage and YTD data flows into HRIS and verification workflows. - T&E and expenseExpensify, Concur, Brex, Ramp, Pleo. Receipts and reimbursements route to the right policy and ledger code. - Onboarding and KYCPersona, Alloy, Sumsub, ComplyAdvantage. ID verification and CIP/CDD packets with sanctions screening triggers. - Custom integrations[REST APIs](api.html), signed webhook events, and API or webhook samples for common development stacks. Enterprise contracts can include scoped integration engineering. Security & compliance ## We treat your documents the way you would. ### Privacy by design Configurable retention, automatic PII anonymization and proper data-processing agreements out of the box. ### End-to-end encryption TLS 1.3 in transit. AES-256 at rest. Keys managed by a dedicated KMS instance per customer. ### Private deployment roadmap Private cloud, BYOC, and on-premises deployment are not generally available. Contract-dependent evaluation requires explicit technical and operational confirmation. FAQ ## Common questions. Can I run Cogneris on document types you don't list? Yes. Custom document types ship from sample documents in 1 business day on Enterprise — send 10 to 50 examples, get a calibrated extractor with the right schema, validators, and confidence thresholds. The same agentic pipeline handles them; the only thing that changes is the schema you target. What's the synchronous vs. async limit? Synchronous extraction is bounded by the 10 MB file limit and the request timeout, not by a page count. For long documents or high volume, submit a document job and receive a signed webhook callback: same per-page rate, just a different request shape. How does Cogneris handle multi-language documents? Thirty-plus languages are supported with field-level localization. Currency conversion, date format normalization (DD/MM vs. MM/DD), and address parsing follow the source-document locale. Output JSON normalizes to ISO formats (ISO 4217 for currency, ISO 8601 for dates, ISO 3166 for countries) so downstream systems get consistent data regardless of source language. What happens to documents that fail validation? Low-confidence fields and schema/business-rule failures are flagged on the extraction result, each carrying the extracted value, its confidence score, the validation error, and the source region it was read from. A manual review step sends the document to a review queue, where reviewers correct values against the original page and approve or reject. Cogneris can preserve tamper-evident audit metadata with model version, prompt and response hashes, and reviewer identity. Can Cogneris run in our cloud or on-premises? Private cloud, BYOC, on-premises deployment, customer-managed keys, AWS PrivateLink, and Google Cloud Private Service Connect are roadmap capabilities, not generally available services. Any contract-dependent evaluation requires explicit technical and operational confirmation before architecture or procurement planning. How fast is the integration? Integrate through REST APIs and signed webhooks. Scope, endpoints, authentication, webhook signing, and receiving-system behavior must be configured for the tenant under the agreed contract. [See the docs](https://docs.cogneris.ai/) or [the API reference](api.html). Walkthrough ## Six capabilities, one recording. Everything above is a description. The walkthrough is a screen recording of the product doing the work on a real document — including the field it was not sure about, which is the part a scripted demo leaves out. Play the walkthrough The recording is not published yet, so there is nothing to play here. We would rather leave the space empty than fill it with a reel that is not the product. - Upload — a real invoice or bank statement, running at the speed the platform actually processes it. - JSON output — the response envelope, with each value's bounding box drawn back onto the source page. - Confidence — the per-field scores, including the one field the model hesitated on. - Human review — the same document waiting in a workflow’s review queue, a reviewer correcting the flagged value, and the change recorded against them. - Webhook — the signed delivery landing, carrying the document id from the first step. Let's talk ## Ready for your first extraction? In 30 minutes we'll show you the platform running on your own documents. [ Book a demo ](contact.html) [See pricing](pricing.html) ## Source: document-extraction-api.html [Skip to content](#main) API · Document extraction # Document extraction API. PDFs in, structured JSON out. Cogneris extracts fields, tables, line items, signatures, clauses, and decisions from PDFs, scans, and images. Send a document with a template or inline schema; get validated JSON with confidence scores, citations, review routing, and webhook delivery. [Start with the API](api.html) [Read extraction docs](https://docs.cogneris.ai/docs/api/document-extraction/) ## Extract the fields your workflow actually needs Raw document text is rarely the target. Your system needs invoice totals, bank-statement transactions, KYC identity fields, contract clauses, claim amounts, policy numbers, and the evidence behind each value. Cogneris lets you request a shipped template or pass your own JSON schema so the response matches the object your application expects. ``` { "document_type": "invoice", "invoice_number": { "value": "INV-2048", "confidence": 0.99 }, "line_items": [{ "description": "Platform usage", "amount": 1480.00 }], "validation": { "status": "passed" } } ``` ## One API for PDFs, scans, images, and document packets Use the same endpoint for native PDFs, scanned PDFs, mobile photos, email attachments, and multi-document packets. Cogneris classifies the document, chooses the right extractor, stitches multi-page files, splits bundled packets, and returns one normalized payload per document. ## Schema-based output instead of raw OCR text The extraction response is typed JSON: values, confidence, page references, bounding boxes, validation status, and audit metadata. That makes the output ready for underwriting systems, AP automation, onboarding flows, claims platforms, data warehouses, and AI agents. ### Financial documents Invoices, receipts, payroll, tax forms, bank statements, and reconciliation packets. ### Identity and onboarding KYC IDs, passports, proof of address, beneficial ownership forms, and supporting evidence. ### Contracts and claims Clauses, obligations, FNOL packets, policy details, repair estimates, and medical bills. ## Confidence, citations, and human review Every field carries confidence and source evidence. High-confidence values can move straight through. Low-confidence or high-risk fields can route to human review without blocking the full document. ## Async jobs and webhooks for production volume Small files can run synchronously. Long packets, batch uploads, and high-volume workflows use async jobs with webhook callbacks, retry semantics, and signed payloads. Start in the [API reference](api.html) or go deeper in the [extraction docs](https://docs.cogneris.ai/docs/api/document-extraction/). ## API sample: extract a document Use your language's standard HTTP client with the REST API. Signed webhook samples show how to verify asynchronous callbacks without implying a published client library. cURL ``` curl -X POST "$COGNERIS_BASE_URL/Document/extraction" \ -H "Authorization: Bearer $COGNERIS_API_KEY" \ -F "template=invoice" \ -F "file=@./invoice.pdf" ``` ## Validation before data reaches your system Cogneris validates extracted fields with totals reconciliation, date checks, regex rules, cross-document consistency, required-field checks, and tenant-specific business rules before the data is approved. ## Common document extraction API use cases Use Cogneris for [invoice extraction](extract-invoices.html), [bank statement extraction](extract-bank-statements.html), [contract extraction](extract-contracts.html), KYC onboarding, insurance claims, payroll verification, and document-heavy workflow automation. ## High-intent document extraction pages ### [PDF data extraction API](pdf-data-extraction-api.html) Extract fields, tables, line items, and citations from native PDFs and scanned PDFs. ### [Convert documents to JSON API](convert-documents-to-json-api.html) Turn document packets into typed JSON with schemas, confidence, validation, and webhooks. ### [Extract tables from PDF API](extract-tables-from-pdf-api.html) Return line items, transactions, rows, columns, totals, and table-level evidence. ### [Document extraction API pricing](document-extraction-api-pricing.html) Compare pricing drivers across page volume, complexity, validation, review, and audit needs. ### [Document extraction benchmark](document-extraction-benchmark.html) Score vendors on field accuracy, citations, table fidelity, schema stability, review load, and webhook behavior. ### [Python API sample](https://docs.cogneris.ai/docs/api/examples/) Upload documents, run async jobs, verify signed webhooks, parse JSON, and route low-confidence fields to review. ## Document extraction API FAQ What is a document extraction API? A document extraction API accepts documents such as PDFs, scans, and images, then returns structured fields, tables, confidence scores, and metadata that software can use directly. Can Cogneris extract tables and line items? Yes. Cogneris extracts line items, table rows, totals, dates, parties, clauses, identifiers, and other schema-defined fields with page evidence and confidence scores. Can I pass my own JSON schema? Yes. You can use shipped templates or pass an inline schema so the response matches the object your application expects. How is this different from OCR-only APIs? OCR returns recognized text. Cogneris turns recognized document content into typed JSON, validates it, attaches evidence, and routes uncertain fields to review. How is document extraction API pricing calculated? Cogneris pricing is page-based, with workflow cost affected by document volume, complexity, validation rules, review routing, and support requirements. What file types does the document extraction API support? Cogneris supports native PDFs, scanned PDFs, common image formats, and document packets used in workflows such as invoices, KYC, claims, contracts, and lending. How fast is document extraction? Small documents can run synchronously, while longer packets and batches run asynchronously. Actual latency depends on page count, document quality, schema complexity, and validation steps. Does the API support webhooks for long documents? Yes. Long documents, batches, and high-volume workflows can run as asynchronous jobs with signed webhook callbacks. Can extracted fields include source citations? Yes. Extracted fields can include source citations, page references, bounding boxes, confidence scores, and validation state for reviewable output. Can low-confidence fields route to human review? Yes. Confidence thresholds and business rules can route only the uncertain fields to review while high-confidence fields continue through automation. Which high-intent keywords match this API? High-intent searches include document extraction API pricing, PDF data extraction API, invoice extraction API, convert documents to JSON API, document extraction API for developers, and comparison queries such as AWS Textract alternative. ## Source: pdf-data-extraction-api.html [Skip to content](#main) API · PDF extraction # PDF data extraction API. Structured output. Cogneris extracts fields, tables, line items, dates, parties, totals, and evidence from native PDFs and scanned PDFs, then returns validated JSON your systems can use. [Talk to our team](contact.html)[Read extraction docs](https://docs.cogneris.ai/docs/api/document-extraction/) ## Built for native PDFs, scanned PDFs, and packets A PDF data extraction API has to handle embedded text, OCR-only scans, rotated pages, multi-page packets, and mixed document types. Cogneris classifies each file, applies OCR when needed, extracts against a schema, validates the result, and keeps page evidence attached to the output. ### Fields Names, dates, totals, IDs, addresses, clauses, balances, policy numbers, and custom schema fields. ### Tables Line items, transactions, row values, columns, subtotals, and table-level confidence. ### Evidence Page references, citations, confidence scores, validation status, and review metadata. ## When this is stronger than OCR alone OCR gives you text. PDF data extraction gives you typed fields, nested arrays, normalized values, validation errors, and workflow state. That difference matters when the data feeds underwriting systems, ERPs, CRMs, compliance workflows, or agent tools. ## Request snippet: PDF to JSON The platform picks the template from the document itself, so there is no schema to declare and no client library to install. Use the optional `ComplementaryPrompt` field to ask for anything the template does not already return. ``` curl -X POST "$COGNERIS_BASE_URL/Document/extraction" \ -H "Authorization: Bearer $COGNERIS_API_KEY" \ -F "file=@./loan-packet.pdf" \ -F "ComplementaryPrompt=Also return the statement period and the ending balance." ``` ## PDF data extraction API FAQ Can a PDF data extraction API process scanned PDFs? Yes. Cogneris can process native PDFs and scanned PDFs by applying OCR, layout understanding, extraction schemas, validation rules, and source citations. Can PDF extraction return tables and line items? Yes. Cogneris can return table rows, line items, transactions, totals, confidence scores, and page-level citations as structured JSON. Does PDF data extraction support async jobs and webhooks? Yes. Long PDFs, packets, and batches can run asynchronously with signed webhook callbacks and retry semantics. Can extracted PDF fields include source citations? Yes. Fields can include page references, source text, bounding boxes, confidence scores, and validation status for review and audit. ## Related pages [Document extraction API](document-extraction-api.html)[Documents to JSON API](convert-documents-to-json-api.html)[Extract tables from PDF API](extract-tables-from-pdf-api.html)[Document parsing API](document-parsing-api.html) ## Source: convert-documents-to-json-api.html [Skip to content](#main) API · Structured JSON # Convert documents to JSON. Without brittle parsing. Send a PDF, scan, image, or packet with a schema. Cogneris returns typed JSON with values, confidence, citations, validation status, and audit metadata. [Start with the API](api.html)[Document extraction API](document-extraction-api.html) ## JSON that matches your workflow Generic OCR text still leaves your team with parsing work. Cogneris converts documents into predictable JSON objects: top-level fields, nested tables, validation outcomes, source evidence, and review state. ``` { "document_type": "bank_statement", "account_holder": { "value": "Northwind LLC", "confidence": 0.98 }, "transactions": [{ "date": "2026-05-14", "amount": -248.50 }], "validation": { "status": "review_required" } } ``` ## Where documents-to-JSON APIs are useful Use this pattern for underwriting packets, invoice ingestion, KYC onboarding, claims, contracts, tax documents, bank statements, payroll files, and any workflow where software needs structured data instead of a PDF attachment. ## Document-to-JSON API FAQ Can I convert documents to JSON with my own schema? Yes. Cogneris can use shipped templates or an inline schema so PDFs, scans, and packets return JSON shaped for your application. Can JSON output include confidence scores and citations? Yes. JSON fields can include values, confidence scores, citations, page references, bounding boxes, validation status, and review state. Can document-to-JSON jobs run asynchronously? Yes. Long documents and batches can run asynchronously with signed webhook callbacks once processing finishes. What file types can be converted to JSON? Cogneris supports native PDFs, scanned PDFs, images, and document packets for workflows such as invoices, KYC, claims, contracts, and bank statements. ## Related pages [PDF data extraction API](pdf-data-extraction-api.html)[Extract PDF fields](faq-api-to-extract-fields-from-pdfs.html)[Document extraction API](document-extraction-api.html)[API reference](api.html) ## Source: extract-tables-from-pdf-api.html [Skip to content](#main) API · Table extraction # Extract tables from PDFs. Rows you can trust. Cogneris extracts table rows, line items, transactions, quantities, prices, balances, and totals from PDFs with citations and validation metadata. [Talk to our team](contact.html)[Invoice extraction](extract-invoices.html) ## Table extraction is more than cell detection Useful table extraction has to understand headers, row boundaries, wrapped descriptions, subtotals, multi-page continuation, and the document type behind the table. Cogneris returns table rows as structured arrays with confidence and source citations. | Document | Tables to extract | Validation | | Invoices | Line items, quantities, unit prices, tax, totals | Subtotal, tax, and total reconciliation | | Bank statements | Transactions, dates, descriptions, balances | Opening and closing balance checks | | Claims packets | Services, repairs, bills, estimates, policy rows | Required fields, dates, and duplicate checks | ## Designed for review and audit Each row can carry page references, source regions, validation status, and reviewer actions. That makes it easier to inspect exceptions without rerunning OCR or manually tracing values through the original PDF. ## Related pages [PDF data extraction API](pdf-data-extraction-api.html)[Invoice extraction](extract-invoices.html)[Bank statement extraction](extract-bank-statements.html)[Tables and line items FAQ](faq-can-document-extraction-extract-tables-and-line-items.html) ## Source: document-extraction-api-pricing.html [Skip to content](#main) Pricing # Document extraction API pricing. What changes cost. A useful pricing comparison looks beyond the OCR call. Page volume, schemas, validation, review labor, audit trail, support, and integration work all affect total cost. [See pricing](pricing.html)[Estimate ROI](roi-calculator.html) ## What usually drives API pricing | Driver | Why it matters | What to ask | | Page volume | Most extraction APIs price by page, document, or committed tier. | Are retries, blank pages, and failed jobs billed? | | Document complexity | Line items, handwriting, packets, and custom schemas require more processing and review. | Do simple and complex documents have different rates? | | Validation and review | Low-confidence fields need routing, UI, audit events, and reviewer actions. | Is review included or priced separately? | | Compliance needs | Audit trail, retention, tenant isolation, and data-processing terms add operational requirements. | Which evidence is included out of the box? | ## Compare cost per resolved document The cheapest per-page call is not always the cheapest workflow. Compare the cost to get from uploaded document to approved, validated, exported data. That includes engineering time, review time, exception handling, and ongoing schema changes. ## OCR API pricing versus extraction pricing OCR API pricing usually measures recognition volume: pages, images, or characters. Document extraction API pricing should be compared against the resolved business object: the invoice approved, the borrower packet validated, the KYC file cleared, or the claim routed with evidence. If a lower OCR rate creates custom parsing, more review, or weak citations, the workflow can cost more even when the API call is cheaper. | Pricing model | Works well for | Watch for | | Per page OCR | Search indexing, archive conversion, simple text capture. | Raw text still needs parsing, validation, and review. | | Per document extraction | Known document types such as invoices, receipts, IDs, and statements. | Confirm how multi-page packets and retries are billed. See the [credit table](pricing.html#how-credits-work) for worked examples. | | Workflow pricing | Regulated processes that need validation and audit evidence. | Compare included review, storage, webhook, and support limits. | ## Related pages [Cogneris pricing](pricing.html)[Document AI pricing comparison](document-ai-pricing-comparison.html)[Document extraction API](document-extraction-api.html)[AWS Textract alternative](compare-aws-textract.html) ## Source: document-parsing-api.html [Skip to content](#main) API · PDF parser # PDF parser API for workflows, agents, and databases. Parse PDFs, scans, images, and document packets into structured data your software can act on. Cogneris works as a PDF parser API for structured JSON, Markdown context, layout-aware chunks, page evidence, citations, confidence, and validation metadata. [Parse your first document](api.html)[See JSON examples](https://docs.cogneris.ai/docs/api/document-extraction/) ## Parse more than text Operational document parsing has to understand structure, not just characters. Cogneris identifies sections, tables, paragraphs, key-value pairs, page boundaries, and document type so downstream systems can use documents as clean data. Use it when a simple PDF parser drops table context, field labels, or citations your workflow needs. ## From PDF layout to typed JSON Native PDFs, scans, photos, and email attachments flow through OCR, layout detection, classification, extraction, and validation. The result is JSON with the context your workflow needs: page evidence, citations, confidence scores, and normalized field values. ## Agent-ready document data AI agents need dependable tool inputs. Cogneris turns document content into structured objects with evidence and validation state, so an agent can reason over invoices, contracts, onboarding packets, and claims without guessing from loose text chunks. ### Layout context Sections, tables, field labels, pages, and reading order are preserved for downstream logic. ### Typed outputs Return JSON objects, arrays, numbers, dates, currencies, booleans, and validation status. ### Evidence links Keep source page references and citations so users can verify parsed data quickly. ## Tables, paragraphs, key-value fields, and citations Cogneris parses document structure and extracts named values in the same workflow. That means a bank statement can return both the transaction table and account metadata, while a contract can return paragraphs, clause spans, and extracted obligations. ## Layout-aware parsing for RAG Retrieval systems work better when the parser preserves the document's structure. Cogneris can keep sections, tables, page numbers, field labels, reading order, embeddings metadata, and citations together so a RAG pipeline retrieves meaningful evidence instead of isolated text fragments. Use layout-aware parsing when documents contain tables, multi-column pages, scanned forms, signatures, footnotes, clauses, attachments, or long packets where page context changes the answer. ## RAG-ready JSON and Markdown context Search and agent workflows often need both typed JSON and readable context. Cogneris can return normalized fields, table rows, source citations, and layout-aware text blocks that can be rendered as Markdown for retrieval, review, or downstream prompt context. That keeps chunking tied to the original document structure: headings stay with their paragraphs, tables stay with captions, and extracted values stay linked to their source page and bounding-box evidence. ## Chunking is not enough for operational workflows Chunking is useful for retrieval, but back-office automation usually needs typed values, validation, routing, and audit evidence. Cogneris is built for the moment after ingestion: when parsed content has to trigger a workflow, update a record, or support a decision. ## When to use parsing, extraction, or Q&A | Job | Best for | Cogneris output | | Parsing | Preparing document structure for systems or agents | Sections, tables, fields, citations, JSON | | Extraction | Pulling named business fields from a document | Typed schema with values and confidence | | Q&A | Asking grounded questions about a document | Answer, citations, reasoning trace | For implementation detail, read the [extraction API docs](https://docs.cogneris.ai/docs/api/document-extraction/), [classification docs](https://docs.cogneris.ai/docs/api/document-classifier/), [custom agents docs](https://docs.cogneris.ai/docs/developers/mcp-server/), or the architecture guide on [ReAct document workflows](pillar-react-architecture.html). ## Document parsing API FAQ What is the difference between document parsing and document extraction? Parsing prepares document structure such as sections, tables, text blocks, and citations. Extraction pulls named business fields into a specific schema. Can Cogneris parse PDFs for AI agents? Yes. Cogneris can return structured document context, page evidence, and validated fields that agents can use as reliable tool inputs. Does Cogneris return Markdown or JSON? Cogneris focuses on JSON for operational workflows. JSON can include sections, paragraphs, tables, citations, field values, and validation metadata. Can I preserve page references and citations? Yes. Parsed and extracted values can include page references, source spans, and bounding boxes for review and auditability. How does parsing work for scanned PDFs? Scanned PDFs run through OCR and layout understanding before the result is normalized into document structure and typed JSON. When should I use Ask the PDF instead? Use Ask the PDF when the user needs conversational answers. Use parsing or extraction when software needs structured output it can store, validate, or act on. ## Source: document-ai-for-rag.html [Skip to content](#main) RAG and document parsing # Document AI for grounded RAG. RAG fails when documents are flattened into loose text. Cogneris keeps layout, tables, headings, page evidence, citations, and validated fields together so retrieval has useful source context. [Talk to our team](contact.html)[Open parsing API](document-parsing-api.html) ## Why generic chunking breaks on business documents Invoices, statements, contracts, policies, claims, and onboarding packets carry meaning in their layout. A table row, signature, clause, checkbox, or page footer can change the answer. Cogneris parses document structure before chunking so retrieval does not separate values from their evidence. ### Layout-aware chunks Keep headings, paragraphs, tables, captions, and page context together. ### Citation-first output Return source page references and bounding boxes alongside extracted values. ### Validated fields Use typed JSON when a workflow needs reliable values, not just search snippets. ## RAG preprocessing API for documents A RAG preprocessing API should do more than split text every few hundred tokens. It should preserve reading order, headings, tables, page references, document type, image-derived text, and source citations so retrieval can explain where an answer came from. Cogneris can prepare both retrieval-friendly context and workflow-ready data from the same document: Markdown for prompts, layout-aware chunks for vector search, JSON for automation, embedding metadata for retrieval filters, and citations for review. ## Outputs for RAG and agents | Need | Output | Why it matters | | Retrieval | Layout-aware text blocks and Markdown | Preserves structure for better prompts and source previews | | Automation | Typed JSON with validation state | Lets software route, approve, or reject records | | Audit | Citations, page references, and bounding boxes | Lets reviewers trace answers to source documents | ## When to use document AI for RAG Use this pattern when your documents contain tables, financial values, dense clauses, multi-page packets, scans, IDs, handwritten fields, or forms where visual proximity matters. The same parsed output can feed vector search, an agent workflow, a review UI, and a downstream system of record. ## RAG parsing vs extraction APIs Unstructured and LlamaParse-style tools are strong when the goal is to prepare documents for retrieval. Extraction APIs are stronger when the answer needs to become a record, approval, exception, or workflow event. Many production systems need both: parsed chunks for search and typed JSON for action. | Decision | Use RAG parsing when | Use extraction APIs when | | Output | Markdown, chunks, layout blocks, images, and table context | Typed JSON fields, arrays, validation state, and audit metadata | | Primary user | AI search, support copilots, research agents, knowledge bases | AP, underwriting, claims, onboarding, compliance, and product workflows | | Success metric | Answer relevance and citation quality | Field accuracy, STP rate, false approvals, and review effort | | Failure mode | Wrong or ungrounded answer | Wrong value posted to a system of record | ## Recommended architecture Parse once, then branch. Send layout-aware chunks and source citations to retrieval. Send validated fields, tables, and business rules to workflow automation. Keep both outputs tied to the same document ID so an agent can answer questions and the application can still trace every decision back to the original page. ## Semantic chunking and embedding metadata For production RAG, store more than chunk text. Keep page number, section heading, document type, table ID, bounding box, citation URL, confidence, extraction job ID, and document metadata with every RAG-ready chunk. That lets the retriever filter by document class, re-rank by source quality, and show an evidence preview instead of a detached snippet. ## Example parsed output The same parse can expose Markdown, chunk metadata, table rows, and extracted fields. That lets a retrieval system show the source paragraph while a workflow system stores the normalized value. | Object | Example fields | Use | | chunk | text, markdown, page, section, citation, bbox | Vector search and grounded answers | | table | caption, columns, rows, page, confidence | Financial tables, line items, schedules, transactions | | field | name, value, type, confidence, citation, validation | Workflow automation and system-of-record updates | ## Related pages [Document parsing API](document-parsing-api.html)[Extraction docs](https://docs.cogneris.ai/docs/api/document-extraction/)[Source citations](document-ai-source-citations.html)[Benchmark guide](document-extraction-benchmark.html)[PDF parsing comparison](compare-pdf-parsing-apis.html)[Compare Unstructured](compare-unstructured.html) ## Source: ocr-api.html [Skip to content](#main) API · OCR # OCR API that returns usable document data. Cogneris reads text, tables, handwriting, and layout from PDFs, scans, and photos, then turns recognition output into validated JSON. Use OCR when you need text; use Cogneris when your workflow needs trustworthy fields, evidence, and decisions. [Try OCR plus extraction](api.html)[Read the OCR guide](term-ocr.html) ## OCR is the first step, not the final output OCR recognizes document text. Most business workflows need more: named fields, tables, totals, identities, document types, validation status, and an audit trail. Cogneris combines OCR, layout understanding, multimodal extraction, and deterministic validation in one API. ## Recognize text, tables, layout, handwriting, and fields Use Cogneris on native PDFs, scanned PDFs, mobile photos, receipt images, invoices, bank statements, contracts, IDs, and onboarding packets. The API preserves the layout context needed to understand labels, rows, columns, and page-level evidence. ### Text and layout Read text, page structure, bounding boxes, sections, and table boundaries. ### Fields and tables Return document-specific JSON rather than making your team parse raw OCR text. ### Review and validation Send uncertain fields to review and validate values before downstream export. ## Turn OCR output into structured JSON For [receipts](extract-receipts.html), [invoices](extract-invoices.html), and [KYC documents](extract-kyc-documents.html), Cogneris can return normalized fields, arrays, dates, amounts, identifiers, and validation metadata. Read the [extraction docs](https://docs.cogneris.ai/docs/api/document-extraction/) for request and response examples. When the workflow needs the document itself rather than fields, the same pipeline can return a [searchable PDF](searchable-pdf-api.html) with the OCR text layer kept behind the original scan. ## Request snippet: OCR plus structured output Teams comparing OCR APIs often start with text extraction, then add tables and JSON when the output reaches a product workflow. There is no client library to install — the call is a multipart POST with a bearer key, and the [language support matrix](https://docs.cogneris.ai/docs/reference/faq/) says the same for every stack. ``` curl -X POST "$COGNERIS_BASE_URL/Document/extraction" \ -H "Authorization: Bearer $COGNERIS_API_KEY" \ -F "file=@./statement.pdf" ``` ## Confidence scores and bounding boxes Every important value can carry confidence, source page, and bounding-box evidence. Reviewers can jump from JSON to the document region that produced the value, which makes QA and audit work much faster. Confidence is only half the picture: our [document AI accuracy SLA](document-ai-accuracy-sla.html) sets out how accuracy is defined per field and measured against straight-through rates. ## Selection marks, signatures, and handwriting Production OCR often needs to understand more than printed text. Cogneris can preserve evidence around checkboxes, selection marks, signature presence, stamps, handwritten notes, and low-confidence handwriting so a reviewer can verify the field instead of re-reading the full document. ## Multimodal extraction for hard documents Some documents are too messy for text recognition alone: rotated scans, unusual templates, broken tables, photos with shadows, handwritten notes, and packets with mixed document types. Cogneris combines OCR with multimodal reasoning and validation rules so the output is built for decisions, not just search. ## Agentic OCR for high-risk workflows For checks, logistics packets, insurance forms, and finance documents, the OCR result is only the first pass. Cogneris can classify the document, extract the right schema, check totals or identifiers, and carry the confidence and evidence each value was read with, so a workflow can stop on the ones it should not trust. That loop is what [document AI agents](document-ai-agents.html) do for back-office work: classify, extract, validate, and hand off the cases they cannot settle. ## Human review for uncertain fields Every value comes back with a confidence score and the region it was read from, so you can decide which ones a person should see. Nothing escalates on the score by itself: review is a step you place in the workflow, and when the run reaches it the documents land in a review queue where a reviewer can correct a value before approving it. That keeps the workflow moving without pretending every OCR result is equally trustworthy. ## OCR API pricing and workflow cost Most OCR APIs look inexpensive when you compare only page recognition. The real cost appears when teams add custom parsers, table cleanup, confidence thresholds, review tools, and audit evidence. For buying criteria, see the [document extraction API pricing guide](document-extraction-api-pricing.html) and the [extraction benchmark](document-extraction-benchmark.html). ## OCR API FAQ What is an OCR API? An OCR API recognizes text from PDFs, scans, photos, and images. Cogneris extends OCR with extraction, validation, confidence, and review workflows. Does Cogneris support scanned PDFs and mobile photos? Yes. Cogneris supports native PDFs, scanned PDFs, mobile photos, and common image formats. Can OCR output include tables and line items? Yes. Cogneris can extract table rows, line items, totals, labels, and field values instead of returning only plain text. How is this different from a text-only OCR API? Text-only OCR gives you recognized characters. Cogneris returns structured JSON, confidence scores, validation status, and evidence for the fields your workflow needs. Can I get bounding boxes and confidence scores? Yes. Field responses can include confidence scores, page references, and bounding boxes for review and audit trails. Can OCR results be validated before export? Yes. Cogneris can validate OCR-derived fields with business rules, totals reconciliation, date checks, and required-field checks before export. ## Source: document-automation-portal.html [Skip to content](#main) Portal · Document automation # Document portals that finish the work. Cogneris turns a document portal into an operating workflow: secure upload, document extraction, validation, review status, reminders, tenant permissions, and audit trail in one system. [Design your portal](contact.html) [See product stack](products.html) ## Move beyond upload and storage A client document portal should not stop after a file lands in a folder. The portal should know which documents are required, whether each upload is readable, what data was extracted, who reviewed it, and what still blocks the workflow. Cogneris is built for document-heavy onboarding, lending, insurance, accounting, procurement, healthcare, and compliance workflows where a missing file or weak extraction creates real operational risk. Request Create required-document checklists by customer, tenant, case, or workflow stage. Upload Accept PDFs, scans, images, and document packets with file-type, quality, and duplicate checks. Extract Classify each document and return structured fields, tables, confidence scores, and source evidence. Validate Run required-field, date, total, identity, policy, and cross-document consistency checks. Review Route low-confidence fields to a reviewer while approved fields continue through automation. Audit Preserve who uploaded, extracted, reviewed, changed, exported, or approved each field. ## Use document AI where the portal needs judgment Most portals can collect files. Fewer can decide whether a W-9 is complete, whether a bank statement covers the requested period, whether an invoice total reconciles with its line items, or whether a KYC packet is missing beneficial ownership evidence. Cogneris combines document classification, extraction, validation, and review routing so the portal can act on document contents, not just filenames. ### Secure upload Signed upload links, role-based access, tenant scoping, expiration controls, and per-document status. ### Extraction and validation Turn PDFs and scans into structured JSON with confidence, citations, and rule checks before export. ### Reminders and status Trigger reminders for missing, expired, invalid, unreadable, or unreviewed documents. ### Review queue Show reviewers the original page, extracted value, confidence, validation error, and edit history. ### Audit trail Log model version, prompt hash, source evidence, reviewer ID, tenant ID, and export event. ### Workflow handoff Send approved data to CRMs, ERPs, loan origination systems, claims platforms, or webhooks. ## Built for tenant and per-tenant controls Document portals often serve many customers, cases, vendors, or business units. Cogneris keeps permissions, encryption, review queues, schemas, and audit events scoped to the tenant so one portal can support many workflows without blending evidence. ## Where teams use it Use Cogneris document portal automation for [lending document portals](lending-document-portal.html), [KYC document portals](kyc-document-portal.html), [accounting and tax document portals](accounting-tax-document-portal.html), [email document workflow automation](email-document-workflow-automation.html), [insurance claims](solutions-insurance.html), [healthcare documentation](solutions-healthcare.html), vendor onboarding, procurement packets, and customer document collection. ## Portal patterns by vertical ### [Mortgage and lending](mortgage-document-extraction-api.html) Borrower uploads, loan packet classification, missing evidence reminders, underwriting fields, and review status. ### [KYC onboarding](kyc-document-portal.html) ID, proof of address, bank letters, beneficial ownership files, validation checks, and audit evidence. ### [Accounting and tax](accounting-tax-document-portal.html) Client checklists, W-2s, 1099s, K-1s, bank statements, reviewer queues, and source-document extraction. ### [Insurance claims](claims-document-processing-api.html) Claim forms, ACORD data, supporting evidence, photos, policy details, and exception routing. ### [Shared inbox intake](email-document-workflow-automation.html) Email attachments, document packets, missing-file follow-up, extraction, and webhook delivery. ## Connect portal intake to extraction APIs Teams can start with the [document extraction API](document-extraction-api.html), parse packets with the [document parsing API](document-parsing-api.html), validate with the [validation docs](https://docs.cogneris.ai/docs/product/core-concepts/), connect systems with the [integration docs](https://docs.cogneris.ai/manuals/admin/workflow/journeys/connect-your-systems/), and expose status through their own portal UI. Cogneris handles the document AI layer behind it. ## Document workflow and case management Many portals are really case-management systems: a borrower file, vendor packet, insurance claim, onboarding checklist, or AP exception queue. Cogneris keeps document requirements, extraction status, validation errors, reviewer actions, reminders, and webhook handoffs tied to the case instead of leaving files in a folder. ## Document automation portal FAQ What is a document automation portal? A document automation portal is a secure workflow where users upload required files, the system extracts and validates data, reviewers handle exceptions, and the business keeps an audit trail. How is a document portal different from file sharing? File sharing stores and transfers files. A document portal tracks required documents, validates uploads, extracts data, sends reminders, manages permissions, and shows review status. Can Cogneris send reminders for missing documents? Yes. Cogneris can trigger reminders based on missing, expired, invalid, unreadable, or unreviewed documents and deep-link users back to the right upload step. Can document extraction run inside the portal? Yes. Uploads can trigger classification, extraction, validation, review routing, webhook delivery, and audit logging in the same workflow. Does the portal support tenant permissions? Yes. Cogneris supports tenant and per-tenant controls, role-based permissions, reviewer queues, and audit metadata for each document event. ## Source: pricing.html [Skip to content](#main) Pricing # One credit pool. No surprises. Every operation spends credits from one monthly pool. Pick a plan by volume — [the credit table](#how-credits-work) shows what each operation costs. No credit card ### Free For evaluating document automation with your own files. € 0for 14 days - 400 credits included - ≈ 95 documents or ≈ 7 verification processes - Stops at zero credits — no overage - No credit card required [Get 400 free credits](https://login.cogneris.ai/signup) Steady volume ### Pro For teams with steady monthly volume. € 99/mo - 5,000 credits / month - ≈ 1,190 documents or ≈ 90 verification processes / month - Overage: €0.03 / credit [Subscribe](https://login.cogneris.ai/signup) Lowest credit rate ### Scale For production volume at the lowest rate per credit. € 499/mo - 40,000 credits / month - ≈ 9,520 documents or ≈ 720 verification processes / month - Overage: €0.01 / credit [Subscribe](https://login.cogneris.ai/signup) Post-paid ### Unlimited No monthly fee. Pay only for the credits you use. € 0.02/credit - No monthly fee, no allowance - ≈ €0.084 per document, ≈ €1.10 per verification process - Billed at the end of each period [Subscribe](https://login.cogneris.ai/signup) Custom ### Enterprise Contracted volume, invoiced and paid by bank transfer. Custom - Negotiated rate (list: €0.02 / credit) - 99.99% uptime SLA + 24/7 support - Dedicated account manager + CSM - HIPAA BAA, GDPR DPA, SOC 2 reports under NDA [Talk to sales](contact.html) Every paid plan includes the integration APIs, document classification, field extraction, structured JSON export and the verified document collection portal. Document and process counts are estimates for 1-page documents classified and extracted (4.2 credits) and a verification process of about 55 credits. Longer documents use more credits per document but fewer per page. Prices in euros: every new account is created and billed in our EU region. The US region is paused since 2026-10-27; accounts in the US region pay the same figures in US dollars. How credits work ## How we charge credits. - One pool for everything — classification, extraction, signatures and portal processes all spend the same credits. - Per operation, not per seat — add as many users as you need; you pay for the work processed. - Base plus per page — each operation charges its base once, so a long document costs less per page. - Failures are not charged — failed processing costs nothing, and a portal process is charged only when it completes. - Monthly reset — credits on prepaid plans reset at the start of each billing month. | Operation | Credits | | Document classification | 1 + 0.2 per page | | Field extraction (trained template) | 2 + 1 per page | | Zero-shot extraction (no template) | 3 + 1 per page | | Document split | 2 + 0.2 per page | | Cross-check between documents | 3 per check | | Face match | 1 | | Electronic signature | 2 | | Verified portal process | 30 per completed process | | Liveness check | 15 | | WhatsApp invitation | 3 | | SMS invitation | 4 | | Document fraud verification | 8 | | Build AI: clarifying questions | 2 per call | | Build AI: market research | 10 per call | | Build AI: build plan | 10 per call | | Build AI: plan revision | 5 per call | - 1-page document — classified and extracted: 1.2 + 3 = 4.2 credits. - 5-page document — 2 + 7 = 9 credits, or 1.8 per page. - KYC/KYB verification — 5 to 6 documents through the portal, classified and extracted, plus the process fee: about 55 credits. ## Estimate your monthly cost USD estimate, not a quote. Assumes every document is classified and extracted with a trained template. Taxes and negotiated Enterprise rates are excluded. Documents per month Pages per document Plan ProScaleUnlimited — / month — credits · base: — + overage: — Scale covers this volume for less. Enter whole numbers: zero or more documents, one or more pages. 50,000+ documents per month? [Talk to sales about volume pricing.](contact.html) [Open the full ROI calculator](roi-calculator.html) to model a document mix, manual review costs, and payback. Availability ## A commitment, and a live record. The uptime SLA is a commitment with service credits behind it. What our monitors record is published live on the status page, for the EU region that serves every new account. One is a promise, the other is history; neither substitutes for the other. Contractual SLA 99.9% Monthly, on the paid self-serve plans. Enterprise contracts carry 99.99%; the Free plan has no SLA. The binding wording lives in your contract. Measured availability · EU region [status-eu.cogneris.ai](https://status-eu.cogneris.ai) Live component health, response times, availability history and incidents for the EU production region, the one that serves every new account. We link to the live record instead of quoting a past window, so this page never shows a stale number. Frequently asked questions ## Pricing questions, answered. How is overage billed? Credits above the monthly allowance are billed at €0.03 each on Pro and €0.01 on Scale. Free never bills overage: processing stops at zero credits. Unlimited and Enterprise have no allowance, so every credit is billed at the end of the period. Pro and Scale also carry a monthly overage cap, 100 in the subscription's currency by default and adjustable by a company admin up to 10,000; past the cap, requests are refused with HTTP 402 until the next billing period or a higher cap. Do you train models on my data? No. Customer data never enters our model training. All LLM API calls run with zero-retention configured. For fine-tuned models on Enterprise, the artifact is isolated per customer tenant and never shared. Can an AI model invent a number? Yes, model output can be uncertain. Cogneris reports field-level confidence and can require source citations, deterministic validation, and multi-model consensus before a critical value moves downstream. Human review is a step you place in the workflow: it holds the document in a review queue, where a reviewer checks the value against the source page and corrects it before it moves on, and nothing escalates on the confidence score by itself. The audit trail records the model version, prompt and response hashes, validation results, and reviewer action. What does a credit buy? A 1-page document classified and extracted uses 4.2 credits; a 5-page document uses 9, because each operation charges its base once. A page is one rendered side: one PDF page or one image. A verification process through the portal uses about 55 credits. Can I process documents synchronously? Yes. The synchronous endpoints are bounded by the 10 MB file limit and the request timeout, not by a page count. For long documents we recommend a document job with a webhook callback. Both modes cost the same credits. How long does implementation take? Free, Pro, Scale and Unlimited are self-serve — most teams are extracting documents in under an hour. Enterprise implementation timing depends on the signed integration and security scope. Private VPC deployment remains a roadmap capability, not a generally available service; any contract-dependent evaluation requires explicit technical and operational confirmation. Is there a free trial? Yes. The Free plan includes 400 credits for 14 days, with no credit card. When the credits run out, processing stops and nothing is billed. If you need more volume to evaluate, talk to our team about a pilot. Can I switch plans? Yes — anytime. Upgrades and downgrades take effect immediately; the prorated difference is credited or charged on your next invoice. Do you offer volume discounts? Yes. Scale is the volume price: 40,000 credits for €499 works out to €0.0125 per credit, against €0.03 for Pro overage. Above that, Enterprise contracts negotiate the rate against committed volume. What payment methods do you accept? Card on the self-serve plans: Pro, Scale and Unlimited. Enterprise is invoiced and paid by bank transfer. We do not currently accept crypto. Still have questions? ### Talk to a human, not a sales bot. Pricing questions on Enterprise, custom schemas, or volume contracts — book 30 minutes with our team. No follow-up sequences, no nurture flows. Just a real conversation. [Talk to our team](contact.html) [Compare alternatives](compare.html) ## Source: roi-calculator.html [Skip to content](#main) Tool · ROI calculator # Document AI ROI calculator. No marketing math. Pick the levers that match your operation. We do the arithmetic. Numbers stay client-side — nothing leaves your browser unless you click "Talk to our team." Your operation Monthly page volume 10,000 Pages per month, across all document types. Plan ProScaleUnlimited Base subscription plus overage, in USD. Each page is priced as a 1-page document, classified and extracted (4.2 credits). Document mix (must sum to 100%) Simple — invoices, receipts, IDs 5 min/page to review % Standard — payslips, statements, KYC 12 min/page to review % Complex — contracts, medical, claims 35 min/page to review % Your current cost structure FTE loaded cost (annual) $72,000 Salary + benefits + overhead. US median for an AP/ops analyst sits around $72k. Current manual error rate 3% Industry studies put manual data-entry error rates at 1–3% per field. Cost of one error (rework + risk) $50 Includes rework, downstream correction, and a small allowance for risk. Copy share link Reset Copied! Estimated savings 50,000+ pages per month? [Talk to sales about volume pricing.](contact.html) These estimates exclude negotiated rates. Monthly savings $0 vs current process Annual savings $0 12 months at this volume Payback — months until breakeven 3-year TCO delta $0 total cost of ownership Current monthly cost$0 Cogneris monthly cost$0 Cost breakdown | Line item | Current process | With Cogneris | | FTE review time | $0 | $0 | | Error rework + risk | $0 | $0 | | Cogneris plan (credits) | — | $0 | | Monthly total | $0 | $0 | ### These numbers feel right? We'll run the same math against your actual document mix in a 30-minute call. If the gap holds up, we'll pilot Cogneris on 50–100 of your real documents and report the actual auto-approval rate, field-level accuracy, and failure modes — at no cost. [Book a 30-min call](contact.html) New to the category? Start with [the IDP buyer's guide](what-is-idp.html), or read [the 2026 State of Document AI report](state-of-document-ai-2026.html) for market context — pricing, accuracy benchmarks, adoption by industry. How this is calculated. We assume the FTE handles only document review (no other tasks), 1,800 productive hours per year, and that Cogneris's auto-approval rate is 80% on simple, 65% on standard, and 35% on complex pages — the rest still needs human review at the original throughput. Error reduction is set at 90% on simple, 85% on standard, and 75% on complex pages — Cogneris's validation catches arithmetic and cross-field issues that manual review often misses, but new AI-specific failure modes emerge that you should still watch for. Pricing uses the selected plan's base subscription and published credit rates, shared with the pricing-page estimator: each page is priced as a 1-page document, classified and extracted, at 4.2 credits. Negotiated volume discounts are excluded; talk to sales at 50,000+ pages/month. Review effort and error assumptions apply per page, so multi-page documents may need different assumptions. Setup cost is treated as zero on the assumption you'd pilot before committing — if you bake in onboarding effort, treat the payback as a floor. None of these numbers are a quote. Use this as a sketch, not a contract. ## Source: mcp.html [Skip to content](#main) Documentation · Integration # Cogneris, inside your code editor. Cogneris runs a Model Context Protocol server. Connect VS Code, Cursor, Claude Code or any MCP client and the assistant in your editor reads our docs and API reference, creates API keys and webhooks, trains classifiers and runs fill jobs — on your tenant, with the same audit trail as the REST API. [← Back to documentation](https://docs.cogneris.ai/) ## What it is A Model Context Protocol server in front of the same platform your API keys already reach. - An MCP endpoint, not a second product — every tool call goes to the same Cogneris services the REST API uses. Same tenant, same metering, same audit trail. - Built for writing integrations — the assistant searches the docs, reads the exact contract of a route with code in Node, Python, C# and curl, then creates the API key and webhook your code needs. - 53 tools in eight areas — account, docs and API reference, API keys, webhooks, classifiers, workflows and their runs, Build AI, and PDF fill. 21 read tools carry `mcp.read` and 10 write tools carry `mcp.write`; API key management, webhook management and Build AI have scopes of their own. - Stay connected with a developer token — generate it in the Cogneris platform, paste it once, and your editor keeps it in its own secret storage for up to 90 days. - Any MCP client — VS Code, Cursor and Claude Code are the three we document below; anything that speaks streamable HTTP works the same way. ## Watch it in VS Code A four-minute walkthrough, recorded in production: set up the server, then ask GitHub Copilot for an API key, Python code that extracts and classifies documents, and a webhook with a verified signature. - [Cogneris in VS Code: MCP, API keys, Python code and webhooks](https://youtu.be/OocTr2fRICk) — in English - [Cogneris no VS Code: MCP, chave de API, código Python e webhooks](https://youtu.be/qU3JFrixlQE) — in Brazilian Portuguese ## Server URL One URL per production region — use the one your tenant lives in. Europe is the default region for new tenants. | Region | MCP server URL | Serving region | | Europe (default) | `https://mcp-eu.cogneris.ai/` | europe-west1 | | United States | `https://mcp-us.cogneris.ai/` | us-central1 (paused since 2026-10-27) | Your tenant region is pinned at sign-up and matches the REST host you already use (`api-eu`; the US region and `api-us` are paused since 2026-10-27) — see [Environments](https://docs.cogneris.ai/docs/api/getting-started/). The MCP page in the platform shows the right URL for your tenant. The JSON-RPC endpoint is the root path `/`, not a sub-path. ## Generate a developer token The recommended way in: one token, scoped and expiring, that keeps your editor connected. - Where — in the Cogneris platform, open Vision → MCP. The page shows your server URL, ready-to-paste configuration for each editor, and the token generator. - What you choose — a name that says where the token is used, a validity of 30 or 90 days, and its scopes. `mcp.read` and `mcp.write` are ticked by default; add `apikeys.manage` or `webhooks.manage` only if the assistant should manage keys or webhooks. - Shown once — copy the token when it appears; it is not shown again. Generate a new one if you lose it. - Tied to you — every action taken with the token is recorded against your user. Revoke it from the same page when you no longer need it. ## Connect your editor Three editors, one server. Swap the URL for your region. ### VS Code Put this in `.vscode/mcp.json`. The token is a password input: VS Code asks for it once, keeps it in its secret storage, and never writes it to the file — so the file is safe to commit. .vscode/mcp.json Copy ``` { "inputs": [ { "id": "cogneris-token", "type": "promptString", "password": true, "description": "Cogneris developer token" } ], "servers": { "cogneris": { "type": "http", "url": "https://mcp-eu.cogneris.ai/", "headers": { "Authorization": "Bearer ${input:cogneris-token}" } } } } ``` ### Cursor Put this in `~/.cursor/mcp.json` and replace the placeholder with your token. This file holds the token in plain text, so keep it out of any repository. ~/.cursor/mcp.json Copy ``` { "mcpServers": { "cogneris": { "url": "https://mcp-eu.cogneris.ai/", "headers": { "Authorization": "Bearer " } } } } ``` ### Claude Code One command registers the server; replace the placeholder with your token. terminal Copy ``` claude mcp add --transport http Cogneris https://mcp-eu.cogneris.ai/ \ --header "Authorization: Bearer " ``` ### A custom MCP client The server is stateless, so a single POST lists the catalog. Send both content types in `Accept` — replies are SSE-framed. Then call `cogneris_whoami` to confirm which tenant, user and scopes the connection carries. tools/list over JSON-RPC Copy ``` curl -sS https://mcp-eu.cogneris.ai/ \ -H "Authorization: Bearer $COGNERIS_MCP_TOKEN" \ -H "Content-Type: application/json" \ -H "Accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}' ``` ## Browser sign-in (OAuth) The alternative to a token: sign in through the browser, for sessions of up to 1 hour. - Configure the URL without a header — a call without credentials answers `401` with a `WWW-Authenticate` header pointing at the server's metadata, which is the signal an MCP client uses to start sign-in. - Authorization code with PKCE — the client sends you to the Cogneris sign-in page for your region (`sso-eu.cogneris.ai` or `sso-us.cogneris.ai`) and exchanges the code for a short-lived bearer token. There is no client secret in your configuration. - The session lasts 1 hour — that cap is a security control of the Cogneris sign-in. To stay connected for days, use a developer token instead. - Covers the two MCP scopes — browser sign-in grants `mcp.read` and `mcp.write`. The API key, webhook management and Build AI tools need a developer token that carries their scopes. - Discover before you configure — `GET /.well-known/oauth-protected-resource` is anonymous and returns the resource identifier (the server's own `https://` URL), the authorization server and the supported scopes, per RFC 9728. Read it rather than hardcoding. ## Scopes A tool whose scope your credential lacks is not listed, and refuses to run if called. | Scope | What it unlocks | | `mcp.read` | Every listing and lookup: account, usage, docs and API reference, classifiers, workflows and their runs, the workflow schema and validation, fill templates and jobs, the webhook event catalogue. | | `mcp.write` | Creating classifiers, adding samples and training them, creating, updating and publishing workflows, starting workflow runs, registering fill templates, uploading sources and starting fill jobs. | | `apikeys.manage` | Listing, creating and revoking API keys. Never implied by `mcp.write`. | | `webhooks.manage` | Managing webhook endpoints and deliveries. Never implied by `mcp.write`. | | `builder.run` | Build AI sessions, which spend AI credits. Production tokens only, and the token's owner must be an administrator of the tenant. Never implied by `mcp.write`. | ## Tool catalog 53 tools, grouped by area. The names are exactly what `tools/list` returns; the scope column is what the credential must carry. Read this before you read the tables. Classifiers and fill templates are two different things, and confusing them is the fastest way to wire an agent to the wrong data. Classifiers are your own document types — the ones the Cogneris app shows as templates. Fill templates are PDF write-back layouts (acroform, overlay or HTML) that a fill job stamps data into. When an agent is asked "what templates do I have?", the answer comes from `cogneris_classifiers_list`, never from `cogneris_templates_list`. ### Account and usage Who the connection acts as, and what the tenant has consumed. | Tool | Scope | What it does | | `cogneris_whoami` | `mcp.read` | Shows who this connection acts as: tenant, user and e-mail, how you signed in (developer token or browser sign-in), sandbox or production, the granted scopes, and the API base URL your code should call. | | `cogneris_usage_summary` | `mcp.read` | Documents, pages and AI tokens per day and per document type over a UTC date range — the same numbers the dashboard shows. | ### Docs and API reference So the assistant in your editor writes code against the real API instead of guessing it. | Tool | Scope | What it does | | `cogneris_docs_search` | `mcp.read` | Searches the public Cogneris documentation and the server's integration guides, and returns the best-matching sections with their URLs. | | `cogneris_docs_get` | `mcp.read` | Reads one documentation page as plain text, in parts of at most 6000 characters. | | `cogneris_api_operations` | `mcp.read` | Lists the API's routes — method, path and a one-line summary, grouped by area — with an optional filter. | | `cogneris_api_reference` | `mcp.read` | Everything needed to call one route: parameters, request and response schemas, error codes, authentication, and ready-to-run Node, Python, C# and curl that read the key from `COGNERIS_API_KEY`. | ### API keys Keys for your own code. These tools are listed only when the developer token carries `apikeys.manage`. | Tool | Scope | What it does | | `cogneris_apikeys_list` | `apikeys.manage` | Lists the tenant's API keys — name, environment, scopes, last four characters, dates — never their values. | | `cogneris_apikeys_create` | `apikeys.manage` | Creates an API key with only the scopes your code needs. Sandbox by default; a production key needs an explicit confirmation. The value is returned once. | | `cogneris_apikeys_revoke` | `apikeys.manage` | Revokes a key by id. Immediate and irreversible: every request made with it fails from then on. | ### Webhooks Endpoints that receive Cogneris events. The event catalogue needs `mcp.read`; everything else needs `webhooks.manage`. | Tool | Scope | What it does | | `cogneris_webhooks_event_types` | `mcp.read` | The event catalogue: each versioned CloudEvents type with its description, an example payload, and whether it is sent today or only planned. | | `cogneris_webhooks_list` | `webhooks.manage` | Lists the tenant's webhook endpoints — URL, subscribed events, active flag. Never returns signing secrets. | | `cogneris_webhooks_get` | `webhooks.manage` | Reads one webhook endpoint by id. | | `cogneris_webhooks_create` | `webhooks.manage` | Registers a public HTTPS endpoint. Returns the signing secret once, with receiver code in Node, Python and C# that verifies the `Cogneris-Webhook-Signature` header. | | `cogneris_webhooks_update` | `webhooks.manage` | Replaces an endpoint's URL, description and subscribed events. Events left out of the list stop being delivered. | | `cogneris_webhooks_disable` | `webhooks.manage` | Stops deliveries to an endpoint. Events raised while it is disabled are not delivered to it. | | `cogneris_webhooks_delete` | `webhooks.manage` | Deletes an endpoint; deliveries to it stop. | | `cogneris_webhooks_rotate_secret` | `webhooks.manage` | Replaces an endpoint's signing secret and returns the new one once. The old secret stops validating new deliveries. | | `cogneris_webhooks_test` | `webhooks.manage` | Queues a signed test delivery to an endpoint, so you can check the receiver before a real event arrives. | | `cogneris_webhooks_deliveries_list` | `webhooks.manage` | Delivery history, newest first — event type, state, attempts, last error. Payloads and secrets are never included. | | `cogneris_webhooks_deliveries_get` | `webhooks.manage` | One delivery with its attempts: the HTTP status your receiver answered, latency and error code. | | `cogneris_webhooks_deliveries_replay` | `webhooks.manage` | Sends a failed delivery again. The endpoint must be active, and a delivery gets at most five attempts in total. | ### Classifiers Your own document types — the ones shown as templates in the Cogneris app. | Tool | Scope | What it does | | `cogneris_classifiers_list` | `mcp.read` | Lists the tenant's classifiers with their status (Draft, Queued, Processing, Training, Finished, Error) and sample files. | | `cogneris_classifiers_get` | `mcp.read` | Reads one classifier by id: name, status, field metadata and its sample files. | | `cogneris_classifiers_status` | `mcp.read` | The current training status of one classifier. | | `cogneris_classifiers_create` | `mcp.write` | Creates a classifier from a name and sample documents. It starts in Draft. Not idempotent. | | `cogneris_classifiers_add_samples` | `mcp.write` | Adds sample documents to a classifier. A Finished classifier is immutable — create a new one to change it. Not idempotent. | | `cogneris_classifiers_train` | `mcp.write` | Starts training from the samples. Training is asynchronous; poll the status until Finished or Error. | ### Workflows Write a workflow graph from a description: read the schema, validate, save a draft, publish when you confirm. No Build AI credits. | Tool | Scope | What it does | | `cogneris_workflows_schema` | `mcp.read` | The JSON Schema of a workflow graph, in pieces an assistant can read: the graph shape and an index of every phase type, then one phase type's configuration at a time, split into parts when it is large. | | `cogneris_workflows_list` | `mcp.read` | Lists the tenant's workflows — name, status, version — page by page. | | `cogneris_workflows_get` | `mcp.read` | One workflow: its published phases and a summary of the draft graph (nodes, phase types, connections). The full graph JSON comes in blocks of 6500 characters. | | `cogneris_workflows_validate` | `mcp.read` | Checks a graph with the same rules publish applies — schema, one start node, no cycle, reachability, the tenant resources it names — without saving anything. Costs no credits. | | `cogneris_workflows_create` | `mcp.write` | Creates a draft workflow from a name, a description and a graph. A draft runs nothing. Retrying the same create returns the same draft. | | `cogneris_workflows_update` | `mcp.write` | Replaces a draft's name, description or graph. The published version keeps running until the next publish. | | `cogneris_workflows_publish` | `mcp.write` | Publishes the draft as a new immutable version that new runs use. The assistant asks you first; a refusal lists every validation issue. | ### Workflow runs Try a published workflow on real documents and see what each step decided. | Tool | Scope | What it does | | `cogneris_workflow_runs_start` | `mcp.write` | Starts a run of a published workflow with 1 to 10 documents (base64, 10 MB each). Steps consume credits like the matching document operations. Retrying with the same documents returns the same run. | | `cogneris_workflow_runs_get` | `mcp.read` | A run's status, documents and steps, with a suggested wait before reading it again while it is in progress. | | `cogneris_workflow_runs_step` | `mcp.read` | What one step concluded: its outcome, the decisions behind it and the extracted fields per document, with long values cut and personal data masked. | ### Build AI Let Cogneris draft the whole process — workflow, classifier and templates — from your description. Listed only when the developer token carries `builder.run`; every step states its cost. | Tool | Scope | What it does | | `cogneris_builder_start` | `builder.run` | Opens a Build AI session from a description of your process and returns follow-up questions. Costs 2 credits. | | `cogneris_builder_answer` | `builder.run` | Saves your answers to the session's questions. Free. | | `cogneris_builder_plan` | `builder.run` | Drafts the plan — document types and fields, form, workflow, templates. Costs 10 to 30 credits: 10 for the plan, 10 more for an automatic market research when none ran yet, and up to 2 automatic revisions of 5. | | `cogneris_builder_revise` | `builder.run` | Changes the latest plan from an instruction in plain words. Costs 5 to 15 credits. | | `cogneris_builder_status` | `builder.run` | The session: questions, a summary of the latest plan and its issues, and credits consumed and pending. Free; while a step runs, reading the session is what moves it forward. | | `cogneris_builder_set_samples` | `builder.run` | Chooses Cogneris library examples, or none, for one document type of the plan. Free. Uploading your own samples stays in the Build AI screen. | | `cogneris_builder_approve` | `builder.run` | Approves one plan version and creates the workflow, classifier and templates in your tenant. Only after your explicit yes. | | `cogneris_builder_apply_status` | `builder.run` | Progress of an approved plan: applying, training, ready or failed. | ### PDF fill: templates, sources, jobs and artifacts Write-back: stamp extracted data into a PDF template and download the result. | Tool | Scope | What it does | | `cogneris_templates_list` | `mcp.read` | Lists PDF fill templates (acroform, overlay or HTML), paged, with an optional name filter. | | `cogneris_templates_get` | `mcp.read` | The full definition of one fill template — fields, strategy, storage key and version. | | `cogneris_templates_register` | `mcp.write` | Registers a fill template from an uploaded PDF and returns its id and detected field count. Not idempotent. | | `cogneris_sources_upload` | `mcp.write` | Uploads a base64 PDF as the source of a fill job or as a template PDF. At most 10 MB decoded. Not idempotent. | | `cogneris_jobs_create` | `mcp.write` | Starts a billable fill job: extract the source PDF, map its fields, fill the template. Every call creates a new job and consumes credits. | | `cogneris_jobs_get` | `mcp.read` | Status and result metadata of one fill job; a completed job carries the artifact id to download. | | `cogneris_jobs_list` | `mcp.read` | Lists fill jobs with optional status and template filters. | | `cogneris_artifacts_download` | `mcp.read` | A signed URL for the filled PDF, valid for 5 minutes. | ## Transport & discovery Stateless streamable HTTP. Worth knowing if you are writing the client yourself. - Stateless by design — there is no `Mcp-Session-Id` to carry. Every request stands on its own, which is what lets the service scale to zero between calls. - JSON-RPC at the root — POST to `/`. There is no `/mcp` sub-path; a client configured with one gets a 404. - Responses are SSE-framed — the reply comes back as `text/event-stream`, so the request must advertise `Accept: application/json, text/event-stream`. - A missing or rejected credential is a 401 — with `WWW-Authenticate: Bearer resource_metadata="…"`, never an empty tool list. - Two anonymous discovery documents — `/.well-known/mcp.json` publishes the protocol version, the server identity and the supported scopes; `/.well-known/oauth-protected-resource` publishes the RFC 9728 metadata. Neither needs a token. - Health is public too — `GET /health` answers without credentials, which is what the regional status probes read. ## Before you wire it up Behaviours that surprise people once an agent is driving. - API keys are sandbox by default — `cogneris_apikeys_create` makes a sandbox key unless you explicitly confirm a production one. Keys created from the editor never carry `apikeys.manage`, `mcp.read` or `mcp.write`. - Secrets are handed out once — a new API key, webhook signing secret or rotated secret appears in one answer only. The assistant is told to write it to your git-ignored `.env` and never repeat it in chat, code or commits. - Fill jobs cost money — `cogneris_jobs_create` consumes credits on your plan. An agent looping over a folder spends real budget, so cap it in the prompt as well as in the tenant cost cap. - Creating is not idempotent — creating a classifier, adding samples, registering a template, uploading a source, starting a fill job, or creating a webhook or an API key each make a new record per call. A retried tool call is a duplicate, not a no-op. Workflow writes, workflow runs and the Build AI steps that cost credits are the exception: a retry with the same input is answered once. - Publishing asks first — new runs use a published workflow version, so the assistant shows you the graph and publishes only after you say yes. Deleting a workflow and promoting it from sandbox are not available over MCP. - Build AI spends credits and needs an admin — the plan costs 10 to 30 credits and a revision 5 to 15, including the automatic research and revisions. It runs on production developer tokens that carry `builder.run`, owned by a tenant administrator; a sandbox token is refused. - Uploads cap at 10 MB decoded — that is the size after base64 decoding, so budget roughly 7.5 MB of file. Larger sources go through the REST upload path instead. - Signed URLs expire in 5 minutes — fetch the artifact when you get the URL. Storing it for later is storing a dead link. - Usage ranges are UTC and half-open — `to` must be strictly greater than `from`; reversed or equal ranges are rejected rather than silently clamped. ### Next: Authentication How keys, scopes and signatures work across the rest of the platform. [ Read Authentication ](https://docs.cogneris.ai/docs/api/authentication/) ## Source: compare-rossum.html [Skip to content](#main) Compare # Cogneris vs Rossum Rossum is a mature, invoice-first document AI platform serving large enterprises. Cogneris is a multi-document, ReAct-architecture platform with pay-per-page pricing. Here's how they compare for buyers in 2026. ## The short version Cogneris is a multi-document, ReAct-architecture document AI platform with pay-per-page pricing, multi-LLM extraction, and a complete platform stack including review queues, per-field confidence and audit trail. Rossum is a mature, invoice-first document AI platform with strong enterprise penetration, EU HQ, and on-prem deployment options. ## Capability comparison | Capability | Cogneris | Rossum | | Primary positioning | Multi-document, agentic ReAct architecture | Invoice automation specialist, expanding to other document types | | Pricing model | Pay-per-page, no setup fee, monthly minimum on usage tiers | Enterprise contracts, typically 6-figure annual commits | | Document type coverage | Tenant-configurable classification and zero-shot extraction | Strong on invoices and finance docs; lighter on HR, contracts, ID | | Custom model training | Few-shot config + active-learning queue, no separate training step | Dedicated training pipeline with annotation tooling | | LLM strategy | Multi-provider (OpenAI + Anthropic) with zero-retention configured | Proprietary deep-learning models, less LLM-forward | | Time to first extraction | Minutes — 6 doc types live on Starter | Weeks — annotation and training cycle for custom docs | | Compliance posture | SOC 2 Type II in progress, GDPR DPA, CCPA service-provider, HIPAA BAA on Enterprise | ISO 27001 certified, SOC 2, GDPR DPA, EU HQ | | Hosting | GCP SaaS; single-tenant VPC is a roadmap capability | EU and US options, on-prem available on Enterprise | | Audit trail | Configurable tamper-evident metadata with model version and prompt/response hashes | Activity log focused on user actions | | API ergonomics | REST + webhooks, OpenAPI spec, sync & async modes | REST + webhooks, mature SDK ecosystem | | Best fit | AI-native teams shipping docs into product workflows | Enterprise AP/AR teams modernising legacy invoice processing | ## When to choose which ### Choose Rossum when… - You're a Fortune-500 AP team with a multi-year invoice-automation budget and need a vendor with deep AP-process expertise. - You need on-prem deployment as a hard contractual requirement for regulatory or security reasons. - You need ISO 27001 today, not in 12 months — Rossum has it and we don't yet. - You're standardising on EU data residency with EU-only inference paths. ### Choose Cogneris when… - You process more than invoices — payslips, statements, contracts, IDs, KYC docs in the same workflow. - You want to start today with pay-per-page pricing and no annual commit. - You need agentic / ReAct workflows — multi-step reasoning where extraction is one step in a longer chain. - You're embedding document AI in your own product rather than standing up an AP automation team. - You need configurable LLM audit metadata — model version plus prompt and response hashes where the workflow supports it. ## Invoice extraction quality Rossum's invoice-specific accuracy on European invoices is excellent and battle-tested over a decade. Cogneris hits comparable accuracy on US invoices via multi-LLM consensus extraction, and its architecture extends to non-invoice documents without separate training pipelines. If your sole use case is European AP invoice processing at high volume, Rossum has the home-court advantage. If you have any other document type in scope, Cogneris's generality starts to win. ## Total cost of ownership Rossum contracts are typically structured as annual commits with implementation services. Total first-year cost for a mid-market deployment usually lands in the high six figures. Cogneris's pay-per-page pricing means a 100k-document/year deployment lands in the low five figures, with no implementation services required for the standard document types. The TCO crossover happens around 5M documents/year — below that, Cogneris is significantly cheaper; above it, the curves flatten. ## Deployment options Rossum offers on-prem deployment, which Cogneris does not. Cogneris currently offers GCP SaaS; single-tenant VPC and customer-managed keys are roadmap capabilities, not generally available services. If your security policy requires private deployment today, verify live capability before selection. ## Switching considerations If you're evaluating Cogneris against Rossum as your incumbent, the practical pieces matter: - Schema portability — Cogneris emits clean JSON; if your pipeline already consumes Rossum's output, mapping is typically a one-day translation layer. - Side-by-side evaluation — we run a 7-day evaluation against your real documents alongside your existing Rossum workflow. You get an accuracy and latency report you can show your CTO. - No annual commit — start with a month-to-month plan, scale up as confidence builds. The pay-per-page pricing means you only pay for what you actually process. For the full security and compliance posture, see our [trust page](security.html). For pricing, see [pricing](pricing.html). For a side-by-side evaluation against your current workflow, [talk to our team](contact.html). ## Rossum alternative FAQ Is Cogneris a Rossum alternative? Yes. Cogneris is a Rossum alternative for teams that want document extraction, validation, review routing, webhooks, and audit evidence in one API-first workflow. When should I choose Cogneris instead of Rossum? Choose Cogneris when you need broader document coverage, schema control, pay-per-page economics, and lightweight implementation instead of an invoice-first enterprise deployment. Choose Rossum when its native product model, cloud ecosystem, or prebuilt workflow is the better fit for your operating constraints. How should I evaluate Rossum against Cogneris? Run the same representative documents through both systems and compare field accuracy, citation quality, schema stability, review effort, webhook behavior, latency, and total production work. Switching from Rossum? ### Talk to our team about a side-by-side evaluation We'll run your real documents through Cogneris with a 7-day evaluation against your current Rossum workflow. No commitment, side-by-side accuracy and latency report at the end. [Book an evaluation](contact.html) [See pricing](pricing.html) ## Source: compare-docsumo.html [Skip to content](#main) Compare # Cogneris vs Docsumo Docsumo is a strong mid-market document AI platform with pre-trained models for finance documents. Cogneris is a multi-document, ReAct-architecture platform with stronger LLM tooling. Here's how they compare. ## The short version Cogneris is a multi-document, ReAct-architecture document AI platform with pay-per-page pricing, multi-LLM extraction, and a complete platform stack including review queues, per-field confidence and audit trail. Docsumo is a strong mid-market document AI platform with pre-trained models for finance documents and a custom-training UI. ## Capability comparison | Capability | Cogneris | Docsumo | | Primary positioning | Multi-document, agentic ReAct architecture, LLM-native | Pre-trained finance documents (invoices, bank statements, KYC) with custom training | | Pricing model | Credits from €0/mo: Free, Pro €99, Scale €499, published rates | Free tier + paid tiers from ~$300/mo, custom enterprise | | Document type coverage | Tenant-configurable classification and zero-shot extraction | Strong on financial docs, growing into other verticals | | Custom model training | No model training: tenant-configurable classification and zero-shot extraction | Dedicated training UI with sample annotation | | LLM strategy | Multi-LLM (OpenAI + Anthropic), zero-retention | ML pipeline with LLM augmentation, less LLM-forward | | Compliance posture | SOC 2 Type II in progress, GDPR DPA, CCPA, HIPAA BAA on Enterprise | SOC 2, GDPR, ISO 27001 | | Workflow tooling | Review queues, validation rules, webhook callbacks, per-field confidence | Review interface, validation rules, integrations marketplace | | Audit trail | Immutable per-request log with model version + prompt hash | Activity log per document | | Hosting | GCP SaaS; single-tenant VPC is a roadmap capability | AWS, multiple regions | | Best fit | AI-native teams shipping docs into product workflows | Mid-market finance and ops teams modernising data entry | ## When to choose which ### Choose Docsumo when… - You're already on AWS and want a vendor that lives on the same cloud as your stack. - You need ISO 27001 today — Docsumo has it. - Your primary documents are invoices, bank statements and KYC exactly — Docsumo's pre-trained models hit these out of the box. - You want a free tier for very low volumes (Docsumo has one; Cogneris's Starter tier has a low monthly fee). ### Choose Cogneris when… - You need agentic / ReAct workflows where extraction is one step in a multi-step reasoning chain. - You're processing many document types beyond Docsumo's finance-doc focus. - You want LLM-native flexibility — multi-provider routing, custom prompts, model versioning controls. - You need configurable audit metadata with model version, prompt and response hashes, and post-processing decisions. - You're embedding extraction in your own product rather than running it through a manual review queue. ## Pre-trained models vs flexible prompts Docsumo's pre-trained models for invoices and bank statements are tight and accurate out of the box, especially for US documents. Cogneris's approach is different: a small set of well-tuned base prompts plus few-shot examples that customers control. The result is comparable accuracy on common docs but more flexibility on edge cases — a customer can ship a new field tomorrow without waiting for a model retrain. If you process mostly common docs and want "it just works," Docsumo is excellent. If you have any document with unusual structure, Cogneris bends rather than breaks. ## Workflow vs API Docsumo includes a substantial review interface as part of the product — operators log in, see queues, validate extractions, push approved data downstream. Cogneris provides the same primitives via API but doesn't ship a full review UI; customers either build their own (most do) or use the included minimal queue. If you want an out-of-the-box review UI for a non-technical operations team, Docsumo is the easier choice. If you're building a product where extraction is invisible to the end user, Cogneris's API-first approach is the better fit. ## LLM control Cogneris exposes more control over the LLM layer: per-tenant provider routing (OpenAI vs Anthropic), model version pinning, custom system prompts, prompt-versioning audit trail. Docsumo's LLM usage is more of an internal implementation detail. For most customers this doesn't matter; for AI teams that want to constrain or extend the model behavior, Cogneris's controls are meaningfully more flexible. ## Switching considerations If you're evaluating Cogneris against Docsumo as your incumbent, the practical pieces matter: - Schema portability — Cogneris emits clean JSON; if your pipeline already consumes Docsumo's output, mapping is typically a one-day translation layer. - Side-by-side evaluation — we run a 7-day evaluation against your real documents alongside your existing Docsumo workflow. You get an accuracy and latency report you can show your CTO. - No annual commit — start with a month-to-month plan, scale up as confidence builds. The pay-per-page pricing means you only pay for what you actually process. For the full security and compliance posture, see our [trust page](security.html). For pricing, see [pricing](pricing.html). For a side-by-side evaluation against your current workflow, [talk to our team](contact.html). ## Docsumo alternative FAQ Is Cogneris a Docsumo alternative? Yes. Cogneris is a Docsumo alternative for teams that want document extraction, validation, review routing, webhooks, and audit evidence in one API-first workflow. When should I choose Cogneris instead of Docsumo? Choose Cogneris when you need developer-controlled extraction APIs, citations, and portal workflows rather than a primarily operations-led document automation suite. Choose Docsumo when its native product model, cloud ecosystem, or prebuilt workflow is the better fit for your operating constraints. How should I evaluate Docsumo against Cogneris? Run the same representative documents through both systems and compare field accuracy, citation quality, schema stability, review effort, webhook behavior, latency, and total production work. Switching from Docsumo? ### Talk to our team about a side-by-side evaluation We'll run your real documents through Cogneris with a 7-day evaluation against your current Docsumo workflow. No commitment, side-by-side accuracy and latency report at the end. [Book an evaluation](contact.html) [See pricing](pricing.html) ## Source: compare-hyperscience.html [Skip to content](#main) Compare # Cogneris vs Hyperscience Hyperscience is a mature IDP platform with strong roots in government and financial services. Cogneris is a multi-document, ReAct-architecture platform with pay-per-page pricing. Here's how they compare for buyers in 2026. ## The short version Cogneris is a multi-document, ReAct-architecture document AI platform with pay-per-page pricing, multi-LLM extraction, and a complete platform stack including review queues, per-field confidence and immutable audit trail. Hyperscience is a mature IDP platform with strong enterprise penetration in government, banking, and insurance, deep human-in-the-loop tooling, and on-prem deployment as a first-class option. ## Capability comparison | Capability | Cogneris | Hyperscience | | Primary positioning | Multi-document, agentic ReAct architecture for AI-native teams | Enterprise IDP platform for highly-regulated industries (government, banking, insurance) | | Pricing model | Pay-per-page, no setup fee, monthly minimum on usage tiers | Enterprise contracts, typically high six-figure to seven-figure annual commits | | Document type coverage | Tenant-configurable classification and zero-shot extraction | Custom-trained per document type via supervised classification + extraction blocks | | Custom model training | Few-shot config + active-learning queue, no separate training step | Dedicated training pipeline with annotation tooling and supervised fine-tuning | | LLM strategy | Multi-provider (OpenAI + Anthropic) with zero-retention configured | Proprietary deep-learning models; LLM integrations available but not LLM-first | | Time to first extraction | Minutes — 6 doc types live on Starter | Weeks-to-months — annotation, training, and validation cycle for each new doc type | | Human-in-the-loop tooling | Review queues with in-place field correction, claim/release, and an audit-tracked timeline; queued by workflow step, not by confidence score | Deep HITL workflows with keyer ergonomics and annotation analytics | | Compliance posture | SOC 2 Type II in progress, GDPR DPA, CCPA service-provider, HIPAA BAA on Enterprise | SOC 2 Type II, ISO 27001, FedRAMP Moderate (HSGov), HIPAA-ready deployments | | Hosting | GCP SaaS; single-tenant VPC and CMEK are roadmap capabilities | SaaS, private-cloud, and full on-prem (air-gapped) deployments | | Audit trail | Configurable tamper-evident metadata with model version, prompt/response hashes, and reviewer identity | Comprehensive activity log with strong human-action lineage | | API ergonomics | REST + webhooks, OpenAPI 3.1 spec, sync & async modes | REST API, mature SDK ecosystem, complex but well-documented | | Best fit | AI-native teams shipping docs into product workflows, mid-market through enterprise | Large regulated enterprises with established ops teams and FedRAMP/air-gap requirements | ## When to choose which ### Choose Hyperscience when… - You're a federal agency or regulated bank with FedRAMP, air-gapped, or fully on-prem deployment as a hard requirement. - You have a dedicated keyer/ops team processing high-volume forms and need the deepest HITL ergonomics on the market. - You're standardizing on supervised models for regulatory defensibility and prefer not to depend on third-party LLM APIs in the critical path. - You need ISO 27001 and FedRAMP today, not in 12 months — Hyperscience has them and we don't yet. - Your budget supports a six- or seven-figure annual commit and a multi-month implementation. ### Choose Cogneris when… - You process more than forms — payslips, statements, contracts, IDs, KYC docs in the same workflow without separate training pipelines. - You want to start today with pay-per-page pricing and no annual commit. - You need agentic / ReAct workflows — multi-step reasoning where extraction is one step in a longer chain. - You're embedding document AI in your own product rather than standing up a back-office ops team to operate it. - You need configurable LLM audit metadata — model version plus prompt and response hashes where the workflow supports it. - You want multi-LLM consensus as a built-in accuracy lever rather than a single proprietary model. ## Low-code IDP vs API-first document AI Hyperscience is strongest when a large operations organization wants a governed IDP suite for back-office work. Cogneris is strongest when product, engineering, and compliance teams need document AI exposed as APIs, schemas, webhooks, review state, and tenant-scoped audit logs. | Decision | Cogneris | Hyperscience | | Implementation motion | API-first, schema-first, faster embed into apps and portals | Enterprise IDP rollout with operations and platform governance | | Best buyer | Engineering-led teams building document intake into products | Large back-office teams standardizing document automation | | Review model | Field-level review, citations, validation, and webhook state | Operational review queues and accuracy programs | | SLA question | Field accuracy, STP rate, false approvals, and audit completeness | Automation rate and operational accuracy at enterprise scale | ## Architecture difference: supervised classification vs. agentic reasoning The fundamental architectural difference matters for buyers. Hyperscience trains a supervised classification + extraction pipeline per document type. The pipeline is fast, predictable, and defensible — you can show an auditor exactly which model version produced each field and what training data it saw. The cost is the cycle time: each new document type requires annotation, training, validation, and deployment. Cogneris's ReAct architecture treats extraction as a multi-step reasoning task: classify, plan, extract, validate, escalate. New document types are tenant-configurable and can use zero-shot extraction rather than a maintained catalog. Supported workflows can record tamper-evident metadata with model version plus prompt and response hashes; coverage must be verified for the configured service. Neither architecture is universally better. If your document set is fixed and your accuracy bar is "regulator-defensible to within a basis point," Hyperscience's supervised approach is hard to beat. If your document set is growing and your team is small, Cogneris's general-purpose approach gets you to production faster. ## Total cost of ownership Hyperscience contracts are typically structured as multi-year commits with substantial implementation services — annotation work, custom training pipelines, and integration. Total first-year TCO for a mid-market deployment usually lands in the high six to low seven figures, with most cost in the first 12 months. Cogneris's pay-per-page pricing means a 100k-document/year deployment lands in the low five figures, with no implementation services required for the standard document types. The TCO crossover happens around 8–10M documents/year of supervised forms — below that, Cogneris is significantly cheaper; above it (and especially with FedRAMP requirements), the curves narrow. ## Deployment options Hyperscience offers fully air-gapped on-prem deployment, which Cogneris does not. Cogneris currently offers GCP SaaS; single-tenant VPC and customer-managed keys are roadmap capabilities, not generally available services. If your security policy requires private deployment or air-gap today, verify a vendor's live capability before selection. ## Switching considerations If you're evaluating Cogneris against Hyperscience as your incumbent, the practical pieces matter: - Schema portability — Cogneris emits clean JSON; if your pipeline already consumes Hyperscience extraction blocks, mapping is typically a 1- to 2-day translation layer. - Side-by-side evaluation — we run a 7-day evaluation against your real documents alongside your existing Hyperscience workflow. You get an accuracy and latency report you can show your CTO. - No annual commit — start with a month-to-month plan. The pay-per-page pricing means you only pay for what you actually process. - Coexistence — many teams run both for a transition period, with Hyperscience handling regulator-defensible forms and Cogneris handling everything else (contracts, IDs, statements, ad-hoc docs). Coexistence is usually easier to sell internally than a hard switch. For the full security and compliance posture, see our [trust page](security.html). For pricing, see [pricing](pricing.html). For a side-by-side evaluation against your current workflow, [talk to our team](contact.html). Switching from Hyperscience? ### Talk to our team about a side-by-side evaluation We'll run your real documents through Cogneris with a 7-day evaluation against your current Hyperscience workflow. No commitment, side-by-side accuracy and latency report at the end. [Book an evaluation](contact.html) [See pricing](pricing.html) ## Source: compare-google-document-ai.html [Skip to content](#main) Compare # Cogneris vs Google Document AI Google Document AI is a powerful OCR + parsing service inside Google Cloud. Cogneris is a complete document AI platform with workflows, audit trail, and multi-LLM extraction on top. Here's how they compare. ## The short version Cogneris is a multi-document, ReAct-architecture document AI platform with pay-per-page pricing, multi-LLM extraction, and a complete platform stack including review queues, per-field confidence and audit trail. Google Document AI is Google Cloud's document extraction service — strong primitives, full compliance footprint, and tight GCP integration, but you assemble the platform yourself. ## Capability comparison | Capability | Cogneris | Google Document AI | | Product type | Complete platform — extraction, validation, workflows, audit trail | Cloud service — extraction APIs you assemble into a workflow | | Pricing model | Pay-per-page, includes review & audit features | Pay-per-page, additional GCP costs for storage / KMS / orchestration | | Document classes | Tenant-configurable classification and zero-shot extraction | 20+ specialized parsers (invoice, receipt, W-2, 1040, etc.) | | LLM strategy | Multi-LLM (OpenAI + Anthropic), zero-retention configured, model-version pinning | Gemini integration available, Google models only | | Workflow / orchestration | Built-in: review queues, validation rules, webhook callbacks, per-field confidence | Bring your own — Cloud Workflows or external orchestrator | | Audit trail | Immutable per-request log with prompt hash, model version, response | Cloud Logging integration; you build the schema | | Custom model training | No model training: tenant-configurable classification and zero-shot extraction | Custom model training in Document AI Workbench | | Compliance | SOC 2 Type II in progress, GDPR DPA, CCPA, HIPAA BAA on Enterprise | Inherits Google Cloud's compliance certifications (extensive) | | Vendor lock-in profile | Standard REST API, portable schema | GCP ecosystem; portability requires migration work | | Best fit | Teams that want a complete platform with batteries included | Teams already deeply on GCP that want raw building blocks | ## When to choose which ### Choose Google Document AI when… - You're already deep in GCP and want a service that lives next to your data and IAM. - You want Google's full compliance stack (FedRAMP, IL5, etc.) without negotiating it from a smaller vendor. - You're building entirely on Gemini and want LLM lineage tied to one provider's model family. - You have engineers to assemble the platform — workflow, review UI, audit schema, alerting — yourself. ### Choose Cogneris when… - You want a complete platform with review queues, validation, and audit trail out of the box. - You need multi-LLM — different providers for different document types, or failover between them. - You want vendor-portable extraction rather than a GCP-native one. - You don't have an engineering team to build review interfaces and audit pipelines on top of a raw API. - You need an audit trail with full LLM lineage per request (prompt, model version, response) without a custom logging build-out. ## Platform vs primitive This is the core distinction. Google Document AI is excellent at the extraction primitive — pull structured data from a document, fast and accurate. It is not a complete document-AI platform. To use it in production you need to build (or buy) review queues, validation rules, audit-trail schemas, alerting, and orchestration. Cogneris bundles those as the product. If you have a strong platform team that prefers raw building blocks, Document AI gives you maximum control. If you'd rather not build the platform, Cogneris is a faster path to production. ## Google processors and pricing notes Google Document AI currently separates work by processor: OCR for text, Layout Parser for structure, Custom Extractor for trained fields, Custom Classifier for routing, and Splitter for multi-document files. Pricing is processor-specific and usually quoted per 1,000 pages, so compare the end-to-end workflow, not only the first extraction call. | Google Document AI processor | Typical buyer intent | Cogneris equivalent | Cost/workflow watch-out | | OCR processor | Recognize text from PDFs, scans, and images. | [OCR API](ocr-api.html) plus structured extraction when needed. | OCR alone still leaves field mapping, validation, review, and downstream JSON work. | | Layout Parser | Preserve layout, tables, and blocks for RAG or search. | [Document AI for RAG](document-ai-for-rag.html) and [parsing API](document-parsing-api.html). | Layout-aware output is useful, but business workflows still need schemas and citations. | | Custom Extractor | Train extraction for custom documents. | [Schema-based extraction API](document-extraction-api.html). | Training effort, evaluation sets, and schema versioning become operating costs. | | Custom Classifier | Route documents by type. | [Classification docs](https://docs.cogneris.ai/docs/api/document-classifier/). | Classifier output still needs packet splitting, exception paths, and audit state. | | Document Splitter | Split packets into component documents. | [Document-type routing](document-types.html) and async extraction jobs. | Packet workflows need per-document status, reviewer queues, and webhook retries. | ## Compliance & certifications Google Cloud's compliance footprint is unmatched — FedRAMP High, ISO 27001, SOC 2, HITRUST, IL5 in some configurations. Cogneris is currently SOC 2 Type II in progress with GDPR DPA, CCPA service-provider framing, and HIPAA BAA on Enterprise. For deeply regulated workloads (federal, defense), Document AI is the answer today. For most commercial enterprise workloads, Cogneris's posture is sufficient and getting stronger. ## Multi-LLM flexibility Google Document AI uses Google's models — Gemini and Google's specialized parsers. Cogneris routes to OpenAI, Anthropic, or both, configurable per tenant or per workflow. Customers who want provider-portable extraction (or who have explicit provider restrictions for compliance) get more flexibility from Cogneris. Customers who are happy on Gemini get tighter integration from Document AI. ## Switching considerations If you're evaluating Cogneris against Google Document AI as your incumbent, the practical pieces matter: - Schema portability — Cogneris emits clean JSON; if your pipeline already consumes Google Document AI's output, mapping is typically a one-day translation layer. - Side-by-side evaluation — we run a 7-day evaluation against your real documents alongside your existing Google Document AI workflow. You get an accuracy and latency report you can show your CTO. - No annual commit — start with a month-to-month plan, scale up as confidence builds. The pay-per-page pricing means you only pay for what you actually process. For the full security and compliance posture, see our [trust page](security.html). For pricing, see [pricing](pricing.html). For a side-by-side evaluation against your current workflow, [talk to our team](contact.html). ## Google Document AI alternative FAQ Is Cogneris a Google Document AI alternative? Yes. Cogneris is a Google Document AI alternative for teams that want document extraction, validation, review routing, webhooks, and audit evidence in one API-first workflow. When should I choose Cogneris instead of Google Document AI? Choose Cogneris when you need a productized workflow layer on top of extraction primitives rather than assembling review queues, validation rules, and audit logs in GCP. Choose Google Document AI when its native product model, cloud ecosystem, or prebuilt workflow is the better fit for your operating constraints. How should I evaluate Google Document AI against Cogneris? Run the same representative documents through both systems and compare field accuracy, citation quality, schema stability, review effort, webhook behavior, latency, and total production work. Switching from Google Document AI? ### Talk to our team about a side-by-side evaluation We'll run your real documents through Cogneris with a 7-day evaluation against your current Google Document AI workflow. No commitment, side-by-side accuracy and latency report at the end. [Book an evaluation](contact.html) [See pricing](pricing.html) ## Source: compare-azure-document-intelligence.html [Skip to content](#main) Compare # Cogneris vs Azure Document Intelligence Azure Document Intelligence is now positioned inside Microsoft Foundry Tools and Azure Content Understanding. Cogneris is a complete document AI platform with workflows, multi-LLM extraction, and audit trail. Here's how they compare. ## The short version Cogneris is a multi-document, ReAct-architecture document AI platform with pay-per-page pricing, multi-LLM extraction, and a complete platform stack including review queues, per-field confidence and audit trail. Azure Document Intelligence is Microsoft's document extraction service inside Azure, now framed as a Foundry Tool and part of Azure Content Understanding. It has strong primitives and Microsoft 365 integration, with Azure OpenAI and Content Understanding for broader multimodal workflows. ## Capability comparison | Capability | Cogneris | Azure Document Intelligence | | Product type | Complete platform — extraction, validation, workflows, audit trail | Foundry Tool / Azure service — extraction APIs and Content Understanding building blocks you compose into a workflow | | Pricing model | Pay-per-page, includes review & audit features | Pay-per-page, additional Azure costs for storage / Key Vault / orchestration | | Document classes | Tenant-configurable classification and zero-shot extraction | Pre-built models for invoices, receipts, ID, W-2, 1040, layout, general forms | | LLM strategy | Multi-LLM (OpenAI + Anthropic), zero-retention configured | Azure OpenAI and Microsoft Foundry-native analyzers, including Content Understanding for broader content types | | Workflow / orchestration | Built-in: review queues, validation rules, webhook callbacks, per-field confidence | Bring your own — Logic Apps, Functions, or external orchestrator | | Custom model training | No model training: tenant-configurable classification and zero-shot extraction | Custom training in Document Intelligence Studio | | Audit trail | Immutable per-request log with prompt hash, model version, response | Azure Monitor integration; you design the schema | | Compliance | SOC 2 Type II in progress, GDPR DPA, CCPA, HIPAA BAA on Enterprise | Inherits Azure's compliance posture (extensive) | | Vendor lock-in profile | Standard REST API, portable schema | Azure ecosystem; portability requires migration work | | Best fit | Teams that want a complete platform with batteries included | Teams deep on Azure that want raw building blocks | ## When to choose which ### Choose Azure Document Intelligence when… - You're already on Azure and want a service that lives next to your IAM, Key Vault and data. - You're standardising on Microsoft 365, Power Platform, or Azure AI Foundry — Azure DI integrates natively. - You need Azure's specific compliance certs (FedRAMP, IL5, GovCloud) for federal workloads. - You have engineers to assemble the platform — review UI, audit schema, orchestration — yourself. ### Choose Cogneris when… - You want a complete platform with review queues, validation rules, audit trail, and orchestration out of the box. - You need multi-LLM across providers, not just Azure OpenAI. - You want vendor-portable extraction that doesn't lock you into Azure. - You need full audit lineage per request (prompt, model version, response) without a custom logging pipeline. - You're embedding document AI in a product shipped on a different cloud than Azure. ## Platform vs Foundry primitive Azure Document Intelligence excels at the extraction primitive — pull structured data from a form or document, with strong accuracy on invoices and US tax forms. The 2026 positioning matters: Microsoft now routes buyers through Foundry Tools and Content Understanding, which broadens the story to multimodal analyzers and agent pipelines. That is powerful if your platform team wants Azure-native building blocks. Cogneris remains the faster path when you need review queues, validation rules, audit-trail schemas, alerting, and orchestration bundled into the product. ## Azure prebuilt models vs Cogneris pages As of July 2026, Azure Document Intelligence still positions prebuilt models as the fast path for common forms, while Azure Content Understanding expands the Microsoft story to complex, unstructured, and multimodal content. That is useful when the model matches the document exactly or your team wants Foundry-native analyzers. Cogneris maps the same buyer intent to document-specific pages, schemas, validation rules, review routing, and audit evidence. | Azure prebuilt model intent | Cogneris target page | When Azure is enough | When workflow/review is needed | | Invoice | [Invoice extraction API](extract-invoices.html) | Single invoice parse into an Azure-native app. | AP approval, PO matching, duplicate checks, GL hints, ERP handoff. | | Receipt | [Receipt extraction API](extract-receipts.html) | Expense receipt field capture. | Policy review, duplicate detection, merchant normalization, T&E routing. | | Bank statement | [Bank statement extraction API](extract-bank-statements.html) | Statement text and transaction extraction. | Lending, reconciliation, multi-page stitching, reviewer evidence. | | ID document | [ID card OCR API](id-card-ocr-api.html) | Basic identity field extraction. | KYC portals, proof-of-address checks, fraud signals, exception handling. | | Tax and paystub forms | [Tax form extraction](extract-tax-returns.html) / [Payroll extraction](extract-payroll.html) | Known US tax/payroll forms. | Underwriting packets, income verification, audit-ready reviewer changes. | | Mortgage documents | [Mortgage document extraction](mortgage-document-extraction-api.html) | Azure-first loan document parsing. | Loan packets with multiple document types, missing evidence, and review state. | | Layout and general document | [Document parsing API](document-parsing-api.html) | Raw layout extraction for an internal Azure pipeline. | Schema output, citations, validation, and workflow handoff. | ## LLM strategy Azure DI's LLM story is now broader than Azure OpenAI alone: Document Intelligence, Azure AI Foundry, and Content Understanding can work together in Microsoft-native agent and analyzer pipelines. Cogneris routes to OpenAI direct, Anthropic, or both, configurable per tenant. Customers who want provider failover or a non-Microsoft control plane find Cogneris easier to reason about. Customers happy on Azure OpenAI and Foundry get tighter integration from Azure DI. ## Compliance & certifications Azure's compliance footprint is comprehensive — FedRAMP, ISO 27001, SOC 2, HIPAA, IL4/IL5, GovCloud. Cogneris is currently SOC 2 Type II in progress with GDPR DPA, CCPA service-provider framing, and HIPAA BAA on Enterprise. For federal or defense workloads requiring IL5+, Azure DI is the answer. For commercial enterprise workloads, Cogneris's posture is sufficient and broadening. ## Switching considerations If you're evaluating Cogneris against Azure Document Intelligence as your incumbent, the practical pieces matter: - Schema portability — Cogneris emits clean JSON; if your pipeline already consumes Azure Document Intelligence's output, mapping is typically a one-day translation layer. - Side-by-side evaluation — we run a 7-day evaluation against your real documents alongside your existing Azure Document Intelligence workflow. You get an accuracy and latency report you can show your CTO. - No annual commit — start with a month-to-month plan, scale up as confidence builds. The pay-per-page pricing means you only pay for what you actually process. For the full security and compliance posture, see our [trust page](security.html). For pricing, see [pricing](pricing.html). For a side-by-side evaluation against your current workflow, [talk to our team](contact.html). ## Azure Document Intelligence alternative FAQ Is Cogneris an Azure Document Intelligence alternative? Yes. Cogneris is an Azure Document Intelligence alternative for teams that want document extraction, validation, review routing, webhooks, and audit evidence in one API-first workflow. When should I choose Cogneris instead of Azure Document Intelligence? Choose Cogneris when you need an API-first document AI workflow outside a Microsoft-only architecture, or you want portal, review, and audit behavior bundled with extraction. Choose Azure Document Intelligence when its native product model, cloud ecosystem, or prebuilt workflow is the better fit for your operating constraints. How should I evaluate Azure Document Intelligence against Cogneris? Run the same representative documents through both systems and compare field accuracy, citation quality, schema stability, review effort, webhook behavior, latency, and total production work. Switching from Azure Document Intelligence? ### Talk to our team about a side-by-side evaluation We'll run your real documents through Cogneris with a 7-day evaluation against your current Azure Document Intelligence workflow. No commitment, side-by-side accuracy and latency report at the end. [Book an evaluation](contact.html) [See pricing](pricing.html) ## Source: best-document-ai-platforms.html [Skip to content](#main) Buyer guide # Best document AI platforms for 2026. A practical shortlist for teams comparing extraction APIs, OCR services, IDP platforms, document automation portals, and agentic document workflows. [Talk to our team](contact.html)[Browse comparisons](compare.html) ## The July 2026 shortlist | Platform | Best fit | Watch-out | | Cogneris | API-first document AI with validation, review, portals, and audit trail | Not the cheapest raw OCR primitive | | Google Document AI | GCP teams that want hyperscaler extraction primitives | You build much of the workflow layer | | Azure Document Intelligence / Content Understanding | Microsoft-first enterprises and Azure AI Foundry stacks | Platform lock-in and orchestration work | | UiPath IXP | UiPath automation estates needing document and communications extraction | Suite fit, licensing shape, and non-UiPath integration ownership | | Tungsten/Kofax | Large operations teams needing mature IDP and workflow orchestration | Implementation weight and suite-specific customization | | AWS Textract | AWS-native OCR, forms, and table extraction | Review, validation, and UX are on you | | Rossum, Docsumo, ABBYY, Hyperscience | Operations teams buying mature IDP suites | Often heavier implementation and commercial motion | | Nanonets, Mindee, Veryfi, Klippa / Doxis | API-first OCR, invoices, receipts, and document-specific extraction | Check review, audit, and multi-document workflow depth | | Reducto, LandingAI ADE, LlamaParse, Mistral OCR, Docling, Tensorlake | Modern API, agentic extraction, parsing, and RAG evaluations | Validate production workflow controls, not only parsing quality | | Unstructured | Document parsing and ETL for RAG and GenAI pipelines | Structured business extraction may still need a workflow layer | ## Emerging API-native and RAG category The fastest-moving group in July 2026 is not classic IDP. It is API-native document AI: schema extraction, document parsing for RAG, visual grounding, agentic extraction, AI-ready and LLM-ready document data, and developer-first OCR. Reducto, Mistral OCR, LandingAI ADE, LlamaParse, Docling, Tensorlake, Firecrawl, and DocuPipe belong in the shortlist when buyers care about chunks, layout, citations, parsing quality, and developer control. Buyers should still compare SDK quality, schema stability, async jobs, webhook retries, validation, human review, and audit evidence before committing to a vendor. ### Agentic extraction LandingAI ADE, UiPath IXP, and similar vendors compete on schema-driven, agentic extraction. Validate citations, review controls, and async behavior. ### RAG parsing Unstructured, LlamaParse, Reducto, Mistral OCR, Docling, and Tensorlake often appear when teams need chunks, tables, layout, and retrieval-ready document data. ### Developer OCR APIs Mindee, Veryfi, Klippa / Doxis, and Nanonets compete on fast integration, SDKs, sample outputs, and document-specific APIs. ## How to choose Shortlist by workflow, not logo size. If you only need text, OCR APIs are enough. If you need business fields with evidence, choose schema extraction. If the extracted data triggers decisions, require validation, review queues, citations, and audit logs. ## Comparison links [UiPath IXP](compare-uipath-ixp.html)[Tungsten/Kofax](compare-tungsten-kofax.html)[Google](compare-google-document-ai.html)[Azure](compare-azure-document-intelligence.html)[AWS Textract](compare-aws-textract.html)[Reducto](compare-reducto.html)[LandingAI ADE](compare-landingai-ade.html)[LlamaParse](compare-llamaparse.html)[Mistral OCR](compare-mistral-ocr.html)[Unstructured](compare-unstructured.html)[Nanonets](compare-nanonets.html)[ABBYY](compare-abbyy-vantage.html) ## Source: best-ocr-apis-for-developers.html [Skip to content](#main) Developer guide # Best OCR APIs for developers. Choosing an OCR API is less about text recognition alone and more about what your app can safely do with the result: parse, validate, review, and audit. [Open OCR API](ocr-api.html)[API reference](api.html) ## What developers should compare OCR accuracy matters, but production teams should also compare table extraction, JSON schemas, confidence calibration, webhook support, SDK quality, rate limits, evidence links, pricing, and whether the API can route uncertain fields to review. | Vendor type | Good fit | Watch closely | | Cloud OCR primitives | High-scale text and table recognition | You build validation, review, and audit trail | | Developer extraction APIs | Invoices, receipts, IDs, bank statements, and typed JSON | Custom document support and citation quality | | IDP platforms | Workflow-heavy operations teams | Implementation time, seats, minimums, and API ergonomics | | Cogneris | API-first extraction with validation, citations, and review routing | Best fit when structured output matters more than raw text | ## OCR API checklist Before buying, run a benchmark with real documents. Include scanned PDFs, mobile photos, tables, handwriting, checkboxes, long packets, missing fields, and documents that should fail validation. Measure clean-through rate, reviewer time, webhook reliability, schema stability, and cost per resolved case. ## Related pages [OCR API](ocr-api.html)[Document extraction API](document-extraction-api.html)[Benchmark guide](document-extraction-benchmark.html)[Pricing comparison](document-ai-pricing-comparison.html)[AWS Textract comparison](compare-aws-textract.html) ## Source: compare-abbyy-vantage.html [Skip to content](#main) Compare # Cogneris vs ABBYY Vantage. A practical comparison for teams evaluating ABBYY Vantage against Cogneris for document extraction APIs, intelligent document processing, and document automation portals. [Talk to our team](contact.html)[See all comparisons](compare.html) ## The short version ABBYY Vantage is strongest around low-code enterprise IDP, OCR, and document skills. Cogneris is built for teams that need document AI to become production workflow infrastructure: schema-based extraction, citations, validation, human review, webhook delivery, tenant controls, and audit trail. | Capability | Cogneris | ABBYY Vantage | | Primary fit | Document AI platform for API-first workflows and portals | low-code enterprise IDP, OCR, and document skills | | Extraction output | Typed JSON, confidence, citations, validation, audit metadata | Structured extraction with vendor-specific strengths | | Workflow layer | Review queues, portal intake, reminders, webhooks, QA state, per-field confidence | Varies by product and deployment | | Engineering control | REST API, schemas, async jobs, validation rules, per-tenant controls | Strong where its product model matches your use case | | Best buyer | Engineering, product, operations, and compliance teams sharing one document workflow | Teams with a use case that maps tightly to ABBYY Vantage's core product | ## When to choose which Choose ABBYY Vantage when a large operations team wants low-code document skills inside a mature IDP suite. Choose Cogneris when an engineering team wants API-first extraction, faster deployment, and explicit LLM workflow control. ## Low-code IDP vs API-first document AI ABBYY Vantage fits teams that want analysts and operations users to configure document skills in a low-code environment. Cogneris fits teams that want extraction to behave like product infrastructure: versioned schemas, REST APIs, an OpenAPI 3.1 spec, webhook delivery, validation rules, and auditable review state. | Decision | Low-code IDP | API-first document AI | | Who configures it | Operations analysts and IDP admins | Engineering, product, and workflow owners | | Best surface | Back-office queues and document-skill configuration | Embedded product workflows, portals, and automation APIs | | Change control | Skill and template changes managed in the IDP suite | Schema versions, test fixtures, CI checks, and API contracts | | Audit focus | Operational review history | Field evidence, validation state, prompt/model lineage, reviewer identity | ### Choose Cogneris when You need extraction, validation, review, audit trail, and portal workflow in one API-first platform. ### Choose ABBYY Vantage when Your use case maps directly to ABBYY Vantage's strongest product surface and you already accept its operating model. ### Evaluation tip Test with your hardest 25 documents, not a demo set. Compare field accuracy, citation quality, latency, review effort, and total platform work. ## Questions to ask during evaluation Ask whether the platform returns source citations with every field, how schema changes are versioned, what happens to low-confidence fields, how webhook retries are signed, and whether audit logs include model version, prompt version, reviewer ID, and validation status. ## Related comparisons [Reducto](compare-reducto.html)[LandingAI ADE](compare-landingai-ade.html)[Nanonets](compare-nanonets.html)[AWS Textract](compare-aws-textract.html)[Google Document AI](compare-google-document-ai.html)[Azure DI](compare-azure-document-intelligence.html) ## ABBYY Vantage alternative FAQ Is Cogneris an ABBYY Vantage alternative? Yes. Cogneris is an ABBYY Vantage alternative for teams that want document extraction, validation, review routing, webhooks, and audit evidence in one API-first workflow. When should I choose Cogneris instead of ABBYY Vantage? Choose Cogneris when you prefer API-first implementation, schema control, and faster product integration over a low-code enterprise IDP suite. Choose ABBYY Vantage when its native product model, cloud ecosystem, or prebuilt workflow is the better fit for your operating constraints. How should I evaluate ABBYY Vantage against Cogneris? Run the same representative documents through both systems and compare field accuracy, citation quality, schema stability, review effort, webhook behavior, latency, and total production work. ## Source: compare-aws-textract.html [Skip to content](#main) Compare # Cogneris vs AWS Textract. A practical comparison for teams evaluating AWS Textract against Cogneris for document extraction APIs, intelligent document processing, and document automation portals. [Talk to our team](contact.html)[See all comparisons](compare.html) ## The short version AWS Textract is strongest around OCR, forms, tables, and AWS-native document analysis. Cogneris is built for teams that need document AI to become production workflow infrastructure: schema-based extraction, citations, validation, human review, webhook delivery, tenant controls, and audit trail. | Capability | Cogneris | AWS Textract | | Primary fit | Document AI platform for API-first workflows and portals | OCR, forms, tables, and AWS-native document analysis | | Extraction output | Typed JSON, confidence, citations, validation, audit metadata | Structured extraction with vendor-specific strengths | | Workflow layer | Review queues, portal intake, reminders, webhooks, QA state, per-field confidence | Varies by product and deployment | | Engineering control | REST API, schemas, async jobs, validation rules, per-tenant controls | Strong where its product model matches your use case | | Best buyer | Engineering, product, operations, and compliance teams sharing one document workflow | Teams with a use case that maps tightly to AWS Textract's core product | ## When to choose which Choose AWS Textract when your team wants a raw AWS primitive and is comfortable building orchestration, validation, review UI, and audit logs. Choose Cogneris when the goal is a complete document AI workflow without assembling the platform yourself. ### Choose Cogneris when You need extraction, validation, review, audit trail, and portal workflow in one API-first platform. ### Choose AWS Textract when Your use case maps directly to AWS Textract's strongest product surface and you already accept its operating model. ### Evaluation tip Test with your hardest 25 documents, not a demo set. Compare field accuracy, citation quality, latency, review effort, and total platform work. ## Questions to ask during evaluation Ask whether the platform returns source citations with every field, how schema changes are versioned, what happens to low-confidence fields, how webhook retries are signed, and whether audit logs include model version, prompt version, reviewer ID, and validation status. ## Related comparisons [Reducto](compare-reducto.html)[LandingAI ADE](compare-landingai-ade.html)[Nanonets](compare-nanonets.html)[AWS Textract](compare-aws-textract.html)[Google Document AI](compare-google-document-ai.html)[Azure DI](compare-azure-document-intelligence.html) ## AWS Textract alternative FAQ Is Cogneris an AWS Textract alternative? Yes. Cogneris is an AWS Textract alternative for teams that want document extraction, validation, review routing, webhooks, and audit evidence in one API-first workflow. When should I choose Cogneris instead of AWS Textract? Choose Cogneris when you need structured JSON, citations, validation, review state, and workflow automation beyond OCR forms and tables. Choose AWS Textract when its native product model, cloud ecosystem, or prebuilt workflow is the better fit for your operating constraints. How should I evaluate AWS Textract against Cogneris? Run the same representative documents through both systems and compare field accuracy, citation quality, schema stability, review effort, webhook behavior, latency, and total production work. ## Source: compare-klippa.html [Skip to content](#main) Compare # Cogneris vs Klippa / Doxis. A practical comparison for teams evaluating Klippa, now presented with Doxis OCR API branding, against Cogneris for document extraction APIs, intelligent document processing, and document automation portals. [Talk to our team](contact.html)[See all comparisons](compare.html) ## The short version Klippa / Doxis is strongest around OCR APIs, SDKs, receipt capture, and fraud detection. Cogneris is built for teams that need document AI to become production workflow infrastructure: schema-based extraction, citations, validation, human review, webhook delivery, tenant controls, and audit trail. | Capability | Cogneris | Klippa | | Primary fit | Document AI platform for API-first workflows and portals | OCR APIs, SDKs, receipt capture, fraud detection, and Doxis-branded document capture | | Extraction output | Typed JSON, confidence, citations, validation, audit metadata | Structured extraction with vendor-specific strengths | | Workflow layer | Review queues, portal intake, reminders, webhooks, QA state, per-field confidence | Varies by product and deployment | | Engineering control | REST API, schemas, async jobs, validation rules, per-tenant controls | Strong where its product model matches your use case | | Best buyer | Engineering, product, operations, and compliance teams sharing one document workflow | Teams with a use case that maps tightly to Klippa's core product | ## When to choose which Choose Klippa / Doxis when receipt capture, mobile SDKs, and fraud checks are the center of the workflow. Choose Cogneris when you need a broader document AI platform with validation, review, and audit trail. ### Choose Cogneris when You need extraction, validation, review, audit trail, and portal workflow in one API-first platform. ### Choose Klippa / Doxis when Your use case maps directly to the Klippa/Doxis product surface and you already accept its operating model. ### Evaluation tip Test with your hardest 25 documents, not a demo set. Compare field accuracy, citation quality, latency, review effort, and total platform work. ## Questions to ask during evaluation Ask whether the platform returns source citations with every field, how schema changes are versioned, what happens to low-confidence fields, how webhook retries are signed, and whether audit logs include model version, prompt version, reviewer ID, and validation status. ## Related comparisons [Reducto](compare-reducto.html)[LandingAI ADE](compare-landingai-ade.html)[Nanonets](compare-nanonets.html)[AWS Textract](compare-aws-textract.html)[Google Document AI](compare-google-document-ai.html)[Azure DI](compare-azure-document-intelligence.html) ## Klippa alternative FAQ Is Cogneris a Klippa alternative? Yes. Cogneris is a Klippa alternative for teams that want document extraction, validation, review routing, webhooks, and audit evidence in one API-first workflow. When should I choose Cogneris instead of Klippa? Choose Cogneris when you need broader document AI workflow control, citations, review routing, and tenant-scoped audit trails around OCR and verification. Choose Klippa when its native product model, cloud ecosystem, or prebuilt workflow is the better fit for your operating constraints. How should I evaluate Klippa against Cogneris? Run the same representative documents through both systems and compare field accuracy, citation quality, schema stability, review effort, webhook behavior, latency, and total production work. ## Source: compare-landingai-ade.html [Skip to content](#main) Compare # Cogneris vs LandingAI ADE. A practical comparison for teams evaluating LandingAI ADE against Cogneris for document extraction APIs, intelligent document processing, and document automation portals. [Talk to our team](contact.html)[See all comparisons](compare.html) ## The short version LandingAI ADE is strongest around agentic document extraction, visual grounding, and visual extraction workflows. Cogneris is built for teams that need document AI to become production workflow infrastructure: schema-based extraction, citations, validation, human review, webhook delivery, tenant controls, and audit trail. | Capability | Cogneris | LandingAI ADE | | Primary fit | Document AI platform for API-first workflows and portals | agentic document extraction, visual grounding, and visual extraction workflows | | Extraction output | Typed JSON, confidence, citations, validation, audit metadata | Structured extraction with vendor-specific strengths | | Workflow layer | Review queues, portal intake, reminders, webhooks, QA state, per-field confidence | Varies by product and deployment | | Engineering control | REST API, schemas, async jobs, validation rules, per-tenant controls | Strong where its product model matches your use case | | Best buyer | Engineering, product, operations, and compliance teams sharing one document workflow | Teams with a use case that maps tightly to LandingAI ADE's core product | ## When to choose which Choose LandingAI ADE when you want a specialized agentic extraction tool and your workflow can be built around its platform. Choose Cogneris when document extraction needs portal intake, validation, multi-document workflow state, and tenant-scoped audit trail. ### Choose Cogneris when You need extraction, validation, review, audit trail, and portal workflow in one API-first platform. ### Choose LandingAI ADE when Your use case maps directly to LandingAI ADE's strongest product surface and you already accept its operating model. ### Evaluation tip Test with your hardest 25 documents, not a demo set. Compare field accuracy, citation quality, latency, review effort, and total platform work. ## Questions to ask during evaluation Ask whether the platform returns source citations with every field, how schema changes are versioned, what happens to low-confidence fields, how webhook retries are signed, and whether audit logs include model version, prompt version, reviewer ID, and validation status. ## Deployment and retention questions LandingAI ADE appears often in searches for agentic document extraction and private deployment. In a regulated evaluation, compare whether each vendor supports private VPC deployment, BYOK, region-pinned inference, zero data retention for source documents, and contract-bound audit metadata retention. ## Visual grounding vs workflow evidence Visual grounding is a useful evaluation term for ADE-style systems because it shows whether extracted values remain tied to where they appeared on the page. In a production workflow, also test whether those references survive reviewer edits, webhook delivery, schema version changes, and audit exports. ## Related comparisons [Private cloud](private-cloud-document-ai.html)[Benchmark guide](document-extraction-benchmark.html)[Agentic extraction](agentic-document-extraction.html)[Reducto](compare-reducto.html)[Nanonets](compare-nanonets.html)[AWS Textract](compare-aws-textract.html) ## Source: compare-llamaparse.html [Skip to content](#main) Compare # Cogneris vs LlamaParse. A practical comparison for teams evaluating LlamaParse against Cogneris for document extraction, parsing, OCR, and production workflow automation. [Talk to our team](contact.html)[See all comparisons](compare.html) ## The short version LlamaParse is strongest around document parsing and RAG preprocessing for LlamaIndex workflows. Cogneris is strongest when parsed or extracted data must become validated, reviewable, auditable workflow state. | Capability | Cogneris | LlamaParse | | Primary fit | Document AI workflows, portals, extraction APIs, validation, and audit trail | document parsing and RAG preprocessing for LlamaIndex workflows | | Output | Typed JSON with citations, confidence, validation, and workflow metadata | Document AI output optimized for its product category | | Best choice | When documents drive operational decisions and exceptions need review | Choose LlamaParse when your main need is ingestion and parsing for a LlamaIndex-centered RAG workflow. | ## RAG parsing vs workflow extraction LlamaParse-style systems are useful when the goal is layout-aware chunks, Markdown, tables, and embeddings for retrieval. Cogneris is a better fit when the same document also has to produce typed JSON, validation results, review state, webhooks, and audit evidence for an operational workflow. ## Evaluation advice Run the same sample packet through both systems and compare citation quality, schema stability, failed-field handling, webhook behavior, review effort, and total integration work. ## Related comparisons [Document AI for RAG](document-ai-for-rag.html)[PDF parser API](document-parsing-api.html)[Unstructured](compare-unstructured.html)[Reducto](compare-reducto.html)[LandingAI ADE](compare-landingai-ade.html)[Benchmark guide](document-extraction-benchmark.html) ## Source: compare-mindee.html [Skip to content](#main) Compare # Cogneris vs Mindee. A practical comparison for teams evaluating Mindee against Cogneris for document extraction APIs, intelligent document processing, and document automation portals. [Talk to our team](contact.html)[See all comparisons](compare.html) ## The short version Mindee is strongest around developer-focused document extraction APIs and custom models. Cogneris is built for teams that need document AI to become production workflow infrastructure: schema-based extraction, citations, validation, human review, webhook delivery, tenant controls, and audit trail. | Capability | Cogneris | Mindee | | Primary fit | Document AI platform for API-first workflows and portals | developer-focused document extraction APIs and custom models | | Extraction output | Typed JSON, confidence, citations, validation, audit metadata | Structured extraction with vendor-specific strengths | | Workflow layer | Review queues, portal intake, reminders, webhooks, QA state, per-field confidence | Varies by product and deployment | | Engineering control | REST API, schemas, async jobs, validation rules, per-tenant controls | Strong where its product model matches your use case | | Best buyer | Engineering, product, operations, and compliance teams sharing one document workflow | Teams with a use case that maps tightly to Mindee's core product | ## When to choose which Choose Mindee when you want a lightweight developer API for extracting known document types. Choose Cogneris when you need the API plus validation, audit trail, review state, and multi-step workflow automation. ### Choose Cogneris when You need extraction, validation, review, audit trail, and portal workflow in one API-first platform. ### Choose Mindee when Your use case maps directly to Mindee's strongest product surface and you already accept its operating model. ### Evaluation tip Test with your hardest 25 documents, not a demo set. Compare field accuracy, citation quality, latency, review effort, and total platform work. ## Questions to ask during evaluation Ask whether the platform returns source citations with every field, how schema changes are versioned, what happens to low-confidence fields, how webhook retries are signed, and whether audit logs include model version, prompt version, reviewer ID, and validation status. ## Related comparisons [Reducto](compare-reducto.html)[LandingAI ADE](compare-landingai-ade.html)[Nanonets](compare-nanonets.html)[AWS Textract](compare-aws-textract.html)[Google Document AI](compare-google-document-ai.html)[Azure DI](compare-azure-document-intelligence.html) ## Mindee alternative FAQ Is Cogneris a Mindee alternative? Yes. Cogneris is a Mindee alternative for teams that want document extraction, validation, review routing, webhooks, and audit evidence in one API-first workflow. When should I choose Cogneris instead of Mindee? Choose Cogneris when you need workflow state, validation rules, audit metadata, and human review in addition to developer-friendly OCR extraction APIs. Choose Mindee when its native product model, cloud ecosystem, or prebuilt workflow is the better fit for your operating constraints. How should I evaluate Mindee against Cogneris? Run the same representative documents through both systems and compare field accuracy, citation quality, schema stability, review effort, webhook behavior, latency, and total production work. ## Source: compare-mistral-ocr.html [Skip to content](#main) Compare # Cogneris vs Mistral OCR. A practical comparison for teams evaluating Mistral OCR against Cogneris for document extraction, parsing, OCR, and production workflow automation. [Talk to our team](contact.html)[See all comparisons](compare.html) ## The short version Mistral OCR is strongest around OCR and document understanding for LLM application builders. Cogneris is strongest when parsed or extracted data must become validated, reviewable, auditable workflow state. | Capability | Cogneris | Mistral OCR | | Primary fit | Document AI workflows, portals, extraction APIs, validation, and audit trail | OCR and document understanding for LLM application builders | | Output | Typed JSON with citations, confidence, validation, and workflow metadata | Document AI output optimized for its product category | | Best choice | When documents drive operational decisions and exceptions need review | Choose Mistral OCR when your main task is model-side OCR or document understanding inside an LLM stack. | ## OCR vs extraction workflow For a fair evaluation, separate raw OCR quality from operational extraction. Test layout-aware parsing, table preservation, structured JSON, source citations, validation rules, webhook delivery, and audit trail. OCR can read the page; a production document AI workflow has to decide what happens next. ## Evaluation advice Run the same sample packet through both systems and compare citation quality, schema stability, failed-field handling, webhook behavior, review effort, and total integration work. ## Related comparisons [Benchmark guide](document-extraction-benchmark.html)[OCR API](ocr-api.html)[Unstructured](compare-unstructured.html)[Reducto](compare-reducto.html)[LandingAI ADE](compare-landingai-ade.html)[AWS Textract](compare-aws-textract.html) ## Source: compare-nanonets.html [Skip to content](#main) Compare # Cogneris vs Nanonets. A practical comparison for teams evaluating Nanonets against Cogneris for document extraction APIs, intelligent document processing, and document automation portals. [Talk to our team](contact.html)[See all comparisons](compare.html) ## The short version Nanonets is strongest around OCR APIs, agentic data extraction, and workflow automation. Cogneris is built for teams that need document AI to become production workflow infrastructure: schema-based extraction, citations, validation, human review, webhook delivery, tenant controls, and audit trail. | Capability | Cogneris | Nanonets | | Primary fit | Document AI platform for API-first workflows and portals | OCR APIs, agentic data extraction, and workflow automation | | Extraction output | Typed JSON, confidence, citations, validation, audit metadata | Structured extraction with vendor-specific strengths | | Workflow layer | Review queues, portal intake, reminders, webhooks, QA state, per-field confidence | Varies by product and deployment | | Engineering control | REST API, schemas, async jobs, validation rules, per-tenant controls | Strong where its product model matches your use case | | Best buyer | Engineering, product, operations, and compliance teams sharing one document workflow | Teams with a use case that maps tightly to Nanonets's core product | ## When to choose which Choose Nanonets when you want a broad OCR automation product with many prebuilt document models. Choose Cogneris when the buying center is engineering-led and needs schema control, citations, review routing, and pay-per-page pricing. ### Choose Cogneris when You need extraction, validation, review, audit trail, and portal workflow in one API-first platform. ### Choose Nanonets when Your use case maps directly to Nanonets's strongest product surface and you already accept its operating model. ### Evaluation tip Test with your hardest 25 documents, not a demo set. Compare field accuracy, citation quality, latency, review effort, and total platform work. ## Questions to ask during evaluation Ask whether the platform returns source citations with every field, how schema changes are versioned, what happens to low-confidence fields, how webhook retries are signed, and whether audit logs include model version, prompt version, reviewer ID, and validation status. ## Related comparisons [Reducto](compare-reducto.html)[LandingAI ADE](compare-landingai-ade.html)[Nanonets](compare-nanonets.html)[AWS Textract](compare-aws-textract.html)[Google Document AI](compare-google-document-ai.html)[Azure DI](compare-azure-document-intelligence.html) ## Nanonets alternative FAQ Is Cogneris a Nanonets alternative? Yes. Cogneris is a Nanonets alternative for teams that want document extraction, validation, review routing, webhooks, and audit evidence in one API-first workflow. When should I choose Cogneris instead of Nanonets? Choose Cogneris when you need extraction, validation, human review, and audit trails wired into your own product or document portal. Choose Nanonets when its native product model, cloud ecosystem, or prebuilt workflow is the better fit for your operating constraints. How should I evaluate Nanonets against Cogneris? Run the same representative documents through both systems and compare field accuracy, citation quality, schema stability, review effort, webhook behavior, latency, and total production work. ## Source: compare-pdf-parsing-apis.html [Skip to content](#main) Compare parsers # PDF parsing APIs compared. A practical comparison of RAG-ready parsing, extraction, citations, table fidelity, schema output, and workflow readiness. [Open benchmark guide](document-extraction-benchmark.html)[RAG architecture](document-ai-for-rag.html) ## Shortlist by job, not category name PDF parsing APIs and document extraction APIs now overlap. LlamaParse, Reducto, Unstructured, Mistral OCR, LandingAI ADE, Docling, Tensorlake, Firecrawl, and DocuPipe are often evaluated for RAG preprocessing and LLM-ready document data. Cogneris belongs in the same evaluation when parsed content also has to become validated JSON, review state, webhooks, and audit evidence. | Vendor | Strongest fit | Watch during evaluation | | LlamaParse | Developer-friendly parsing for LlamaIndex/RAG workflows | How much production workflow logic you still need to build | | Reducto | Schema extraction, citations, and RAG-oriented parsing | Whether review, validation, and portal workflows are first-class enough | | Unstructured | Partitioning and preprocessing heterogeneous files for RAG | Table fidelity, field validation, and downstream operational state | | Mistral OCR | OCR-to-Markdown and multimodal document understanding | Workflow controls, audit metadata, and low-confidence routing | | LandingAI ADE | Agentic extraction, grounding, and visual document workflows | API contract, pricing shape, and integration fit | | Docling | Open-source parsing and document conversion in developer pipelines | Operational review, hosted reliability, and audit evidence | | Tensorlake | Benchmark-driven document parsing and extraction for AI systems | Workflow controls, schema governance, and production exception handling | | Firecrawl | Web and document content extraction feeding AI apps | Business-document validation, citations, review state, and portal workflows | | DocuPipe | Document extraction pipeline for PDFs, spreadsheets, and RAG-ready content | How extraction, review, and workflow ownership map to your operating model | | Cogneris | RAG-ready parsing plus schema extraction, validation, review, and audit trail | Best fit when document AI must feed both search and operations | ## Evaluation checklist Test each API on scanned PDFs, native PDFs, PPTX and PowerPoint exports, Word documents, tables, multi-page packets, handwritten fields, checkboxes, signatures, and documents with repeated labels. Score layout-aware chunks, Markdown quality, semantic chunking, table preservation, visual grounding, source citations, JSON schema compliance, latency, retry behavior, and review workload. ## RAG-ready is not enough RAG-ready output helps answer questions. Operations-ready output also validates totals, normalizes dates, flags low-confidence fields, deduplicates packets, and explains every value with source evidence. If a wrong value can trigger a payment, claim, approval, or compliance failure, include workflow metrics in the bakeoff. ## Related pages [Document AI for RAG](document-ai-for-rag.html)[Benchmark guide](document-extraction-benchmark.html)[Compare Reducto](compare-reducto.html)[Compare LlamaParse](compare-llamaparse.html)[Compare Unstructured](compare-unstructured.html)[Compare LandingAI ADE](compare-landingai-ade.html)[Compare Mistral OCR](compare-mistral-ocr.html) ## Source: compare-reducto.html [Skip to content](#main) Compare # Cogneris vs Reducto. A practical comparison for teams evaluating Reducto against Cogneris for document extraction APIs, intelligent document processing, and document automation portals. [Talk to our team](contact.html)[See all comparisons](compare.html) ## The short version Reducto is strongest around schema extraction, citations, and developer-first parsing. Cogneris is built for teams that need document AI to become production workflow infrastructure: schema-based extraction, citations, validation, human review, webhook delivery, tenant controls, and audit trail. | Capability | Cogneris | Reducto | | Primary fit | Document AI platform for API-first workflows and portals | schema extraction, citations, and developer-first parsing | | Extraction output | Typed JSON, confidence, citations, validation, audit metadata | Structured extraction with vendor-specific strengths | | Workflow layer | Review queues, portal intake, reminders, webhooks, QA state, per-field confidence | Varies by product and deployment | | Engineering control | REST API, schemas, async jobs, validation rules, per-tenant controls | Strong where its product model matches your use case | | Best buyer | Engineering, product, operations, and compliance teams sharing one document workflow | Teams with a use case that maps tightly to Reducto's core product | ## When to choose which Choose Reducto when your main job is document-to-LLM preprocessing with citation-rich structured extraction. Choose Cogneris when extraction has to continue into validation, review queues, portal intake, and audit-ready workflow automation. ### Choose Cogneris when You need extraction, validation, review, audit trail, and portal workflow in one API-first platform. ### Choose Reducto when Your use case maps directly to Reducto's strongest product surface and you already accept its operating model. ### Evaluation tip Test with your hardest 25 documents, not a demo set. Compare field accuracy, citation quality, latency, review effort, and total platform work. ## Questions to ask during evaluation Ask whether the platform returns source citations with every field, how schema changes are versioned, what happens to low-confidence fields, how webhook retries are signed, and whether audit logs include model version, prompt version, reviewer ID, and validation status. ## Benchmark Reducto against Cogneris Run the same document AI bakeoff for both vendors: PDF parser output, layout-aware chunks, schema extraction, citation quality, table fidelity, auto-approval rate, and reviewer minutes per case. If the workflow touches regulated data, include private VPC, zero data retention, and audit-retention requirements in the scorecard. ## Related comparisons [Benchmark guide](document-extraction-benchmark.html)[Document AI for RAG](document-ai-for-rag.html)[Private cloud](private-cloud-document-ai.html)[LandingAI ADE](compare-landingai-ade.html)[Nanonets](compare-nanonets.html)[Unstructured](compare-unstructured.html) ## Source: compare-tungsten-kofax.html [Skip to content](#main) Compare # Cogneris vs Tungsten/Kofax. A practical comparison for teams evaluating Tungsten Automation, formerly Kofax, against Cogneris for intelligent document processing, orchestration, and document AI APIs. [Talk to our team](contact.html)[See all comparisons](compare.html) ## The short version Tungsten Automation, formerly Kofax, is strongest around mature intelligent document processing, workflow orchestration, governance, and enterprise automation estates. Cogneris is stronger when the evaluation is API-first, product-embedded, and focused on getting document extraction, validation, review state, webhooks, and audit trail live quickly. | Capability | Cogneris | Tungsten/Kofax | | Primary fit | API-first document AI platform for workflows and portals | Enterprise IDP and workflow automation suite | | Document coverage | Structured JSON with citations, validation, and reviewer evidence | Broad capture, classification, extraction, and process automation tooling | | Workflow layer | Review queues, portal intake, webhooks, audit trail, per-tenant controls | End-to-end orchestration and governance across larger automation programs | | Implementation path | REST API, page-based pricing, and product-team ownership | Best with enterprise implementation, operations ownership, and suite adoption | | Best buyer | Teams embedding document AI into a product, portal, or custom workflow | Large operations teams modernizing established capture and process estates | ## When to choose which Choose Tungsten/Kofax when you need a broad enterprise automation suite, have existing Kofax estate knowledge, or want IDP tightly coupled with large-scale process orchestration. Choose Cogneris when the work is narrower: document extraction APIs, reviewable JSON, portal workflows, and audit evidence that engineering can ship directly. ### Choose Cogneris when You need a focused document AI platform with schema control, citations, validation, webhooks, tenant permissions, and audit metadata. ### Choose Tungsten/Kofax when Your operating model already depends on a broad IDP and workflow automation suite with governance and orchestration controls. ### Evaluation tip Score total time-to-production, licensing shape, reviewer effort, data retention, integration ownership, and the cost of suite-specific customization. ## Questions to ask during evaluation Ask whether the deployment requires suite adoption, how reviewer changes are audited, how extraction schemas move between environments, how webhook retries are handled, and whether product teams can own integrations without a dedicated automation center of excellence. ## Related comparisons [UiPath IXP](compare-uipath-ixp.html)[ABBYY Vantage](compare-abbyy-vantage.html)[Hyperscience](compare-hyperscience.html)[Rossum](compare-rossum.html)[Document automation portal](document-automation-portal.html)[Private cloud](private-cloud-document-ai.html) ## Source: compare-uipath-ixp.html [Skip to content](#main) Compare # Cogneris vs UiPath IXP. A practical comparison for teams evaluating UiPath Intelligent Xtraction and Processing against Cogneris for document AI APIs, agentic automation, validation, and audit trail. [Talk to our team](contact.html)[See all comparisons](compare.html) ## The short version UiPath IXP is the current UiPath direction for Intelligent Xtraction and Processing: multimodal extraction from documents and communications for automation and analytics. Cogneris is a narrower API-first document AI platform for teams that need structured JSON, citations, validation, review state, webhooks, tenant controls, and audit trail without adopting a broad RPA suite. | Capability | Cogneris | UiPath IXP | | Primary fit | Document AI APIs, portals, validation, review, and audit trail | Document and communications extraction inside UiPath automation programs | | Output model | Typed JSON, confidence, source citations, validation status, audit metadata | Structured outputs for automation, analytics, and agentic workflows | | Workflow layer | Portal intake, review queues, webhooks, reminders, per-tenant controls | Works best when paired with UiPath automation, validation apps, and governance | | Implementation path | REST API and pages-based pricing for product and engineering teams | Best for organizations already standardizing on UiPath's automation platform | | Best buyer | Engineering, product, operations, and compliance teams sharing one document workflow | Automation centers of excellence that want document understanding inside UiPath | ## When to choose which Choose UiPath IXP when document understanding is part of a larger UiPath automation estate. Choose Cogneris when the immediate job is to embed document extraction, validation, review, and audit evidence in your own product or document portal. ### Choose Cogneris when You need an API-first document AI layer with citations, schema control, webhooks, tenant permissions, and audit metadata. ### Choose UiPath IXP when You already run UiPath at scale and want document and communications extraction feeding bots, agents, and process orchestration. ### Evaluation tip Compare total implementation work: extraction accuracy, validation app effort, reviewer handoff, webhook behavior, audit lineage, and licensing shape. ## Questions to ask during evaluation Ask whether low-confidence fields keep page-level source evidence, how schema versions are promoted, whether reviewer changes are exported with audit metadata, and whether a non-UiPath product can consume the same structured output without workflow lock-in. ## Related comparisons [Tungsten/Kofax](compare-tungsten-kofax.html)[ABBYY Vantage](compare-abbyy-vantage.html)[Hyperscience](compare-hyperscience.html)[Rossum](compare-rossum.html)[Document automation portal](document-automation-portal.html)[Agentic extraction](agentic-document-extraction.html) ## Source: compare-unstructured.html [Skip to content](#main) Compare # Cogneris vs Unstructured. A practical comparison for teams evaluating Unstructured against Cogneris for document extraction APIs, intelligent document processing, and document automation portals. [Talk to our team](contact.html)[See all comparisons](compare.html) ## The short version Unstructured is strongest around document partitioning and preprocessing for RAG pipelines. Cogneris is built for teams that need document AI to become production workflow infrastructure: schema-based extraction, citations, validation, human review, webhook delivery, tenant controls, and audit trail. | Capability | Cogneris | Unstructured | | Primary fit | Document AI platform for API-first workflows and portals | document partitioning and preprocessing for RAG pipelines | | Extraction output | Typed JSON, confidence, citations, validation, audit metadata | Structured extraction with vendor-specific strengths | | Workflow layer | Review queues, portal intake, reminders, webhooks, QA state, per-field confidence | Varies by product and deployment | | Engineering control | REST API, schemas, async jobs, validation rules, per-tenant controls | Strong where its product model matches your use case | | Best buyer | Engineering, product, operations, and compliance teams sharing one document workflow | Teams with a use case that maps tightly to Unstructured's core product | ## When to choose which Choose Unstructured when the job is preparing mixed document formats for retrieval and LLM applications. Choose Cogneris when parsed content must become validated business data with workflow state, permissions, and review evidence. ### Choose Cogneris when You need extraction, validation, review, audit trail, and portal workflow in one API-first platform. ### Choose Unstructured when Your use case maps directly to Unstructured's strongest product surface and you already accept its operating model. ### Evaluation tip Test with your hardest 25 documents, not a demo set. Compare field accuracy, citation quality, latency, review effort, and total platform work. ## Questions to ask during evaluation Ask whether the platform returns source citations with every field, how schema changes are versioned, what happens to low-confidence fields, how webhook retries are signed, and whether audit logs include model version, prompt version, reviewer ID, and validation status. ## Related comparisons [Reducto](compare-reducto.html)[LandingAI ADE](compare-landingai-ade.html)[Nanonets](compare-nanonets.html)[AWS Textract](compare-aws-textract.html)[Google Document AI](compare-google-document-ai.html)[Azure DI](compare-azure-document-intelligence.html) ## Unstructured alternative FAQ Is Cogneris a Unstructured alternative? Yes. Cogneris is a Unstructured alternative for teams that want document extraction, validation, review routing, webhooks, and audit evidence in one API-first workflow. When should I choose Cogneris instead of Unstructured? Choose Cogneris when you need operational extraction, typed schemas, validation, review queues, and audit evidence alongside RAG preprocessing. Choose Unstructured when its native product model, cloud ecosystem, or prebuilt workflow is the better fit for your operating constraints. How should I evaluate Unstructured against Cogneris? Run the same representative documents through both systems and compare field accuracy, citation quality, schema stability, review effort, webhook behavior, latency, and total production work. ## Source: compare-veryfi.html [Skip to content](#main) Compare # Cogneris vs Veryfi. A practical comparison for teams evaluating Veryfi against Cogneris for document extraction APIs, intelligent document processing, and document automation portals. [Talk to our team](contact.html)[See all comparisons](compare.html) ## The short version Veryfi is strongest around receipt, invoice, bank statement, and financial document APIs. Cogneris is built for teams that need document AI to become production workflow infrastructure: schema-based extraction, citations, validation, human review, webhook delivery, tenant controls, and audit trail. ## What if a model invents a number? That risk is real: model output is not inherently deterministic. Cogneris reports field-level confidence and treats every generated field as a candidate that must carry evidence and satisfy the controls configured for the workflow. - 1Source citation — connect the extracted value to its page and document region. - 2Deterministic validation — apply schema, type, arithmetic, and cross-document rules after extraction. - 3Multi-model consensus — require configured models to agree on critical fields. - 4Human review — route low-confidence values and rule conflicts to a reviewer. - 5Audit trail — retain model version, prompt and response hashes, validation results, and reviewer action. | Capability | Cogneris | Veryfi | | Primary fit | Document AI platform for API-first workflows and portals | receipt, invoice, bank statement, and financial document APIs | | Extraction output | Typed JSON, confidence, citations, validation, audit metadata | Structured extraction with vendor-specific strengths | | Workflow layer | Review queues, portal intake, reminders, webhooks, QA state, per-field confidence | Varies by product and deployment | | Engineering control | REST API, schemas, async jobs, validation rules, per-tenant controls | Strong where its product model matches your use case | | Best buyer | Engineering, product, operations, and compliance teams sharing one document workflow | Teams with a use case that maps tightly to Veryfi's core product | ## How the two pricing models differ This is the sharpest practical difference between the two products, and it is not a matter of who is cheaper overall. Veryfi bills per document: one API transaction covers a single document of up to 15 pages, so a 10-page bank statement costs the same as a 1-page one. Cogneris bills per page. Which model is cheaper depends entirely on how long your documents are, and there is a crossover point you can calculate. Below, [our published per-page overage rates](pricing.html#how-credits-work) carried out to whole documents, against [Veryfi's published Starter per-document rates](https://www.veryfi.com/pricing/), both checked 28 August 2026. | Document | Pages | Cogneris Professional | Veryfi Starter | Cheaper | | Invoice | 1 | $0.06 | $0.16 | Cogneris | | W-9 | 1 | $0.06 | $0.16 | Cogneris | | Bank check | 1 | $0.06 | $0.25 | Cogneris | | Receipt | 1 | $0.06 | $0.08 | Cogneris | | Invoice with line-item annex | 3 | $0.18 | $0.16 | Veryfi | | Bank statement | 6 | $0.36 | $0.25 | Veryfi | | Bank statement | 12 | $0.72 | $0.25 | Veryfi | | Contract | 40 | $4.80 | Not published | — | ### Where Veryfi is cheaper On multi-page financial documents that fit inside 15 pages, Veryfi wins on price, and the gap widens with every page. A 12-page bank statement is $0.72 on our Professional rate against $0.25 on theirs — nearly three times more. Our Enterprise standard rate starts at $0.03 per page, which takes the 6-page statement to $0.18 and back under their document rate, but the 12-page case still lands at $0.36. If your volume is mostly long statements, Veryfi's model is cheaper and you should weigh that against what the rest of this page describes. ### Where Cogneris is cheaper On single-page documents we are consistently cheaper, and on checks and bank statements at one page the difference is roughly four times. The crossover is arithmetic: against a $0.25 document rate our $0.06 page rate is cheaper up to four pages; against a $0.16 invoice, up to two. Separately, Veryfi's Starter plan carries a $500 monthly minimum commitment, while Cogneris Starter is $99 with 1,000 pages included — so at low volume the per-document rate is not what decides your bill. A 40-page contract exceeds Veryfi's 15-page transaction and needs custom configuration, so there is no public price to compare. We have left that cell empty rather than estimate one on their behalf. ## When to choose which Choose Veryfi when you need a narrow, high-speed financial document API with fraud signals. Choose Cogneris when you need broader document coverage, cross-document validation, review queues, and portal automation. ### Choose Cogneris when You need extraction, validation, review, audit trail, and portal workflow in one API-first platform. ### Choose Veryfi when Your use case maps directly to Veryfi's strongest product surface and you already accept its operating model. ### Evaluation tip Test with your hardest 25 documents, not a demo set. Compare field accuracy, citation quality, latency, review effort, and total platform work. ## Questions to ask during evaluation Ask whether the platform returns source citations with every field, how schema changes are versioned, what happens to low-confidence fields, how webhook retries are signed, and whether audit logs include model version, prompt version, reviewer ID, and validation status. ## Related comparisons [Reducto](compare-reducto.html)[LandingAI ADE](compare-landingai-ade.html)[Nanonets](compare-nanonets.html)[AWS Textract](compare-aws-textract.html)[Google Document AI](compare-google-document-ai.html)[Azure DI](compare-azure-document-intelligence.html) ## Veryfi alternative FAQ Is Cogneris a Veryfi alternative? Yes. Cogneris is a Veryfi alternative for teams that want document extraction, validation, review routing, webhooks, and audit evidence in one API-first workflow. When should I choose Cogneris instead of Veryfi? Choose Cogneris when you need multi-document workflow automation, portal intake, and audit evidence beyond financial-document API extraction. Choose Veryfi when its native product model, cloud ecosystem, or prebuilt workflow is the better fit for your operating constraints. How should I evaluate Veryfi against Cogneris? Run the same representative documents through both systems and compare field accuracy, citation quality, schema stability, review effort, webhook behavior, latency, and total production work. ## Source: compare.html [Skip to content](#main) Compare # Cogneris vs the alternatives. Side-by-side comparisons against the document AI platforms buyers ask us about most often. Each comparison is honest about strengths and weaknesses on both sides — including ours. Pick the one closest to your incumbent or evaluation shortlist. New to the category? Start with [the IDP buyer's guide](what-is-idp.html), or read [the 2026 State of Document AI report](state-of-document-ai-2026.html) for market context before you shortlist. If analyst coverage is part of your evaluation, compare each vendor's Gartner Magic Quadrant, Forrester Wave, PEAK Matrix, customer references, and security evidence directly. Cogneris does not claim those placements here; we use those terms so procurement teams can map the same proof checklist across vendors. [ Cogneris vs ### UiPath IXP Intelligent Xtraction and Processing for UiPath automation estates. Compare against Cogneris for API-first document AI, review state, webhooks, and audit evidence. Read the comparison ](compare-uipath-ixp.html) [ Cogneris vs ### Tungsten/Kofax Enterprise IDP and workflow automation suite. Compare against Cogneris for focused document AI APIs, portal workflows, implementation speed, and ownership model. Read the comparison ](compare-tungsten-kofax.html) [ Cogneris vs ### Rossum Mature, invoice-first, enterprise-priced. The default for Fortune-500 AP teams. We compare on document coverage, agentic architecture, pricing model, and on-prem options. Read the comparison ](compare-rossum.html) [ Cogneris vs ### Docsumo Strong mid-market platform with pre-trained finance-doc models. We compare on multi-document coverage, LLM control, workflow tooling, and pricing tiers. Read the comparison ](compare-docsumo.html) [ Cogneris vs ### Hyperscience Mature IDP with deep human-in-the-loop tooling and FedRAMP-grade deployments. We compare on architecture, time-to-production, LLM strategy, and TCO. Read the comparison ](compare-hyperscience.html) [ Cogneris vs ### Google Document AI Powerful extraction primitive inside GCP. We compare platform-vs-primitive tradeoffs, multi-LLM flexibility, and the workflow build-out you'd otherwise do yourself. Read the comparison ](compare-google-document-ai.html) [ Cogneris vs ### Azure Document Intelligence Microsoft's Foundry Tools document service, formerly Form Recognizer and now part of Azure Content Understanding. We compare platform completeness, LLM strategy, and Microsoft lock-in. Read the comparison ](compare-azure-document-intelligence.html) [ Cogneris vs ### AWS Textract AWS's OCR and document-analysis API. We compare Textract's primitives against Cogneris's complete platform for review queues, validation, webhooks, and audit trail. Read the comparison ](compare-aws-textract.html) [ Cogneris vs ### Reducto Schema extraction, citations, and developer-first parsing. Compare Reducto against Cogneris for workflow depth, validation, review state, and audit evidence. Read the comparison ](compare-reducto.html) [ Cogneris vs ### LandingAI ADE Agentic document extraction for visually complex documents. Compare against Cogneris for portal workflows, tenant controls, validation, and audit trail. Read the comparison ](compare-landingai-ade.html) [ Cogneris vs ### Nanonets OCR APIs and agentic data extraction. Compare against Cogneris for API control, review routing, document coverage, and pricing model. Read the comparison ](compare-nanonets.html) [ Cogneris vs ### Veryfi Financial document APIs for receipts, invoices, and bank statements. Compare against Cogneris for broader workflow automation and evidence handling. Read the comparison ](compare-veryfi.html) [ Cogneris vs ### Mindee Developer-first extraction APIs and custom documents. Compare against Cogneris for workflow state, validation, and audit-ready operations. Read the comparison ](compare-mindee.html) [ Cogneris vs ### ABBYY Vantage Low-code enterprise IDP suite. Compare against Cogneris for API-first deployment, LLM control, implementation speed, and engineering ownership. Read the comparison ](compare-abbyy-vantage.html) [ Cogneris vs ### Klippa OCR APIs, SDKs, receipt capture, and fraud detection. Compare against Cogneris for broader document AI workflow automation. Read the comparison ](compare-klippa.html) [ Cogneris vs ### Mistral OCR OCR and document understanding for LLM app builders. Compare against Cogneris for structured extraction, validation, workflow state, and audit trail. Read the comparison ](compare-mistral-ocr.html) [ Cogneris vs ### LlamaParse Parsing and RAG preprocessing for LlamaIndex workflows. Compare against Cogneris for operational extraction, validation, review, and audit evidence. Read the comparison ](compare-llamaparse.html) [ Cogneris vs ### Unstructured Document partitioning and RAG preprocessing. Compare against Cogneris for structured extraction, validation, review, and business workflows. Read the comparison ](compare-unstructured.html) [ Guide ### PDF parsing APIs Compare LlamaParse, Reducto, Unstructured, Mistral OCR, LandingAI, Docling, Tensorlake, Firecrawl, DocuPipe, and Cogneris for RAG-ready parsing and workflow readiness. Read the guide ](compare-pdf-parsing-apis.html) Our approach ## Honest comparisons. Including our weaknesses. Most vendor comparison pages are thinly-disguised hit pieces. Ours aren't. Each comparison spends as much time on when to choose the other vendor as on when to choose us — because most procurement decisions are nuanced, and pretending otherwise burns credibility with the buyers we want to work with. For a side-by-side technical evaluation against your current vendor with your real documents, start with the [document extraction benchmark guide](document-extraction-benchmark.html) or [talk to our team](contact.html). We run a 7-day evaluation and produce an accuracy and latency report you can show your CTO. No commitment, no sales pressure. ## Source: document-ai-pricing-comparison.html [Skip to content](#main) Pricing # Document AI pricing without the fog. The headline per-page price is rarely the real cost. Compare OCR primitives, IDP suites, agentic extraction APIs, review labor, implementation work, and audit requirements together. [Talk to our team](contact.html)[See Cogneris pricing](pricing.html) ## Common pricing models | Model | Where it appears | Hidden cost | | Per-page OCR | AWS Textract, Google, Azure, Klippa-style APIs | You still build extraction, validation, review, and audit workflow | | Per-document extraction | Veryfi, Mindee, modern extraction APIs | Complex document types and custom schemas can change pricing | | Annual IDP suite | Rossum, ABBYY, Hyperscience, Docsumo | Implementation, seats, minimums, and change requests | | Workflow platform | Cogneris and similar production platforms | Review volume and validation complexity matter more than OCR alone | ## What to include in TCO Model costs, page volume, retries, review labor, exception handling, monitoring, compliance evidence, integration work, and vendor support all belong in the calculation. A cheap OCR call can become expensive if the engineering team has to build everything around it. ## How Cogneris is priced Cogneris uses transparent pay-per-page pricing with platform features attached: extraction, validation, review routing, webhooks, and audit trail. For high-volume or regulated workflows, compare cost per resolved case instead of cost per OCR call. Per-page and per-document billing cross over as documents get longer. Cogneris charges a fixed base per operation plus a small per-page amount, so a long multi-page document costs less per page than a short one. The [credit table](pricing.html#how-credits-work) shows that arithmetic on a 1-page and a 5-page document. ## Source: document-types.html [Skip to content](#main) Document types # Document type extraction by workflow. One place to map the documents your users upload to the Cogneris extraction page, schema, validation pattern, and review workflow that fits it. [Read extraction docs](https://docs.cogneris.ai/docs/api/document-extraction/)[Open API overview](document-extraction-api.html) ## Capacity coverage, without marketing estimates Cogneris document classes and extraction schemas are configured per tenant. That makes a single global count misleading. The figures below are limited to capabilities that have a current, traceable product source; variable axes are named explicitly instead of left blank. ### Document types Tenant-configured There is no fixed product-wide count. Teams define the document classes their workflows require and validate them against representative files. ### Extraction fields Schema-defined There is no fixed field total. Each tenant template defines its fields, types, required values, validation rules, and review thresholds. ### Interface languages 15 localized languages The application interface ships in 15 languages. This is not an extraction-language claim: extraction coverage depends on the configured model, schema, and the customer's validated document set. ### Currencies Schema-defined There is no fixed product-wide currency allowlist. Currency codes and monetary values are extracted and validated according to the tenant schema. ### File and page limits 10 MB per file The public synchronous Document AI API accepts files up to 10 MB. There is no single page ceiling across every endpoint; route-specific limits apply, and document splitting has a separate contract. Evidence checked 26 August 2026: the [interface locale registry](https://github.com/cogneris-ai/cogneris-platform-fe/blob/db3f0d7aeb6841157ab9feed85ce18ad3656994c/src/src/app/services/language.service.ts), the [API upload validator](https://github.com/cogneris-ai/cogneris-api-be/blob/f0bf5b0967553aa67a95f1aba956f62e22fcfbd1/src/Web/Validators/FileValidator.cs), and the machine-readable [capacity evidence snapshot](capacity-evidence.json). ## High-intent document extraction pages ### Invoices Vendor, invoice number, line items, tax, totals, PO references, GL-code hints, and match status. [Open invoice extraction](extract-invoices.html) ### Receipts Merchant, date, itemized lines, subtotal, tax, tip, payment method, currency, and duplicate signals. [Open receipt extraction](extract-receipts.html) ### Bank statements Account metadata, balances, transactions, statement period, fees, deposits, and reconciliation fields. [Open bank statement extraction](extract-bank-statements.html) ### Checks Payer, payee, amount, date, memo, MICR context, remittance data, and payment-review evidence. [Open check OCR API](check-ocr-api.html) ### ID cards and passports Name, date of birth, document number, issue/expiry dates, address, MRZ fields, and source evidence. [Open ID card OCR API](id-card-ocr-api.html) ### Tax forms W-2, 1099, 1040 source documents, K-1, payer/payee fields, wages, withholding, and review flags. [Open tax form extraction](extract-tax-returns.html) ### Mortgage packets Loan packets, paystubs, statements, IDs, tax forms, disclosures, and underwriting evidence. [Open mortgage extraction](mortgage-document-extraction-api.html) ### ACORD forms Policy, applicant, vehicle, property, coverage, premium, loss data, and insurance workflow checks. [Open ACORD extraction](extract-acord-forms.html) ### Bills of lading Carrier, consignee, shipment IDs, weights, dates, charges, line items, and logistics handoff data. [Open BOL OCR API](extract-bill-of-lading.html) ### Proof of delivery POD numbers, delivery dates, addresses, carriers, signatures, exception notes, and delivery evidence. [Open POD OCR API](proof-of-delivery-ocr-api.html) ### EOBs Patient, provider, payer, claim, service lines, allowed amounts, adjustments, and denial codes. [Open EOB OCR API](eob-ocr-api.html) ### Contracts Parties, term, dates, governing law, clauses, obligations, signatures, and citation-backed answers. [Open contract extraction](extract-contracts.html) ### Purchase orders PO numbers, vendors, line items, totals, delivery terms, approvals, and three-way-match evidence. [Open PO OCR API](purchase-order-ocr-api.html) ### Payroll and W-2 Payslips, W-2s, paystubs, 1099 forms, employer details, income fields, and lending-ready audit trail. [Open payroll extraction](extract-payroll.html) ### Proof of address Names, addresses, issuers, dates, freshness signals, residency evidence, and validation state. [Open proof-of-address API](proof-of-address-extraction-api.html) ### KYC document packets IDs, passports, proof of address, business records, beneficial owners, and cross-document validation. [Open KYC document extraction](extract-kyc-documents.html) ### Health insurance cards Member and group IDs, payer, plan, coverage dates, copay fields, and eligibility workflow inputs. [Open health insurance card OCR](health-insurance-card-ocr-api.html) ### Insurance claims Claim forms, policies, loss evidence, estimates, invoices, parties, dates, and review-ready fields. [Open insurance claim extraction](extract-insurance-claims.html) ### Claims processing Mixed claim packets classified and extracted for intake, adjudication, exception routing, and audit trails. [Open claims processing API](claims-document-processing-api.html) ### Financial spreading Statements, tax returns, schedules, normalized periods, ratios, and source-backed credit analysis. [Open financial spreading](financial-spreading-document-ai.html) ## How to choose the right extraction route | Workflow need | Use this pattern | Start here | | Known document type | Use a shipped template and tenant-specific validation rules. | [Extraction docs](https://docs.cogneris.ai/docs/api/document-extraction/) | | Unknown upload packet | Classify, split, extract, and return one normalized payload per document. | [Classification docs](https://docs.cogneris.ai/docs/api/document-classifier/) | | Custom schema | Pass a schema with required fields, field types, and validation expectations. | [Document extraction API](document-extraction-api.html) | | Reviewable output | Require citations, confidence scores, bounding boxes, reviewer changes, and audit metadata. | [Source citations](document-ai-source-citations.html) | ## Why a document-type hub matters Competitor pages often rank by document type: invoice OCR API, receipt OCR API, bank statement extraction, ID card OCR, mortgage document extraction, purchase order OCR, proof of delivery OCR, and EOB OCR. This hub gives buyers and search engines a clean map from each document type to the Cogneris page that explains fields, output, validation, review, and integration depth. ## Source: blog-loan-estimate-closing-disclosure-tolerance.html [Skip to content](#main) EngineeringOctober 27, 2026 · 11 min read # The tolerance follows the bucket. A Loan Estimate and a Closing Disclosure are two snapshots of the same transaction. To explain a changed charge, trace the service, provider and shopping right to its tolerance bucket before comparing the numbers. By Cogneris Marketing TeamPublished 2026-10-2711 min read The useful comparison is not “did this line go up?” It is “what rule governs this charge, what evidence puts it there, and what changed between the estimate and closing?” Under the TILA-RESPA Integrated Disclosure (TRID) rule, the answer depends on the charge’s category and the facts behind it. ## Start with the same transaction and the right versions The Loan Estimate (LE) is an early good-faith estimate. The Closing Disclosure (CD) gives the final terms and costs. Build a comparison from the most recent LE that applies and the CD delivered for consummation; retain each document’s issue date, version and page. A corrected form should not silently overwrite the version that came before it. Before comparing amounts, align the underlying service. Normalize labels only to propose a match, then preserve the printed descriptions, provider, amount, estimate date and source location on both sides. “Title services” on one form may cover several services on the other; a similar label alone does not establish that two rows are the same charge. ## Classify each charge before testing its change For each LE charge, capture the service, the disclosed provider or provider category, whether the consumer was permitted to shop, the amount and any written provider-list context. Link it to a CD charge only when the service and transaction facts support that match. Then classify it under 12 CFR 1026.19(e)(3): - Zero tolerance: charges paid to the creditor, mortgage broker or their affiliates; charges for a service the consumer was not permitted to shop for; and transfer taxes generally cannot increase above the applicable estimate, subject to the rule’s limited provisions. - 10% aggregate tolerance: eligible charges for services the consumer was permitted to shop for, when the consumer selects a provider from the written list, plus recording fees. The limit applies to the sum of applicable charges, not separately to every row. - Permitted to vary: specified charges such as prepaid interest, property-insurance premiums, initial escrow deposits, certain services when the consumer selects a provider outside the written list, and services not required by the creditor. Keep the exact regulatory conditions with the classification. These labels are a rule-based grouping, not an amount-based guess. The same service name can land differently when the provider relationship or shopping facts differ. If those facts are absent, preserve the uncertainty for review rather than assigning a bucket from the dollar movement. ## The 10% test is a pooled calculation Suppose eligible recording and third-party shopping charges total $1,000 on the applicable LE. On the CD, one eligible fee rises by $40 and another falls by $20. The comparison is $1,020 against the $1,000 aggregate baseline: the pool is up 2%, even though one individual line rose 40%. A per-line alert would misstate the rule. The converse matters too. A single line can be unchanged while the eligible group, taken together, exceeds its baseline by more than 10%. Keep the included rows, amounts and sum visible so a reviewer can reproduce the test. Do not place charges from the zero-tolerance or permitted-to-vary categories into this pool. ## Preserve the baseline and any permitted reset A changed estimate does not automatically reset every tolerance. Section 1026.19(e)(3)(iv) permits a revised estimate for specified changed circumstances, consumer-requested changes, rate locks, certain delayed settlements involving new construction, and other listed events. The creditor must meet the rule’s timing and disclosure conditions. Record the stated event, when it became known, the revised LE date and the specific charges affected. Compare the CD to the applicable disclosed baseline for each charge, not simply to the first LE or the latest file in a folder. Keep both original and revised values and the reason a revised estimate applies. Where no qualifying event and timely disclosure can be evidenced, do not treat a later number as a valid reset. ## Turn the comparison into an audit trail A useful reconciliation record ties each conclusion to its inputs: - Identity: LE and CD versions, dates, loan and property identifiers, and page references. - Charge match: printed descriptions, service, provider, consumer shopping status and evidence used to link rows. - Bucket: zero, 10% aggregate or permitted to vary, with the facts supporting that classification. - Calculation: applicable baseline, CD amount, included 10% pool rows and aggregate result. - Change evidence: any claimed changed circumstance or other permitted event, dated notice and affected amounts. - Disposition: within tolerance, potential excess requiring review, or not measurable from the documents provided. Keep OCR output linked to its page and region, and distinguish a blank or unreadable field from a confirmed zero. Extraction can organize the evidence and perform a reproducible arithmetic comparison; a qualified human should determine how the rule applies to the transaction. ## Keep the two reviews distinct This article compares the LE with the CD to organize TRID tolerance analysis. A separate closing review may compare the CD with the title commitment to connect payoffs, vesting, requirements and recorded exceptions. That is the scope of [the Closing Disclosure and title commitment reconciliation](blog-closing-disclosure-title-commitment-reconciliation.html); it does not answer whether an LE charge fits a tolerance bucket. For the governing text, see the CFPB’s [12 CFR 1026.19(e)(3)](https://www.consumerfinance.gov/rules-policy/regulations/1026/19/), its [consumer explanation of final mortgage-cost changes](https://www.consumerfinance.gov/ask-cfpb/can-my-final-mortgage-costs-increase-from-what-was-on-my-loan-estimate-en-172/), and the [Closing Disclosure guide](https://www.consumerfinance.gov/owning-a-home/closing-disclosure/). This overview is educational and does not provide legal advice or decide whether a specific charge complies. Mortgage operations ### A reviewable trail from form to finding. See how Cogneris turns complex document packages into structured records with source evidence and reviewer-ready exceptions. [Explore document extraction](solutions-real-estate.html) Read next[ ### The fees balance. The exceptions do not. A separate reconciliation of the Closing Disclosure with the title commitment. ](blog-closing-disclosure-title-commitment-reconciliation.html)[ ### A balance does not prove its source. Trace deposits, ownership and records before treating funds as ready for closing. ](blog-proof-of-funds-source-tracing.html) ## Source: state-of-document-ai-2026.html [Skip to content](#main) Annual report · 2026 # State of Document AI 2026. Numbers, not hot takes. Where the document AI market is in May 2026 — adoption, accuracy, pricing, vendor concentration, regulation, and what we expect by Q1 2027. Aggregated from public market research, vendor disclosures, our own benchmarks, and analysis of customer-facing security questionnaires across the category. Read time: 12 minutes. Published 2026-05-07 · Methodology in §10 ## 1. Executive summary The document AI market entered 2026 in its first full year as a foundation-model-native category. The five most important things to know: Finding 1 · Market The global IDP market sits at roughly $4.0–4.4B in 2026, with North America at ~47% share and Europe growing 13.4% CAGR. The category is real, the growth is real, and the buyer pool is broad — not concentrated in tech. Finding 2 · Accuracy Field-level accuracy on common document types (invoices, statements, payslips, KYC documents) crossed the 99% threshold in 2025 and is no longer the differentiator. The gap that matters now is auto-approval rate — what fraction of documents can ship to downstream systems without human review. That sits at 70–90% on simple documents and 30–50% on complex ones. Finding 3 · Pricing Per-page pricing dropped roughly 40% from 2024 to 2026 as LLM inference costs collapsed. Per-call pricing is fading; pay-per-page is now the dominant SMB/mid-market motion, with annual commits returning at $50k+ deal sizes. Finding 4 · Regulation August 2, 2026 is the deadline that shapes every enterprise procurement conversation right now. The EU AI Act's high-risk obligations and Article 50 transparency requirements apply on that date to any AI system serving EU customers — including US-headquartered SaaS. ~50% of enterprises lack a systematic AI inventory; the gap is the opportunity. Finding 5 · Vendor concentration The category remains fragmented. ABBYY, Hyperscience, Rossum, Docsumo, and Nanonets each hold under 8% share; cloud-platform primitives (Google Document AI, Azure Document Intelligence) take a larger but workflow-thin slice. Multiple architectural approaches still compete: single-pass, multi-pass with verification, agentic ReAct, hybrid OCR-first. No clear consolidation winner yet. ## 2. Market size & growth Public market research firms differ on absolute numbers — by a factor of 4× in some cases, depending on whether the analyst counts cloud-platform OCR usage in the IDP envelope. Triangulating across Fortune Business Insights, Grand View Research, Mordor Intelligence, Precedence Research, and Market Data Forecast (see §9), the consensus midpoint for 2026 is: ~$4.0–4.4B Global IDP market, 2026 ~47% North America market share ~$0.91B European IDP market, 2026 13.4% EU CAGR 2026–2034 ### What's driving the growth Cost compression unlocks new use cases. Per-page pricing fell roughly 40% from 2024 to 2026; document types that were unit-economically marginal at $0.20/page (long-form contracts, multi-page medical bills, full underwriting bundles) become viable at $0.10/page. Compliance as a buying trigger. The EU AI Act's August 2026 deadline forces every regulated enterprise to inventory their AI systems. IDP — where AI touches financial, legal, and personal data — is the highest-priority audit target. Compliance documentation that vendors didn't need in 2024 is now procurement-blocking. Native LLM accuracy on long-tail formats. Pre-2023 IDP required template training per format. Modern foundation models generalize to unseen formats out of the box, which collapses the time-to-value from months (template training) to days (schema configuration). ### Geographic split | Region | 2026 size | CAGR | Defining force | | North America | ~$1.5–2.0B | ~22% | API-first adoption; speed-to-deploy | | Europe | ~$0.91B | 13.4% | EU AI Act + GDPR; data-residency requirements | | Asia-Pacific | ~$0.7–0.9B | ~18% | Banking and government digitization | | LATAM + Rest of World | ~$0.4–0.6B | ~15% | LGPD (Brazil); fintech adoption | ## 3. Adoption by industry Document AI adoption isn't uniform. The seven industries driving 2026 demand, ranked by share of paid IDP spend (estimated): - Banking, lending, and insurance (BFSI) — ~38% of category spend. KYC, claims processing, AML, mortgage underwriting, bank-statement extraction. Strong regulatory tailwind. - Healthcare — ~14%. Medical-bill line-item parsing, prior authorization, clinical-document extraction. HIPAA BAA is procurement-blocking. - Logistics and supply chain — ~11%. Bill of lading, customs documents, freight invoices, POD reconciliation. - Legal and professional services — ~9%. Contract review, M&A diligence, regulatory filings. - Government and public sector — ~8%. Tax processing, benefit applications, regulatory submissions. Slow buying cycles, large contract sizes. - Hospitality, travel, and retail — ~7%. Identity verification, expense management, supplier onboarding. - Finance & ops at every other industry — ~13%. AP automation as a horizontal use case across the long tail. Adoption maturity varies dramatically by industry. Banking is in the "second-vendor" phase — most large banks ran a first IDP project in 2022–2024, are now consolidating onto fewer vendors, and rebid every 24–36 months. Healthcare is still in the "first-vendor" phase for most providers and payers — selection happens in 2026 for first-time adopters. Government is on a 5–7 year cycle and most procurements signed in 2026 will be deployed in 2027–2028. ## 4. Accuracy benchmarks "99% accuracy" is the most over-claimed number in the category, and the metric every vendor leans on. The realistic 2026 picture, based on independent benchmarks and our own production data: | Document type | Field accuracy | Auto-approval rate | Note | | Invoices (clean PDF) | 99.0–99.5% | 80–90% | Mature; line-item table the long pole | | Bank statements (digital PDF) | 98.5–99.3% | 75–88% | Multi-page stitching the long pole | | Payslips | 98.5–99.5% | 78–90% | Year-codes and special pay items still tricky | | KYC documents (passports, IDs) | 99.0–99.7% | 72–85% | Damaged-image robustness is the differentiator | | Receipts (smartphone photo) | 96–98% | 62–78% | Thermal print + glare drop accuracy 3–5pts | | Tax returns (1040, Schedule C) | 98–99% | 65–80% | Cross-form references catch many errors | | Medical bills (CPT/ICD-10) | 97–98.5% | 50–70% | Coding precision needs domain models | | Contracts (MSA, SOW, NDA) | 95–98% | 30–50% | Citations and deviation flags > raw extraction | The metric buyers should ask for is not field-level accuracy. It is auto-approval rate on a benchmark you picked from your documents. Document-level accuracy — whether every field in a document is correct — is roughly (field-accuracy)^N for an N-field document, ignoring correlation. A 99% field-level rate on a 12-field document gives you ~88.6% document-level accuracy, meaning 1 in 9 documents has at least one wrong field. This is why HITL routing matters: the model handles the easy 70–90%, humans handle the rest, and the combined system reaches 99.9% effective accuracy. ## 5. Pricing trends Per-page pricing fell from a ~$0.10–0.30 typical band in 2024 to ~$0.05–0.18 in 2026. The drop tracks LLM inference costs (down roughly 10× from 2023 to 2025, another 3–5× expected by end of 2026) and increased competition. Three pricing models still compete: - Pay-per-page — dominant for SMB and mid-market. Most transparent. Predictable as volume scales. - Pay-per-call (or pay-per-document) — common at OCR-first vendors. Misaligned with multi-page documents — a 12-page bank statement counts as one call but is 12× the work. - Annual subscription with included volume — returning at the high end ($50k–$1M deal sizes). Discounts kick in at 100k+ pages/month. Setup and integration fees fell faster than per-page rates — most vendors now waive setup for self-serve plans and charge $5–25k for enterprise deployments that include custom connector work, security review, and dedicated support onboarding. Procurement is increasingly comfortable with per-unit AI pricing (compared to traditional SaaS seat pricing). The shift maps to how AI cost behaves — variable with usage, not fixed per user. ## 6. Vendor landscape The category is fragmented. No vendor holds more than ~8% category share by revenue. The main groupings: | Group | Examples | Strength | Weakness | | Established enterprise IDP | ABBYY, Hyperscience | Compliance posture, large-bank references, complex docs | Long sales cycles, dated UX, on-prem-leaning | | API-first invoice/AP | Rossum, Docsumo, Nanonets, Klippa | Time-to-deploy, transparent pricing, mid-market fit | Less depth on multi-document workflows; thinner enterprise references | | Cloud-platform primitives | Google Document AI, AWS Textract, Azure Document Intelligence | Inherits cloud compliance; deep integration with same-cloud tooling | Workflow-thin; you assemble the rest | | Vertical specialists | Ocrolus (lending), Ephesoft (gov), Indico (insurance) | Domain-specific accuracy + buyer fit | Limited beyond their vertical | | Agentic / multi-LLM newer entrants | Cogneris (us), Reducto, Unstract, Mendable's Doc | ReAct architecture, multi-LLM, audit-trail depth | Smaller customer footprints, fewer trust signals | Two cuts that matter to buyers: Architecture. Single-pass extraction (one LLM call per doc) vs multi-step / agentic (planner + extractor + validator + post-processor). Single-pass is fast and cheap; multi-step catches errors single-pass can't. The architectural choice is increasingly disclosed in security questionnaires, but rarely in marketing copy. Compliance posture. SOC 2 Type II, GDPR DPA with EU SCCs, HIPAA BAA on Enterprise, EU AI Act Article 50 disclosure. Vendors that have all four are a small subset; vendors that have none of the four are off the table for regulated enterprise procurement. ## 7. Regulatory landscape The compliance picture in 2026 is more complex than at any point since the GDPR came into force in 2018. ### The EU AI Act (Regulation 2024/1689) The high-risk obligations and Article 50 transparency requirements apply from August 2, 2026. Any AI system placed on the EU market (including by a US-headquartered SaaS) must meet the obligations applicable to its risk classification. For document AI: - Article 50(1) — direct-interaction AI must inform users they're interacting with AI. Document Q&A interfaces are clearly in scope. - Article 50(2) — synthetic content (AI-generated summaries, AI-drafted explanations) must be marked as such, with machine-readable provenance. - Annex III — high-risk uses — creditworthiness assessment, life and health insurance pricing, recruitment, education, law enforcement, and critical infrastructure are explicitly listed. If your customer's workflow falls in Annex III, deployer-side obligations (Fundamental Rights Impact Assessment under Art. 27) apply. Enterprise readiness gaps are wide. ~50% of enterprises lack a systematic inventory of AI systems in production or development. Vendors that ship procurement-ready Provider Cards and Article 50 disclosures pull ahead. ### GDPR and post-Schrems II transfers The EU-US Data Privacy Framework adequacy decision remains in force, but is under legal challenge. Most enterprise customers continue to require EU SCCs (Decision 2021/914) plus a Transfer Impact Assessment as the primary protection — DPF on top, not instead. The UK IDTA and Swiss FDPIC adaptations remain the standard. ### US patchwork and CCPA/CPRA The state-by-state US privacy regime continued to grow. By 2026, ~20 US states have comprehensive privacy laws. Service-provider terms (Cal. Civ. Code §1798.140(ag)) are the most common feature; AI-specific provisions are emerging in some states (Colorado AI Act, NYC AEDT). ### Sector-specific - HIPAA — BAA required for any PHI processing. Most IDP vendors offer BAAs only on Enterprise plans, which gates SMB healthcare adoption. - SOX 404 — auditors increasingly request audit-trail evidence from AI systems used in financial reporting workflows. Retention should be contract-specific and aligned to the customer's applicable record-keeping duties. - NIS2 (EU) — cybersecurity obligations on essential entities, applicable since October 2024. Cascades to processors and sub-processors. - DORA (EU financial services) — operational resilience obligations from January 2025. ICT risk management and incident reporting cascade to vendors. ## 8. 2026–2027 predictions Five predictions for the 12 months from May 2026 — graded as high, medium, or low confidence. 1. Per-page pricing falls another 25–40% by Q2 2027. (High confidence.) LLM inference costs continue compressing; the savings flow through to per-page rates with a 6–9 month lag. 2. The first wave of EU AI Act enforcement actions arrives by Q4 2026. (High confidence.) Article 50 transparency violations are the most likely first-target — they're easy to detect, easy to prove, and don't require a deep technical audit. Expect a mix of "guidance letters" early and one symbolic fine before the end of 2026. 3. Agentic / multi-step IDP becomes the procurement default for high-stakes workflows. (Medium-high confidence.) Single-pass extraction will continue to dominate low-stakes use cases (expense reports, retail receipts) on cost grounds, but the regulated workflows where errors have real consequences (lending, claims, KYC, medical) will increasingly require demonstrably multi-step architectures with audit trails. 4. At least one major IDP-vendor consolidation or shutdown. (Medium confidence.) The category is fragmented and the cost compression squeezes margins. The vendors with the weakest compliance posture or the thinnest moat above commodity inference will be acquired or wind down. We don't predict which. 5. "AI-native" CLM, AP, and KYC products bundle IDP rather than buy from separate IDP vendors. (Medium confidence.) The build-vs-buy line moves toward "buy a workflow product with extraction included" for some segments. IDP vendors that can be embedded (white-label, OEM) capture this; pure-play IDP loses some greenfield deals to bundled offerings. ## 9. Sources Public market sizing in this report is triangulated across: - Fortune Business Insights — IDP Market Size, Trends 2034. - Grand View Research — IDP Market Size Report, 2030. - Mordor Intelligence — IDP Market Size, Share & Industry Trends Report, 2031. - Precedence Research — IDP Market to Hit USD 43.92B by 2034. - Market Data Forecast — Europe IDP Market Size 2034. - Strategic Market Research — IDP Market Report, 2026. - Holland & Knight — US Companies Face EU AI Act's Possible August 2026 Compliance Deadline. - Public sub-processor lists, transparency reports, and DPA templates from ~25 vendors in the category. Accuracy benchmarks are aggregated from our own production data on 18M+ documents processed across 6 industries; vendor-published benchmarks where independently verifiable; and the limited public benchmarks available (the most cited remains the SROIE receipt benchmark, now dated, and the FUNSD form benchmark). ## 10. Methodology & corrections What this report is. A synthesis of public market research, our own production data, vendor-published claims (cross-checked where possible), and analysis of customer-facing security questionnaires across the category. We've tried to call out where our numbers come from and where confidence is lower. What this report isn't. A primary research study. We didn't run a survey of buyers; we didn't interview 100 CIOs. The numbers reflect the public information environment plus our own dataset — not an exhaustive sweep. Conflicts of interest. Cogneris builds in this category. We compete with vendors named in §6. We have not framed any vendor more favorably or unfavorably than the public record supports — but a healthy reader will discount our category-positioning claims accordingly. Corrections. If you spot an error or have data that would tighten a number, email [research@cogneris.ai](mailto:research@cogneris.ai). We'll update the page and credit substantive corrections in the change log. Citation. Cite as: Cogneris, "State of Document AI 2026", May 7 2026, https://cogneris.ai/state-of-document-ai-2026.html ## Want the data behind a specific finding? If you're scoping a procurement, building a board deck, or writing an analyst note, we'll share the underlying tables and the per-vendor compliance comparison on request. [Talk to our team](contact.html). ## Further reading For the category fundamentals, see [the IDP buyer's guide](what-is-idp.html) — what intelligent document processing is, how it differs from OCR, what to look for in 2026. To size a project against your own volume, use [the ROI calculator](roi-calculator.html): monthly savings, payback period, and 3-year TCO. For terminology, the [document AI glossary](glossary.html) covers IDP, OCR, HITL, ReAct, audit trail, and the rest of the vocabulary. ## Source: blog.html [Skip to content](#main) Blog # Engineering, product and documents. For market context, read [the 2026 State of Document AI report](state-of-document-ai-2026.html). For the category basics, see [the IDP buyer's guide](what-is-idp.html). [All](blog.html) [Engineering](blog-tag-engineering.html) [Regulation](blog-tag-regulation.html) [Product](blog-tag-product.html) [Case](blog-tag-case.html) [ Engineering pillar Apr 30, 2026 · 14 min ## ReAct architecture for documents A technical deep-dive on why Cogneris's document extraction is built on a ReAct agentic architecture rather than pure RAG, classical OCR, or single-pass LLM extraction — with three worked examples and an honest take on when not to use it. Read article ](pillar-react-architecture.html) [ FinanceNov 6, 2026 ### Form 1098 reports the interest, not the deduction. Map each box of Form 1098, link the forms a sale or refinance creates, and leave the deduction to the return. 10 min read ](blog-form-1098-mortgage-interest-statement.html) [ FinanceNov 5, 2026 ### A deposit should leave the final invoice once. Tell a prepayment invoice from a pro forma, deduct the advance with its tax, and keep a balance so it comes off only once. 11 min read ](blog-prepayment-invoice.html) [ EngineeringNov 4, 2026 ### A subrogation demand has to prove each number. Trace each amount to its document, put the deductible in the demand, and catch a waiver before the letter goes out. 10 min read ](blog-subrogation-demand-package.html) [ FinanceNov 3, 2026 ### An underwriting condition is a document request. Match each condition with its requested evidence, record what is still missing, and leave sufficiency decisions with the reviewer. 6 min read ](blog-mortgage-condition-document-evidence.html) [ FinanceNov 1, 2026 ### The payment arrives with a list of invoices. One payment can pay several invoices, short-pay two and cover an order never billed. Post it the way the remittance advice says, line by line. 11 min read ](blog-cash-application-remittance-advice.html) [ EngineeringOct 31, 2026 ### A freight invoice needs a rate check. Compare carrier charges with the agreed rate, shipment records, and delivery evidence before routing exceptions for review. 7 min read ](blog-freight-invoice-rate-audit.html) [ FinanceOct 31, 2026 ### The 1099-K shows the volume, not the income. A 1099-K counts gross payments before fees and refunds. Bridge it to the deposits and the tax return before calling any of it income. 12 min read ](blog-1099-k-income-verification.html) [ FinanceOct 30, 2026 ### The new balance on a card statement has its own math. Payments and credits lower a card balance; purchases, fees and interest raise it. Read every section of the statement and check the new balance closes. 11 min read ](blog-credit-card-statement-extraction.html) [ FinanceOct 29, 2026 ### The sell sheet sets the rebate terms. Match a retailer’s deduction to the promotion’s dates, eligible products, calculation base and cap, then make exceptions reviewable. 7 min read ](blog-trade-promotion-rebate-deduction.html) [ EngineeringOct 28, 2026 ### Every option has a date that counts. Track talent agreement options and cancellation fees from the clause to the trigger, deadline, and payment evidence. 7 min read ](blog-talent-agreement-option-dates.html) [ FinanceOct 28, 2026 ### The BEO is the contract for the event. The sales contract sets the minimum; each signed BEO sets the guarantee, price per head and service charge. Check the final event invoice against them. 11 min read ](blog-banquet-event-order.html) [ EngineeringOct 27, 2026 ### The tolerance follows the bucket. Compare Loan Estimate and Closing Disclosure charges by zero, 10% aggregate and permitted-to-vary categories, with the evidence behind each match. 11 min read ](blog-loan-estimate-closing-disclosure-tolerance.html) [ FinanceOct 27, 2026 ### The invoice without VAT can still be right. A reverse charge invoice is zero-tax on purpose. Read the mention and both VAT numbers, tell it from an exempt supply, and return it when they are missing. 13 min read ](blog-reverse-charge-invoice.html) [ EngineeringOct 26, 2026 ### The ASN is checked one pallet at a time. Join the advance ship notice to the dock by the SSCC on each pallet label, compare product and count pallet by pallet, and roll up to the order line last. 12 min read ](blog-advance-ship-notice-reconciliation.html) [ FinanceOct 25, 2026 ### The blank 1099-B box does not mean zero. Keep missing cost basis distinct from zero, link a noncovered lot to its source records, and route unresolved values for review. 6 min read ](blog-1099-b-cost-basis.html) [ EngineeringOct 25, 2026 ### The registration number has to match every document. Use the registration number as the key across a KYB packet, match every name on the certificate, tax form and bank letter to it with its date, and route each disagreement. 13 min read ](blog-kyb-document-verification.html) [ EngineeringOct 24, 2026 ### The group contract date that costs money. Room-block cutoffs, attrition thresholds and food-and-beverage minimums each have a date, trigger and calculation to extract. 8 min read ](blog-hotel-group-contract-attrition.html) [ EngineeringOct 24, 2026 ### A condo questionnaire answer needs its document. Read each condo questionnaire answer beside the budget, aging report, reserve study or minutes that should support it, recompute the ratios and flag every disagreement. 13 min read ](blog-condo-questionnaire.html) [ EngineeringOct 23, 2026 ### A rights window is territory, media and dates. Extract each grant in a license agreement as a row of territory, media, start, end and exclusivity, with the clause and page beside it, and surface overlaps. 12 min read ](blog-content-rights-window.html) [ EngineeringOct 22, 2026 ### The invoice does not name the account. Propose the GL account and cost center from purchase orders and posting history, keep the evidence beside each suggestion, and route disagreements to a person. 13 min read ](blog-invoice-gl-coding.html) [ EngineeringOct 20, 2026 ### The K-1 number is in the statement. Link each Schedule K-1 box marked with an asterisk or STMT to the statement line that carries its value, tie the totals that can tie, and name what is missing. 12 min read ](blog-k1-supplemental-statement.html) [ EngineeringOct 19, 2026 ### The return packet has to match the sale. Read the RMA, return label, packing slip and serial number as fields against the original order, and route each mismatch to a person instead of a verdict. 12 min read ](blog-return-merchandise-authorization.html) [ EngineeringOct 18, 2026 ### The estoppel certificate closes the association’s account. Read an HOA estoppel certificate for its date of issuance, per-day schedule and other associations, then check each field against the title file. 14 min read ](blog-hoa-estoppel-certificate.html) [ EngineeringOct 17, 2026 ### A check date needs a second look. Separate stale-dated, post-dated and unreadable check dates before a payment is routed, and preserve the evidence for review. 10 min read ](blog-check-date-review.html) [ EngineeringOct 17, 2026 ### The deadline starts when the order is served. Read a levy or garnishment order for its receipt date, debtor identifiers and attachments, and match it to account holders before any funds move. 18 min read ](blog-bank-levy-garnishment-order.html) [ EngineeringOct 16, 2026 ### The commercial invoice is the customs record. Read a commercial invoice against the packing list and the certificate of origin, field by field, and route each mismatch to its owner. 16 min read ](blog-commercial-invoice-hs-code.html) [ EngineeringOct 15, 2026 ### The declarations page is the policy on one sheet. Read an insurance declarations page against the application, the certificate and the lender’s terms, and route each mismatch to its owner. 14 min read ](blog-insurance-declarations-page.html) [ EngineeringOct 14, 2026 ### The statement is what the supplier thinks you owe. Match a vendor statement to the payables ledger by invoice reference, and route each unmatched line to the owner of its cause. 13 min read ](blog-vendor-statement-reconciliation.html) [ EngineeringOct 13, 2026 ### The itemized bill should add up to the claim. Roll an itemized hospital bill up by revenue code the way the UB-04 instructions say, match each total, and route every difference with both sources. 12 min read ](blog-itemized-bill-ub04-reconciliation.html) [ EngineeringOct 12, 2026 ### The appraisal has a date and comparables. Effective date, contract price, concessions and comparable sales: what extraction must capture from an appraisal report so each field can be checked against the loan file. 16 min read ](blog-appraisal-report-extraction.html) [ EngineeringOct 11, 2026 ### A royalty statement starts with the contract. Compare a statement’s royalty base, rate, threshold, recoupment and reporting period with the terms that apply. 6 min read ](blog-royalty-statement-reconciliation.html) [ EngineeringOct 11, 2026 ### The purchase order changed after you matched it. Revisions, change orders and blanket releases: which version of a purchase order an invoice line has to be matched against, and what extraction must capture. 13 min read ](blog-purchase-order-change-orders.html) [ EngineeringOct 10, 2026 ### The first EOB starts the second claim. Carry the primary payer’s allowed amount, payment, adjustments and member responsibility into a reviewable secondary-claim handoff. 12 min read ](blog-secondary-eob-coordination-benefits.html) [ EngineeringOct 9, 2026 ### A fuel bill of lading is a tax record. Reading a rack bill of lading for fuel tax reporting: gross and net gallons, destination state, the shared document number and the dyed fuel notice. 13 min read ](blog-fuel-bill-of-lading-tax-reporting.html) [ EngineeringOct 8, 2026 ### A hotel folio is not a receipt. Reading a hotel folio line by line: the room against a nightly limit, tax that follows the room, meals, credits and split folios. 12 min read ](blog-hotel-folio-ocr-line-items.html) [ EngineeringOct 7, 2026 ### The return declares. The transcript confirms. Reading a borrower’s tax return against the IRS transcript: what Form 4506-C asks for, what each transcript covers, and which gaps are expected. 14 min read ](blog-tax-transcript-document-verification.html) [ InsuranceOct 7, 2026 ### The valuation date changes the loss run. Compare paid claims, reserves, carrier totals and policy years without turning missing data into a guess. 9 min read ](blog-loss-run-valuation-date.html) [ EngineeringOct 6, 2026 ### The fees balance. The exceptions do not. Reconciling a Closing Disclosure with the title commitment: vesting, payoffs, Schedule B exceptions and the requirements that clear them. 16 min read ](blog-closing-disclosure-title-commitment-reconciliation.html) [ EngineeringOct 5, 2026 ### The OCR was right. SAP still said no. Accurate invoice fields can still fail posting checks when supplier, company-code, tax or account-assignment data does not resolve. 8 min read ](blog-sap-invoice-master-data-validation.html) [ EngineeringOct 5, 2026 ### A lab value without its range is not data. Value, unit, reference interval, flag and collection time are one statement. What lab report extraction has to keep together, read from the CLIA rules. 14 min read ](blog-lab-report-extraction-reference-range.html) [ EngineeringOct 4, 2026 ### A balance does not prove its source. A bank balance shows what is there. Trace deposits, ownership and source records before treating a number as a complete funds story. 11 min read ](blog-proof-of-funds-source-tracing.html) [ EngineeringOct 4, 2026 ### Know the institution before the data travels. The Travel Rule sends customer details to another institution. Read its registration, licences and questionnaire as one packet, and keep their dates. 13 min read ](blog-travel-rule-counterparty-due-diligence.html) [ EngineeringOct 3, 2026 ### Approved hours meet the invoice. Match supplier bills to approved time by worker, role, date, rate, and policy. 10 min read ](blog-timesheet-invoice-reconciliation.html) [ EngineeringOct 3, 2026 ### A claim does not end at first notice. Estimates, supplements and invoices keep revising each other after the first notice. Read each for its role and derive the claim’s amounts from the sequence. 13 min read ](blog-claims-document-processing-after-fnol.html) [ EngineeringOct 2, 2026 ### A tax form is a claim, not a fact. A W-8BEN or CRS self-certification is the holder’s own statement. Compare it with the rest of the packet and route each contradiction to review with its cure. 14 min read ](blog-tax-residency-self-certification.html) [ EngineeringOct 1, 2026 ### CMS-1500 and UB-04 are not one schema. A professional claim line is a service; an institutional line is a charge category. Classify the form before you decide what units, dates and NPIs mean. 13 min read ](blog-cms-1500-ub-04-claim-form-extraction.html) [ EngineeringSep 30, 2026 ### The document you extracted is not the contract. An MSA, its order forms and amendments are one contract, ranked by a precedence clause. Read that clause first, or cite the value that does not govern. 13 min read ](blog-contract-order-of-precedence.html) [ EngineeringSep 29, 2026 ### A rent roll is a forecast, not a statement. A rent roll lists what each unit should pay, not what the property collected. Vacancy, concessions, delinquency and reimbursements are where the two part. 12 min read ](blog-rent-roll-extraction-underwriting.html) [ EngineeringSep 28, 2026 ### A valid MRZ can still be the wrong document. An MRZ check digit proves a field was read consistently, not that it is right or genuine. The visual zone is the second reading that catches what it misses. 13 min read ](blog-mrz-check-digit-visual-zone.html) [ EngineeringSep 27, 2026 ### The document that has not arrived yet. A missing 1099 or K-1 is an availability signal. Keep the expected source, review evidence and follow-up visible without inventing a tax conclusion. 9 min read ](blog-1099-k1-document-availability.html) [ EngineeringSep 27, 2026 ### A credit note is not a negative invoice. A credit note carries its sign in its type, not its amounts, and applies to an origin invoice: in part, across periods, or through review when it names none. 13 min read ](blog-credit-memo-processing.html) [ EngineeringSep 26, 2026 ### A certificate of insurance expires without notice. A certificate lists one expiry per policy and announces none of them: read every row, keep additional insured as a claim and renew before the earliest date. 11 min read ](blog-certificate-of-insurance-tracking.html) [ EngineeringSep 25, 2026 ### The payroll register does not close per employee. A payroll register is posted, not verified: map every code to a ledger account, carry the employer taxes and stop the close on a code nobody mapped. 11 min read ](blog-payroll-register-gl-reconciliation.html) [ EngineeringSep 24, 2026 ### The document is real. It still does not qualify. Why most rejected proofs of address are genuine: currency, holder, type and address rules fail real documents, and each needs its own reason code. 12 min read ](blog-proof-of-address-acceptance-rules.html) [ EngineeringSep 23, 2026 ### An adjustment code is not the cause. Why CARC and RARC codes explain an adjustment without deciding the clinical, administrative or payer workflow behind it. 10 min read ](blog-eob-carc-rarc-routing.html) [ EngineeringSep 23, 2026 ### The net profit is not the income. Why a self-employed borrower’s return has no income line: Schedule C and K-1 figures were computed to settle tax, not to describe capacity. 11 min read ](blog-tax-return-extraction-self-employed-borrower.html) [ EngineeringSep 22, 2026 ### The receipt proves the purchase, not the policy. Why receipt OCR can read every field and still leave the expense undecided: purpose, attendees, the trip and a dated limit. 11 min read ](blog-receipt-ocr-expense-policy.html) [ EngineeringSep 21, 2026 ### Deprecation is a contract, not an email. What to require before a document API retires a version, when the schema stays put and the values move. 10 min read ](blog-api-deprecation-policy-checklist.html) [ EngineeringSep 20, 2026 ### Two deposits can be one check. Why the duplicate key is the serial on the MICR line, and never the picture of the page. 11 min read ](blog-duplicate-check-deposit-detection.html) [ EngineeringSep 19, 2026 ### A delivery is proven by its exceptions. Why a shortage claim is argued from the counts, exception notes and timestamps beside the signature. 10 min read ](blog-proof-of-delivery-dispute-evidence.html) [ ProductSep 18, 2026 ### Eligibility starts at the front desk. Why both sides of an insurance card belong in one evidence bundle before intake asks a payer a question. 10 min read ](blog-health-insurance-card-eligibility.html) [ EngineeringSep 18, 2026 ### The paystub and the W-2 should not match. Why an income packet disagrees with itself by design, and which residual is worth a reviewer’s time. 10 min read ](blog-income-verification-paystub-w2-deposit.html) [ EngineeringSep 18, 2026 ### Invoice matching needs context. Why OCR is only the first witness, and how purchase orders, receipts, and tolerances explain the line that needs review. 10 min read ](blog-invoice-reconciliation-three-way-match.html) [ EngineeringSep 17, 2026 ### The text layer can say something the page doesn’t. Why a searchable PDF can disagree with its own scan, and how to test the layer before search depends on it. 10 min read ](blog-searchable-pdf-text-layer.html) [ EngineeringSep 16, 2026 ### A submission is more than a form. Why ACORD forms, loss runs, schedules, and evidence belong in one reviewable submission packet. 10 min read ](blog-acord-submission-packet.html) [ EngineeringSep 16, 2026 ### Four documents, one spread, one cited cell. Why a credit spread is a reconciliation, and every cell should name its source line. 10 min read ](blog-financial-spreading-credit-packet.html) [ EngineeringSep 15, 2026 ### Structured invoices still leave exceptions. Why e-invoicing reduces keying but still needs matching, evidence, and review. 10 min read ](blog-e-invoicing-exception-queue.html) [ EngineeringSep 15, 2026 ### Retrieve the passage, or extract the field. Decide per question when a RAG system needs chunks, validated fields, or both. 10 min read ](blog-rag-chunking-vs-field-extraction.html) [ EngineeringSep 14, 2026 ### Three ways to reduce document exposure. Compare redaction, pseudonymization, and least-data extraction by residual risk. 9 min read ](blog-document-anonymization.html) [ EngineeringSep 14, 2026 ### The statement is worth its lines, not its balance. Classify recurring income, obligations and reversals line by line, with evidence. 9 min read ](blog-bank-statement-transaction-categorization.html) [ EngineeringSep 13, 2026 ### Test the pipeline without borrowing customer documents. Separate plumbing, synthetic and real-sample tests, and approve the data first. 11 min read ](blog-document-ai-sandbox-test-data.html) [ EngineeringSep 12, 2026 ### Discover the fields before you define them. Use field discovery as evidence for a reviewed, versioned schema. 11 min read ](blog-zero-shot-field-discovery.html) [ EngineeringSep 11, 2026 ### Two photos, one person, a stated threshold. A one-to-one comparison, a threshold you can defend, and a third answer that needs a budget. 13 min read ](blog-face-match-document-verification.html) [ EngineeringSep 10, 2026 ### The limit is part of the contract. A fixed window, a queue on your side, and two refusals that mean different things. 12 min read ](blog-api-rate-limits-backpressure.html) [ EngineeringSep 9, 2026 ### The language is not the hard part. Date order, decimal marks, name order and reading order break long after the alphabet stops being a problem. 9 min read ](blog-multilingual-document-processing.html) [ EngineeringSep 8, 2026 ### A credential should have a small blast radius. Bound each API key by workload, environment and enforced scope, then rotate it with evidence instead of hope. 7 min read ](blog-api-key-scope-rotation.html) [ EngineeringSep 8, 2026 ### The record is the episode, not the file. Assemble the note, result, authorization and EOB into one episode of care instead of four unrelated extractions. 9 min ](blog-medical-record-abstraction.html) [ EngineeringSep 7, 2026 ### Return the gap, not the workflow. Reopen only the evidence a replacement invalidates while preserving every approval whose inputs did not change. 12 min ](blog-partial-document-resubmission.html) [ EngineeringSep 7, 2026 ### Retention is a schedule, not a checkbox. A clock per document class, a legal hold that outranks it, and deletion you can prove without keeping the data. 13 min ](blog-configurable-retention-legal-hold.html) [ EngineeringSep 6, 2026 ### Same applicant, a cleaner file. Blocklists are decisions: defensible keys, tiered expiry, a real appeal path, and a measured false-positive budget. 11 min ](blog-repeat-applicant-blocklist.html) [ EngineeringSep 5, 2026 ### Price the workflow before it runs. Estimate every step, retry and review path before execution, then reconcile assumptions against the first measured run. 11 min ](blog-workflow-cost-estimation.html) [ Engineering Sep 5, 2026 ### Agents do not read your docs. They fetch them. How to measure an agent time to first call, in fetches and bytes, and what the measurement found on our own machine-readable surface. 10 min ](blog-machine-readable-documentation.html) [ Engineering Sep 4, 2026 ### Region is a decision, not a setting. What data residency actually decides, what stays identical across regions, and why the choice is fixed at tenant creation. 11 min ](blog-document-ai-data-residency.html) [ Engineering Sep 3, 2026 ### A connector is not a migration. Live queue, backfill, reprocessing, portable access and cost — the systems a document connector needs after the first sync. 12 min ](blog-continuous-document-processing.html) [ Engineering Sep 2, 2026 ### A field without context is a guess. Keep line items, totals, semantic relationships and source evidence connected in structured document extraction output. 12 min ](blog-structured-json-evidence-relationships.html) [ Engineering Sep 2, 2026 ### The callback is the contract. Verify the signature over raw bytes, dedupe on the delivery id, return 2xx fast, and replay what the retry schedule missed. 11 min ](blog-webhook-callback-contract.html) [ Product Sep 1, 2026 ### A lease is a calendar with money attached. Turn clauses into portfolio operations: critical dates, rent schedules, option windows, obligations, amendment lineage and evidence. 12 min ](blog-lease-abstraction-portfolio-data.html) [ Engineering Sep 1, 2026 ### The seams are where long documents break. Process long, multi-class files as parts: split on structure, budget context per part, retry only what failed and reassemble the evidence. 13 min ](blog-long-document-deep-split.html) [ Product Aug 31, 2026 ### Your labels should leave with you. Make ground truth portable: define ownership, export format, provenance and cost per accepted field before annotation begins. 12 min ](blog-ground-truth-dataset-ownership.html) [ Engineering Aug 31, 2026 ### The file parsed. The table did not. Nested tables, columns and rotation break in parsing, not in capture. Measure layout fidelity with a declared test set and per-class scores. 12 min ](blog-layout-fidelity-parsing-metric.html) [ Engineering Aug 30, 2026 ### Four documents. One shipment. One truth. Reconcile air waybills, packing lists, shipping labels and customs declarations across weight, quantity and consignee before cargo moves. 11 min ](blog-air-waybill-shipment-reconciliation.html) [ Product Aug 30, 2026 ### The first mile is a product Case completion, not upload success, is the metric for a document collection portal: checklist contracts, expiring links, guided resend and reopening. 11 min ](blog-document-collection-portal.html) [ Engineering Aug 30, 2026 ### Catch the bad photo on device A blurred document fails days later in a review queue. Two convolutions on the device and one round trip move that rejection to the moment of capture. 11 min ](blog-capture-time-quality-classification.html) [ Engineering Aug 30, 2026 ### Liveness without keeping a face A live face has a pulse the camera can see. Remote photoplethysmography turns liveness into signal processing, with no biometric template retained. 12 min ](blog-passive-liveness-rppg.html) [ Engineering Aug 30, 2026 ### Publishing a flow is a release Publishing a workflow is a release, not a setting. Immutable versions, pinned runs and a graph that survives publication keep conditional routing honest. 12 min ](blog-workflow-versioning-audit.html) [ Engineering Aug 30, 2026 ### From extracted field to finished document Extracted fields populate the document you send back. Structured markers, a published data dictionary and strict rendering stop a placeholder shipping. 11 min ](blog-document-generation-templates.html) [ Product Aug 30, 2026 ### Signature belongs inside the flow Buying signature separately means exporting a PDF and losing the case. Envelope, signer authentication, seal and timestamp belong in the same run. 11 min ](blog-signature-as-workflow-step.html) [ Engineering Aug 30, 2026 ### Handwriting is where models guess Handwriting fails as a clean, plausible, wrong value. Measure accuracy per writing class, calibrate the score, and make abstention an explicit outcome — so doubt reaches review instead of the database. 12 min ](blog-handwriting-ocr-icr.html) [ Engineering Aug 29, 2026 ### Classify before you extract Document classification is a routing decision: it selects the schema, validation rules, model and exception path. Measure it on route quality, latency, cost and review load — not label accuracy alone. 12 min ](blog-document-classification-routing.html) [ Engineering Aug 29, 2026 ### Accuracy per field is not completion A completion rate with no stated denominator is a marketing number. This is the definition that survives an audit: a written predicate, itemized exclusions, per-type segments, a correction window and a cost per document. 13 min ](blog-document-ai-completion-rate.html) [ Engineering Aug 28, 2026 ### One parser needs a contract A multi-format document parser should give every consumer one stable interface without flattening PDFs, Office files, images and spreadsheets into the same lossy text stream. This contract keeps native structure, typed provenance, partial errors and per-format fidelity tests explicit. 12 min ](blog-document-ingestion-contract.html) [ Engineering Aug 28, 2026 ### Your inbox is not a workflow Forwarding an attachment is transport, not automation. A production inbox needs explicit message identity, idempotent delivery, separate threading and deduplication rules, hostile-file containment, named exception states, and an audit trail from source bytes to the system-of-record receipt. ](blog-inbox-document-processing.html) [ Engineering Aug 27, 2026 ### Coordinates are part of the contract Bounding boxes usually ship as a convenience for highlighting, and three widely used OCR APIs already return three different coordinate systems. This is the version that survives an audit: a region schema that names its space and origin, separate page and region rotation, an explicit page rule per format, a versioned convention, and tests that render the boxes instead of trusting the numbers. 12 min ](blog-bounding-boxes-api-contract.html) [ Engineering Aug 26, 2026 ### Self-hosting moves the boundary Moving OCR into your cloud account or data centre does not remove operational risk. It moves the ownership boundary. This practical decision matrix compares managed SaaS, private VPC, private cloud and on-prem across data paths, operating responsibilities, model updates, observability, threat boundaries and total cost. 11 min ](blog-self-hosted-document-ai.html) [ Engineering Aug 25, 2026 ### A tuning loop needs a leash An optimizer let loose on an intelligent document processing pipeline will find a configuration nobody chose and nobody can explain. This is the version that stays inspectable: a declared parameter search space, a four-term objective function that prices review load and abstention, guardrails checked before promotion, and a rollback that is one action. 12 min ](blog-autonomous-idp-optimization.html) [ Engineering Aug 24, 2026 ### One score hides two failures A document intelligence benchmark can report the right answer for the wrong reason, or the wrong answer from perfectly read text. This reproducible framework separates parsing fidelity from reasoning quality, scores evidence and abstention, reports by document characteristic, and turns the result into a diagnostic queue instead of a single leaderboard number. 12 min ](blog-document-intelligence-benchmark.html) [ Product Aug 20, 2026 ### Someone has to maintain the model. Decide who. The prebuilt-or-custom question gets run as an accuracy contest on a folder of sample documents and settled, months later, by maintenance. The four different artifacts vendors call a model, the eight dimensions that decide between them, why layout variability rather than volume is the first question, why the answer is per field rather than per document type, why labelling and not training time is the cost line, the regression set you need on either path, what governance asks of each, the portability test to run before the first label, the hybrid that ships, the four ways this goes wrong, and when each answer is honestly the right one. 17 min ](blog-prebuilt-vs-custom-extraction-models.html) [ Engineering Aug 19, 2026 ### Not every page deserves a frontier model One pipeline, one model, every page is easy to reason about and it sets your gross margin to whatever the largest model in your stack costs. Routing across OCR, layout models, vision language models, frontier reasoning and CPU or GPU is the right idea, and it is usually described in a way that hides every hard part. The four separate decisions a router makes, why the signal has to be evidence about the page rather than a label on the document, what each tier is honestly good at, why escalation must trigger on a failed check and never on a confidence score, how residency and latency constrain placement before cost gets a vote, cost per accepted field as the unit that makes the economics legible, why the routing policy is a change-control event rather than a config tweak, the four ways it goes wrong including fallback that hides the failure, and when one model for everything is the correct answer. 17 min ](blog-inference-routing-document-ai.html) [ Product Aug 18, 2026 ### Your exception queue is the real product Every document platform reports a straight-through rate, and the number describes the cheap half of the work. The documents that pass cost almost nothing; the missing page, the figure that contradicts the purchase order, the certificate that expired last month and the scan that arrived at an angle consume the staff, set the cycle time and produce the complaint — and none of them are extraction problems. They are workflows that have to leave the building, ask a person outside your company for something specific, and come back. What an exception actually is once you classify it by who can resolve it rather than by error type, the six steps that close one, why the explanation sent to a counterparty has to be derived from the check that failed rather than generated afterwards, what authority an agent needs before it is allowed to contact anyone at all, the four ways the loop goes wrong — including closing by timeout and measuring exceptions closed — and what the record has to contain when the other side keeps their own. 15 min ](blog-exception-automation-idp.html) [ Regulation Aug 17, 2026 ### Your policy now excludes the thing you automated Coverage does not get withdrawn with an announcement. It gets withdrawn with an endorsement attached to a renewal, in wording naming artificial intelligence or algorithmic decision-making that nobody in engineering will read. Document operations sit exactly where it bites: an extraction error is not obviously a professional service failure, not a security incident and not a defective product, so it falls between the three policies that would each have absorbed it. Worse, the answer given on the insurance application — that a person reviews the output — starts behaving less like a description of the pipeline and more like a condition of the cover, tested against what the system did rather than what the diagram said. What the three shapes of AI wording actually do, where the error lands across each line of cover, the engineering questions underwriters have started asking, the four artifacts that answer them — versioned thresholds, abstention as a recorded outcome, approval bound to consequence, and a log that reconstructs one decision from 14 months ago — plus the four ways teams find the gap only after a claim, and the cost stated honestly. 15 min ](blog-ai-insurance-exclusions.html) [ Engineering Aug 16, 2026 ### Your extractor cannot say I don't know Every improvement in document understanding made the same failure quieter. A model that reads the page well returns a total in the right format, in the right currency, plausibly close to the line items above it — and sometimes that number is not on the page at all. Nothing downstream can tell, because the output is well-formed either way. Hallucination is not one defect but five: fabrication, misattribution, normalization drift, completion of absence, and inherited error, each caught by a different control and none of them by a firmer prompt. The six layers that work — source binding for every field, a schema that permits absence, deterministic validation that cannot itself hallucinate, abstention as a real output, thresholds priced per field rather than per platform, and an explanation log of why each value was released. Plus what confidence actually measures, four ways the architecture gets built wrong, and the straight-through rate you give up on purpose. 14 min ](blog-anti-hallucination-document-ai.html) [ Regulation Aug 15, 2026 ### The dossier cannot be written afterwards High-risk obligations became enforceable on August 2, 2026, and the change that matters is sequencing rather than substance: technical documentation, conformity assessment and registration are preconditions of go-live, not artifacts assembled for the first audit. For a workflow reading payslips into a credit decision or bank statements into an affordability check, the paperwork is now the gate. What the dossier has to establish — intended purpose stated narrowly rather than as a product description, data governance with real lineage, accuracy per document type and per field instead of one platform number, human oversight built as a capability, a risk-management process that is running rather than filed, and records that reconstruct a past output. Why four of the eight sections depend on measurements that cannot be produced retroactively, which artifacts to demand during the proof of concept, where the provider and deployer split runs, and the overhead stated honestly — including when it buys nothing. 14 min ](blog-ai-technical-documentation-dossier.html) [ Engineering Aug 14, 2026 ### Your pipeline has no control plane Reading got good enough that the model stopped deciding outcomes. What decides them now is everything wrapped around the read — which model version runs for which document type, which schema the output must satisfy, which queue the job enters, what a tenant may spend, how long artifacts are kept, where the webhook goes and how often it retries, what escalates to a person. Each of those is a policy decision, and in most stacks each one lives somewhere different: a config file, an environment variable, a dashboard toggle, a constant compiled into a worker. Nobody can answer "why was this document processed that way" without reading code. The nine registries a control plane holds, why policy resolves along two axes — tenant and document type — rather than one, the four ways the plane gets built wrong, and the engineering overhead it costs stated honestly, including when it buys nothing. 14 min ](blog-document-ai-control-plane.html) [ Engineering Aug 13, 2026 ### Reading a document is a loop Classical OCR is a forward pass: pixels in, characters out, a confidence score attached, and the result handed downstream whether or not it makes sense — a total that contradicts its line items, a column read one position off, a table that lost its header at the page break. None of those defects lowers character confidence, which is exactly why a threshold never catches them. Agentic OCR changes the category rather than the accuracy number: reading becomes a loop that perceives the page region by region, parses its structure, grounds every value in the region it came from, validates the result against what the document itself asserts, and corrects the named failure or escalates it with the contradiction attached. The five stages and why their order is not decorative, the four rules that keep the loop terminating — budget it, correct narrowly, require evidence for every revision, instrument every iteration — the four ways it goes wrong, and the cost-versus-defensibility trade-off stated honestly. 13 min ](blog-agentic-ocr-loop.html) [ Product Jul 3, 2026 ### Catch the bad scan while it's still fixable A person photographs their ID with a thumb over the corner, uploads it, and moves on — and three days later the pipeline bounces it back: image too dark, re-submit. By then most of those people are gone. The document was bad the instant it was captured, but nobody checked at the instant that could have fixed it: while the camera was still open and the person was still there. This is validation as a property of the capture moment, not a downstream checkpoint — the three checks that belong at capture (is the image usable, is it the right document, does it belong to the right person), how to build them without wrecking the experience, the specific re-capture prompt that turns a rejection into a two-second fix, thresholds set by risk rather than vanity, the four ways it goes wrong, and the friction-at-capture-versus-failure-downstream trade-off stated honestly. 12 min ](blog-capture-time-validation.html) [ Product Jul 2, 2026 ### Your newest hire isn't human In 2026 the digital worker stopped being a metaphor — large firms run tens of thousands of humans alongside tens of thousands of agents, the HCM platforms added digital-worker management, and finance teams swapped headcount for compute-and-output as the unit they plan against. A digital worker owns a process end to end under bounded autonomy, as a member of a team, not a script someone triggers. That is a different thing to govern than agent sprawl (dormant privilege) and different from attributing cost per successful task (FinOps): the axis here is the workforce itself — how you provision, onboard, supervise and measure a hybrid team, and who becomes the system of record for the worker that never had an HR file. What a digital worker actually is, the three lenses people confuse, the record nobody assigned, how to onboard an agent like you hire someone, the metric that retires headcount, the four ways it goes wrong, and the leverage-versus-governance-debt trade-off stated honestly. 13 min ](blog-digital-labor-workforce.html) [ Product Jul 1, 2026 ### Your AI vendor rents its compute The continuity risk that reached the board in 2026 isn't where your SaaS vendor is headquartered — it's whose balance sheet the inference actually runs on. A wave of GPU-native providers, the so-called neoclouds, grew at double- and triple-digit rates and took on tens of billions in forward capacity commitments, financed largely by debt that bets on continued refinancing, with revenue concentrated in a few anchor customers and operations younger than the balance sheet. When demand cools, that structure consolidates fast — and the vendor that anchored its stack on a fragile provider inherits the default risk, the SLA degradation and the forced migration. Distinct from the grid-capacity ceiling and the cost of energy per decision, this is the credit and continuity risk of the compute provider itself: what a neocloud is and why it's fragile, why it reaches your SLA, the diligence frame that treats compute as a supply input, the four ways buyers get surprised, and the cost-versus-continuity trade-off stated honestly. 13 min ](blog-compute-provider-risk.html) [ Regulation Jun 30, 2026 ### The agent became a regulated entity In 2026 agentic AI stopped being filed under "software that has AI in it" and started getting rules of its own: jurisdictions began publishing agent-specific regimes, some carrying a mandate for third-party security assessment of covered systems, while analysts put a large share of enterprise applications on a path to embedded agents within the year. Adoption ran well ahead of governance — most organisations already run agents in production, a minority say their oversight keeps pace — and that gap is why the market started repeating that agent governance is the new cybersecurity. Distinct from agent sprawl (the dormant-privilege liability), from red teaming (a testing technique) and from agent identity (who the agent is), this is the regime itself: the agent as an entity a board owns and an assessor asks about. The framework forming — a business owner, technical steward and risk sponsor per agent; an inventory of purpose, scope and review date; a decision trail that survives an assessor; a tested intervention path — the four ways it goes wrong, and the trade-off stated honestly. 13 min ](blog-agent-governance-regulatory-category.html) [ Product Jun 29, 2026 ### The ceiling isn't price — it's the grid The supply risk that reached the board in 2026 is not the price of a token — it is whether there is enough power to produce it. Grid interconnection queues, transmission capacity and transformer lead times became the bottleneck that sets the maximum size of an AI data centre, and the largest operators spent the year locking continuous power under long-term nuclear contracts, including direct reactor allocation and small modular reactors, because firm electricity turned scarce and contested. This is energy supply, distinct from cost per decision (the sustainability line in the RFP): your SaaS vendor depends on frontier inference, which depends on compute capacity, which depends on energy that is physically limited — and that chain is now a risk to your SLA, roadmap and continuity. The due-diligence frame that treats power as a supply input — map where inference runs, place workload by energy availability, reserve capacity in the contract, move the sensitive and steady to the edge — the four ways it goes wrong, and the capacity-versus-lock-in trade-off stated honestly. 13 min ](blog-energy-capacity-ceiling-ai.html) [ Product Jun 28, 2026 ### The power bill behind the token The AI cost that bit in 2026 was not the price of the token — it was the energy bill underneath it. Inference quietly overtook training as the dominant energy draw in a model's lifecycle (roughly two-thirds of frontier energy now goes to the running system, not the build), data-centre power demand grew double digits a year, and electricity prices climbed where compute concentrated — putting power per query inside the unit economics of every high-volume document pipeline. Meanwhile the efficiency frontier moved the other way: inference-optimised chips, mixture-of-experts, KV-cache reuse and liquid cooling cut energy per query even as capacity doubled. This is cost per decision, not grid supply. The FinOps-of-energy frame — energy per decision on the dashboard, placement by intensity times SLA times jurisdiction, right-sizing the model per call — the four ways it goes wrong, and why sustainability disclosure became an RFP asset instead of an ESG footnote. 13 min ](blog-inference-energy-economics.html) [ Regulation Jun 27, 2026 ### Encrypted today, readable tomorrow The cryptographic clock stopped being academic in 2026: NIST's post-quantum standards are final and transition guidance now deprecates RSA and elliptic-curve cryptography toward 2030. But the threat that removes the option to wait is "harvest now, decrypt later" — an adversary capturing encrypted documents today to decrypt when the machine matures — which puts exactly the long-retention document (contract, medical record, tax filing, legal exhibit) at immediate risk. The exposure is double: confidentiality (what is encrypted today can be read tomorrow) and integrity (an ECC signature or timestamp loses future validity, threatening the non-repudiation your audit trail depends on). This is algorithm migration, not content provenance. The framework forming — a cryptographic inventory, crypto-agility, hybrid signatures, re-sealing the archive, and triage by sensitivity times retention — the four ways it goes wrong, and the trade-off stated honestly. 13 min ](blog-post-quantum-document-migration.html) [ Regulation Jun 26, 2026 ### The agent nobody decommissioned Creating an agent got trivial in 2026; retiring one did not — and projections put the average large enterprise on a path to more than 150,000 agents by 2028, while only a minority say they have the governance to match. The agent nobody aged out does not disappear: it keeps its API key, cached token, memory, embeddings and integrations, becoming a dormant identity with live privilege nobody is watching. This is agent lifecycle, distinct from machine-identity security (is the credential a breach vector) and observability (what is it doing now): does every agent have an owner, a review date and a decommissioning gate. A central registry so the fleet is visible, a named sponsor and review cadence, a gate that revokes credential, memory, embedding and integration at once, and orphan detection. The four ways it goes wrong, and the trade-off stated honestly. 13 min ](blog-agent-lifecycle-governance.html) [ Regulation Jun 25, 2026 ### When the agent acts, who answers? The risk bill hardened in 2026 at the exact point the agent started acting alone: insurers began excluding AI from general liability by default — CGL endorsements that surface quietly at renewal — opening a coverage gap precisely when the agent approves a payment, decides, and writes to a system of record. The market answered with standalone AI-liability policies, and governance became an underwriting requirement: the company with a documented programme gets cover and a limit the ungoverned one cannot, and D&O now reads a board's AI-governance maturity. This is risk transfer, distinct from agent identity and observability. The framework forming — allocation named in the MSA, governance evidence that unlocks the limit, a signed trail that makes the claim payable, runtime guardrails sized to the loss — the four ways it goes wrong, and the trade-off stated honestly. 13 min ](blog-agent-liability-insurance.html) [ Regulation Jun 24, 2026 ### Deleting the file isn't forgetting The requirement that walked out of the research lab and into the AI vendor contract in 2026 is blunt: erasing the source document does not remove the influence that data left behind. Once information has passed through a system it lives in more than the file store — in fine-tuned weights, an indexed embedding, a cache, a log — and the right of erasure has stopped accepting "we deleted the record" as proof. Machine unlearning is the emerging discipline that removes a specific data point's influence without retraining from scratch, and analysts now project the auditable "unlearn API" as a platform table-stakes item. Why "we deleted the file" stopped being an answer, why unlearning is neither privacy-preserving AI nor deleting the index, the architecture that makes forgetting provable, the four ways it goes wrong, and the trade-off stated honestly. 13 min ](blog-machine-unlearning.html) [ Regulation Jun 23, 2026 ### Transparency is a feature, not a footnote The disclosure obligation that turned into a product requirement in 2026 catches teams off guard: the newest transparency rules — most visibly the EU AI Act transparency track, landing this year — no longer accept a paragraph in the terms of service. They require the product to tell a person when they are interacting with an AI system, and to mark, machine-readably, that content was generated or altered by AI. This is not cryptographic provenance and not fraud detection on the way in; it is the obligation to label your own output. Why the disclaimer stopped being enough, why labeling your output differs from catching someone else's fake, the architecture that makes the mark survive export, the four ways it goes wrong, and the trade-off stated honestly. 13 min ](blog-mandatory-ai-transparency.html) [ Regulation Jun 22, 2026 ### Prove your controls every single day The governance shift that hardened in 2026 is quiet but total: the regulator and the enterprise buyer stopped accepting the snapshot. The annual audit proves you were compliant for one instant and says nothing about the other 364 days — a gap that stopped being tolerable the moment AI started making consequential decisions in production. Continuous controls monitoring (CCM) is the architectural answer: instrument each control, pull the evidence automatically from infrastructure, identity, HR systems and the pipeline, and keep the posture current and provable every day. Analyses this year put the manual-work saving as high as 90% and the payback at 8 to 18 months. Why the point-in-time audit broke, how CCM differs from a GRC checklist, the architecture that makes continuous evidence hold, the four ways it goes wrong, and the trade-off stated honestly. 13 min ](blog-continuous-controls-monitoring.html) [ Regulation Jun 21, 2026 ### Keep the sensitive document on the device The architecture that gained real traction in 2026 for the most sensitive documents is the one where the data never leaves the perimeter: run extraction and classification on-device or on an edge node, with compact quantized models doing the work locally. AI PCs and edge servers made local inference routine, and a growing share of CIOs now route some inference to the edge under privacy pressure in healthcare, finance and legal. The pay-off is threefold — residency taken to its limit (zero egress, not just "the right region"), latency without a cloud round-trip, and a much lower unit cost on predictable classify-and-extract load. Why the document stopped going to the cloud, how the tier differs from sovereign-cloud and air-gapped, the hybrid pipeline and model fleet that make it hold, the four ways it goes wrong, and the trade-off — a slice of raw capability on the hard cases — stated honestly. 13 min ](blog-on-device-edge-idp.html) [ Regulation Jun 20, 2026 ### Now the RFP asks if you're certified For most of the last two years, "are you certified in AI management?" was a question no procurement team knew to ask. In 2026 it moved onto the form. ISO/IEC 42001 — the first international standard for an AI management system, a Plan-Do-Check-Act discipline for organizations that build, deploy or use AI — went from a governance nicety to a line item, appearing in roughly 40% of enterprise AI vendor questionnaires in Europe and near 25% in North America, with Fortune 500 buyers writing "certified, or a dated roadmap to it" straight into due diligence. The reason is not a fashionable logo — it is that the certificate answers a question self-attestation cannot: how do you govern the AI lifecycle, and who checked? What a 42001 certificate actually claims across the model lifecycle, why scope and operating evidence separate a real certification from a wall decoration, the framework that clears the procurement gate, the four ways the certificate goes wrong, the cost-vs-gate trade-off stated honestly, and why a third party's signature is the new answer to a buyer who stopped accepting self-attestation about AI governance. 13 min ](blog-iso-42001-procurement-baseline.html) [ Regulation Jun 19, 2026 ### Procurement wants your bill of materials For most of the last decade the AI Bill of Materials was a security nice-to-have — a diagram someone drew for an architecture review and never updated. In 2026 it became a line in the RFP. Regulated buyers stopped accepting "we follow best practices" as an answer to "what is your AI system made of," and started asking for a structured, machine-readable inventory of every component: the models, the training and reference data, the software dependencies, the frameworks, the hardware, the processing pipelines and the governance metadata around all of it. The driver is not fashion — the EU AI Act's Annex IV technical-documentation requirement lands on August 2, 2026, formats like CycloneDX ML-BOM and the SPDX 3.0 AI Profile matured enough to be named in contracts, and auditors converged on three questions a slide deck cannot answer: does the inventory exist, is it current, and can each component be traced to its source? What the AI-BOM inventories across six areas, why provenance on data and third-party models is the part that survives scrutiny, the generate-in-CI-not-draft-at-deal-time framework, the four ways the inventory goes wrong, the build-early-vs-lose-the-gate trade-off stated honestly, and why the parts list is the new answer to a buyer who stopped accepting self-attestation. 13 min ](blog-ai-bill-of-materials.html) [ Regulation Jun 18, 2026 ### One decision, fifty state clocks The EU AI Act gave a vendor one thing the United States refused to: a single clock. More than 1,200 AI-related bills were introduced across the states in one legislative cycle, and while Congress debated a federal standard that never arrived, Colorado, Texas, California, Illinois and New York each passed their own requirements, deadlines and penalties — overlapping in intent, conflicting in every operational detail. There is no Annex III to read, because there is no single Act. The exposure doesn't follow the vendor's address or the customer's; it follows the person the automated decision is about, so a document AI pipeline that screens an application inherits the regime of the state where that applicant lives, and one enterprise customer serving all fifty states exports the same decision into fifty regimes at once. The distinction from the single clock and from data residency, what Colorado's duty of care and Texas's NIST-RMF affirmative defense actually turn on, why you classify by use and jurisdiction rather than technology, the framework that answers many regulators with one body of evidence, the four ways the mosaic catches vendors out, the build-for-fifty-vs-exposed-in-two trade-off stated honestly, and why the durable answer is architecture on the product clock — not a policy slide filed the week an attorney general writes. 13 min ](blog-us-ai-state-law-mosaic.html) [ Regulation Jun 17, 2026 ### The deadline that classifies your pipeline On August 2, 2026 the EU AI Act's high-risk obligations become enforceable, and the reaction we hear most on sales calls — "that's a European problem, we're not in Europe" — stops being useful. The Act's scope clause is written around where the output is used, not where the vendor is incorporated: an extraction that feeds a hiring decision, a credit line or a benefit determination for a person in the EU is in scope, through the customer if not directly. The regime classifies the use, not the technology, so the same model is minimal-risk reading a receipt and high-risk when the field it extracts feeds a creditworthiness assessment under Annex III. The obligations are concrete — risk management, data governance with lineage, technical documentation, human oversight as architecture, a per-decision audit trail, conformity assessment and CE marking — and the penalty ceiling reaches €35M or 7% of global turnover for the worst violations. The framework that survives an assessment, the four ways the deadline catches vendors out, the early-compliance-vs-market-exclusion trade-off stated honestly, and why the high-risk deadline is a product decision made quarters ahead, not a legal task done the week before. 13 min ](blog-eu-ai-act-high-risk-deadline.html) [ Product Jun 16, 2026 ### The seat stopped measuring the work Software priced itself by the seat for twenty years because headcount tracked value — until the agent broke the proxy. When one operator does the work of five, seat count stops rising with the value delivered, and the agent itself never takes a seat: it runs as a service, does the volume of ten analysts, and shows up on the license report as one seat or none. The early-2026 software selloff repriced the sector as investors worked out that seats and work had come apart. What replaced per-seat isn't pure pay-per-outcome but a hybrid — a predictable base plus a variable component tied to the result — which for a document-AI vendor changes the billable unit from users and pages to decisions, cases and processed documents. The emerging frame (a measurable billable unit, a per-outcome margin guardrail, base-plus-variable, outcome telemetry, a seat-to-outcome migration), the four ways outcome pricing goes wrong, the alignment-vs-predictability trade-off stated honestly, and why charging for the result without knowing what it costs trades predictable revenue for variable loss. 13 min ](blog-death-of-per-seat-pricing.html) [ Product Jun 15, 2026 ### The data changed. Nobody told the agent Data quality is the most-cited blocker to agentic AI in 2026 — around 42% of organisations name it the number-one obstacle — but most teams fight the symptom with quality dashboards and cleanup pipelines instead of the root cause: there's no formal contract between the team that produces a dataset and the agent that consumes it. A column gets renamed, a field arrives late, an enum gains a value, and the agent decides on the new shape with logic built for the old one. The layer that fixes it is the data contract — a strictly versioned schema with freshness and completeness SLAs, a breaking-change policy, and a defined fallback on violation. The frame component by component (schema registry with semver, formal owner per dataset, deprecation policy, in-pipeline validator, broken-contract alert, observability), the four ways contracts go wrong, the velocity-vs-safety trade-off stated honestly, and why the team that skips this layer ships an agent whose reliability is hostage to every upstream sprint it never sees. 13 min ](blog-data-contracts-agentic-foundation.html) [ Product Jun 14, 2026 ### Cheap per run. Expensive per result The unit economics question the board started asking in 2026 isn't what the agent costs per token or per run — it's what a completed outcome costs, adjusted for how often the agent actually completes it. An agent that's cheap per execution and fails 30% of the time has worse unit economics than one that costs 4x more and succeeds 95% of the time, because every failure carries human rework, retries and eroded trust the inference bill never shows. The frame that replaced the token report: cost-per-successful-task with success — not execution — in the denominator, a success-rate multiplier that prices rework and escalation into the unit cost, agent FinOps that sums inference plus tools plus review plus the cost of error, and an honest human baseline per task — plus the four ways the token metric sinks the board conversation, the cheaper-per-run vs cheaper-per-result trade-off stated honestly, and why "we spent X on tokens" is the sentence that loses the meeting. 13 min ](blog-cost-per-successful-task.html) [ Product Jun 13, 2026 ### You didn't buy an agent. You bought the demo The number that repositioned agentic AI in 2026 isn't about model capability — it's about production reliability: fewer than 15% of organisations have scaled an agent to broad use, and the rest sit in pilot purgatory with agents that shine in the demo and fail on dirty input, uncooperative users and off-script scenarios. The bottleneck has almost nothing to do with model quality and everything to do with the gap between capability — the happy path — and reliability — the real case, repeatedly, under exception. The emerging discipline that closes it: success-rate SLOs per intent measured in production, continuous evaluation plus action observability plus runtime guardrails as one stack, a maturity model that gates autonomy on the numbers, deterministic fallback and human review designed in rather than patched on — plus the four ways the gap swallows a programme, the honest launch-speed trade-off, and why "which model is best" stopped being the question that matters. 13 min ](blog-agent-reliability-gap.html) [ Product Jun 12, 2026 ### Your customers are training someone's moat The moat story of 2024–2025 — "we fine-tuned our own model" — quietly expired in 2026, when frontier inference prices collapsed and open-weight models reached parity on domain extraction tasks. What stayed defensible is the data flywheel: the pipeline that turns every correction a customer makes in production — a wrong field, a reverted decision, a reclassified exception — into a retrained validator, an adjusted threshold, a refined prompt or an expanded schema. Mature flywheels report human review dropping 40–60% within 12 months without changing the base model, and a competitor entering the same account 6 months later starts behind on a curve it cannot buy. The five-component anatomy — correction telemetry, governed ingest, targeted updates, production A/B sandbox, auto-resolution rate as the KPI — the four ways the flywheel breaks, the data-rights trade-off, and why "which model do you use" is a commodity question while "how fast does your system learn my domain" decides the deal. 13 min ](blog-data-flywheel-moat.html) [ Product Jun 11, 2026 ### Your vendor's exit is your problem now In roughly twelve months document AI went from a crowded field to an active M&A board: RPA majors absorbed document-compliance vendors that had raised hundreds of millions, European ECM groups bought tax-focused OCR houses, data platforms acquired agentic capabilities — while mid-stage funding concentrated in vertical specialists. The thesis is blunt: by Q4 2026, a horizontal IDP vendor under $50M ARR with no vertical anchor and no platform partnership is a cut candidate. For the buyer that lands in three places at once — vendor-viability due diligence became strategy, the continuity clause became contract baseline, and the vertical platform with a real agentic engine earned a premium over the horizontal toolkit. The consolidator–target–survivor read on every profile, the four ways buyers and founders misread the wave, the honest trade-off, and why "what is your plan if you are acquired" now belongs in the RFP one line under "what is your SLA". 13 min ](blog-document-ai-ma-wave.html) [ Product Jun 10, 2026 ### Sixty tools, half unused. Five are about to go The market thesis of 2026 put a nine-figure margin pool inside one move — converting human labour into agentic software — and the side effect is vendor consolidation. The average enterprise runs 60 to 80 active SaaS tools with under half the licenses used, and once a category becomes an agent that decides and acts rather than a feature that shows a result, the buyer stops wanting five point tools and starts wanting one platform that extracts, decides, writes back and orchestrates. The hard implication: the vendor of one isolated feature is the obvious cut at the next renewal, while the platform that closes the loop captures the budget of three or four tools at once. Why the point tool goes first, the platform-fit gap that sinks the pretender — a pretty UI over a weak decision engine — the frame for what survives consolidation, the four ways a vendor loses the renewal it thought was safe, the honest trade-off, and why the buyer's RFP stopped asking what feature you have and started asking what workflow you close end to end. 13 min ](blog-agentic-vendor-consolidation.html) [ Product Jun 09, 2026 ### Your buyer stopped searching. They asked the model The quiet demand-gen turn of 2026: the B2B buyer starts vendor research inside an answer engine, not a search box. Close to half of software buyers begin in an AI assistant, the visitors those answers refer convert well above organic, and traditional search volume is projected to fall by double digits. Generative engine optimization is the discipline of structuring content and data so the model cites and recommends you in its answer — and because no platform sells paid placement inside a generated answer, the citation is earned by citable fact, retrievable structure and off-site trust. The break from SEO is hard: instead of ten ranked positions you compete to be one of the two to seven sources the model names in a single answer. For a document AI vendor it's a funnel one — if the buyer's agent doesn't cite you, you're not on the shortlist and the demo never happens. The frame for being citable, the four ways teams get GEO wrong, the honest trade-off, and why a team that still optimises only to rank loses the top of the funnel to the one worth quoting. 13 min ](blog-generative-engine-optimization.html) [ Product Jun 08, 2026 ### Stop charging for the page. Charge for the decision The quiet commercial turn in document AI in 2026 follows the turn in B2B SaaS as a whole: seat pricing fell from around a fifth of the market to about a seventh in twelve months, and pricing moved toward consumption and outcome because the agent broke the per-seat equation — one human plus three agents doesn't consume three seats, it consumes work. In document AI the per-page price repeats the same error: it charges for the input, the volume processed, when the mature buyer only wants to pay for the result — a decision approved with no rework, an invoice reconciled, a supplier cleared, a claim settled. How to define a billable outcome unit, how to keep vendor and customer incentives aligned so more automation means more revenue not less, the price floor and collar that protect margin when the hard-document mix rises, the per-outcome telemetry that makes the bill auditable, the four ways outcome pricing goes wrong, the honest trade-off between recurring-revenue predictability and value alignment, and why a vendor still billing per page loses the economic buyer who stopped buying throughput and started buying a measurable result per workflow. 13 min ](blog-outcome-based-pricing-idp.html) [ Product Jun 07, 2026 ### Growth bought the round. Margin keeps it B2B SaaS spent 2026 trading growth-at-any-cost for margin discipline. The median software business now grows in the low teens at a low-twenties operating margin — figures that read as weakness in 2021 and as health once cheap capital stopped subsidising unprofitable growth. Capital didn't leave the category; it moved toward retention, efficient acquisition and a gross margin the business can hold, and the vendor still pitching feature breadth loses the deal to the one that can show its unit economics. The metrics a cap-table buyer now checks before the term sheet — NDR above 110%, payback under 18 months, magic number above 1, customer concentration under 20% — how the discipline rewrites hiring toward solution and customer engineering, the four ways the growth-first reflex still sinks the model, the honest trade-off between speed and durability, and why software anchored to paid specialist work is the most defensible thesis left when money is no longer free. 13 min ](blog-saas-margin-over-growth.html) [ Product Jun 06, 2026 ### It's software. The margin disagrees The number that reframed the 2026 board conversation about AI wasn't growth or retention — it was gross margin. Agentic software carries a token tax classic SaaS never had: inference lands directly in COGS and pulls gross margin down to the 50–60% band against the 75–85% of traditional SaaS, and reasoning models inside agentic loops multiply the consumption at every level of autonomy. The implication is a valuation one, not just an operating one — a revenue multiple calibrated for SaaS margin, applied to a business with an agent's COGS, overprices the asset. Why inference is first-order COGS, where the token tax actually comes from, the levers that recover margin in points rather than vibes, why "margin per decision" belongs next to "cost per decision", the four ways the tax sinks a business, the honest trade-off between autonomy and the margin floor, and why "it's software, so 80% margin" loses credibility on the first results cycle. 13 min ](blog-token-tax-agentic-margin.html) [ Product Jun 05, 2026 ### The agent is ready. The data isn't The most-repeated readiness number of 2026 keeps landing on the same verdict: only about 15% of enterprises can sustain agentic AI in production — yet 41% already run agents and roughly 60% have spent millions getting there. Read together, the figures say most of the money is sitting on a foundation that can't hold it. The barriers that dominate aren't the model: data quality and lineage (42%), compliance and sovereignty (39%), security and privacy (39%). The five layers of a foundation an agent can act on — always-on pipelines, end-to-end lineage, data contracts, interoperable governance, per-field provenance — the four ways the foundation quietly fails, the honest trade-off between waiting for a clean foundation and shipping on the one you have, and why a board grading maturity by agent count is reading the wrong number. 13 min ](blog-agentic-ai-readiness-data-foundation.html) [ Product Jun 04, 2026 ### A real agent, or RPA in a new coat A forecast reorganised the year: more than 40% of agentic projects will be cancelled by the end of 2027 — for runaway cost, diffuse value, or inadequate risk control — and the aggravating factor has a name, "agent washing", rebranding an assistant, a chatbot or an RPA bot as an "agent" with no real autonomy. For the board, what separates the project that survives from the write-off isn't the demo — it's measurable autonomy, ROI attributed per flow, and risk control proportional to blast radius. The autonomy test that breaks the happy path on purpose, ROI per flow as the continuity gate, controls sized to what the agent can touch, the total cost that includes the reasoning loop, the four ways a board ends up in the 40%, and why the document desk is the cleanest place to run the test. 13 min ](blog-agent-washing-real-agents.html) [ Product Jun 04, 2026 ### Build the moat, partner for the plumbing One figure from the last year cut through a lot of internal pride: programs run with a specialised partner reach production at roughly twice the rate of those built purely in-house — about two thirds versus one third. The naïve reading is "outsource it", and that's wrong; so is "we build everything". The build-vs-partner call decided layer by layer — rent the model, partner for orchestration and evaluation, own the domain data and the corrected flow — the portability clause that keeps a head-start from becoming lock-in, who should own the customer's correction, the four ways a partner program quietly fails, the honest trade-off between speed now and margin later, and why the board metric that predicts who ships isn't ML headcount. 13 min ](blog-build-vs-partner-ai.html) [ Product Jun 04, 2026 ### The gap isn't the model. It's the instrumentation Roughly 95% of GenAI pilots deliver no measurable P&L impact, almost everyone has a pilot and almost nobody has a fleet in production. The cause isn't model quality and it isn't regulation — it's the three layers underneath the technology that never got built: measurement that proves the task works, integration that wires it into the flow that creates value, and ownership that keeps the system learning after go-live. The three layers in detail, the build-vs-partner signal that roughly doubles the production rate, the four ways the evaluation layer quietly rots, the honest trade-off of instrumenting before you see a number, and why "which model" is the wrong opening question. 13 min ](blog-pilot-production-gap.html) [ Product Jun 03, 2026 ### Templates or zero-shot, decided by the tail Two honest ways to get fields out of a document, and most of the noise comes from pretending one has to lose. Template-based extraction is deterministic, fast and near-free per page — and breaks the instant a layout drifts. Zero-shot absorbs the long tail of formats you'll never template, at a per-call cost and with quiet, plausible failures. The eight dimensions a buyer actually weighs, where each genuinely wins, the hybrid that ships — route by class and confidence, zero-shot as cold-start and fallback, templates as a cache of what zero-shot learned — and the failure mode that bites each when nobody's watching. 12 min ](blog-template-vs-zero-shot-extraction.html) [ Product Jun 03, 2026 ### Upskilling at the speed of automation In 12 months agentic-AI upskilling went from an optional internal program to a board investment — consultancies certifying tens of thousands of professionals in a single deal, banks certifying whole operations functions. The dual thesis (operating agentic AI needs a different skill stack; the firm that certifies at scale first captures the talent supply and builds a switching cost), the certification-as-moat case and exactly where it rots into paper, a competence matrix by role that survives an audit, the four ways an upskilling program quietly fails, and why the document desk is the cleanest place to make the certificate predict the work. 13 min ](blog-agentic-upskilling-certification.html) [ Product Jun 02, 2026 ### Intelligence per worker, measured honestly Frontier firms run ~3.5x more intelligence per worker than the median, and in 2026 the metric jumped from HR slide to board agenda — arriving in the same news cycle as the AI-restructuring layoffs. It is a real measurement and an easy euphemism, one keystroke apart. The four components that make it auditable instead of anecdotal — output per FTE-equivalent, delegable-task automation rate, time recovered with a destination, quality with vs without the agent — why measuring only the numerator books six months of gain against a year of cost, and where the document desk measures it honestly first. 13 min ](blog-intelligence-per-worker.html) [ Engineering May 24, 2026 ### Always-on agents and the back-office that stops opening tickets 2026 shipped a category that 2025 treated as roadmap: agents that run continuously, watch a queue, decide and execute without a human trigger. The five back-office workflows that absorbed the pattern first, the four design patterns that consolidated — subscription / idle, per-agent budget, kill switch and dead-man's switch, ground-truth-of-when-to-stop — and the cost trade-off that survives a CFO's diligence. 13 min ](blog-always-on-agents-workflow-design.html) [ Product May 23, 2026 ### Capability contracts for the post-app era The 2026 device-and-OS shift moved the buyer one layer up: the assistant in the operating system, the notebook and the phone decides which portal to call. The three consequences for SaaS in the next two quarters, the four-artefact capability contract buyers are asking for by name — capability manifest, intent registry, tool catalog with permission scope, per-agent telemetry — and the anti-patterns that quietly retire a portal. 12 min ](blog-post-app-ai-first-interfaces.html) [ Regulation May 22, 2026 ### Regulator-in-the-loop: pre-launch model access as the 2026 baseline Frontier-model providers in the US, UK and EU now hand pre-release weights to government safety institutes before public availability. The five RFP questions enterprise procurement put on the table in 2026, the model-card fields buyers check, the rollback architecture that holds when a regulator pulls a version — and the three anti-patterns that disqualify in regulated diligence. 13 min ](blog-regulator-in-the-loop-pre-launch.html) [ Product May 22, 2026 ### When inference owns your margin: compute as the 2026 moat for document AI Record-breaking AI capital — an US$ 122B round at US$ 852B post-money, US$ 200B+ in committed compute, an OCR floor sliding to US$ 2/1,000 pages. The three decisions a CIO and a CFO have to make now, the five-line cost build-up procurement is asking vendors to show, and the two vendor profiles most likely to get squeezed in the next 12–18 months. 12 min ](blog-compute-moat-inference-margin.html) [ Engineering May 21, 2026 ### When reasoning carries a proof: verifiable frontier models for enterprise documents 2026 moved the frontier-model line from "better answers" to "checkable steps". The four enterprise places verifiable reasoning already landed — audit, compliance, legal, regulated extraction — the five-component architecture that ships, the 78–92% auditor-accepted rates in early deployments, and the limits we are not going to pretend away. 12 min ](blog-frontier-reasoning-verifiable-enterprise.html) [ Product May 20, 2026 ### Agent-as-user: the post-app B2B portal A measurable share of B2B portal sessions in 2026 originate from an assistant acting for a logged-in buyer. Six surfaces an agent actually needs, four moves in the playbook that ships, and the anti-patterns that quietly disqualify a portal — read against what changes for the document AI behind it. 11 min ](blog-agent-as-user-portals.html) [ Regulation May 19, 2026 ### Sovereign AI: residency as a contract clause, not a policy slide Sovereign AI consolidated as an enterprise category in 2026, with a 30–60% premium and six clauses regulated buyers now want by name — reserved compute, jurisdiction-bound models, BYOK/BYOM, signed audit trail, kill switch and bordered sub-processors. The architecture that holds and the anti-patterns that fail diligence. 12 min ](blog-sovereign-ai-document-pipelines.html) [ Product May 18, 2026 ### AI operating model: six artifacts, three anti-patterns, 90 days 76% of large orgs hired a CAIO, but the title alone does not survive an RFP. The six artifacts buyers now ask for, the three anti-patterns we keep seeing, and a 90-day build plan that fits inside one quarter. 12 min ](blog-ai-operating-model.html) [ Regulation May 17, 2026 ### AI governance for document AI: ISO 42001, EU AI Act and audit evidence in 2026 Gartner expects 70%+ of enterprises to adopt a formal AI governance standard by end of 2026, and the platforms market goes from US$ 492M to US$ 1B by 2030. The five frameworks, three audit shapes and the evidence envelope that holds. 12 min ](blog-ai-governance-audit-evidence.html) [ Engineering May 16, 2026 ### Generative document fraud: why detection became a board item in 2026 AI-generated document fraud grew nearly 5x in eight months, and 97.8% of risk leaders say it already keeps them up at night. The four shapes hitting the queue, the hybrid stack that holds, and what an investigation agent actually does. 12 min ](blog-generative-fraud-document-ai.html) [ Engineering May 15, 2026 ### Multi-agent systems in the backoffice: practical orchestration in 2026 One agent stops scaling at the third edge case. Squads of specialists coordinated by a maestro hit 35–55% more throughput on complex flows. The three patterns, the five failure modes, and the observability shape that ships. 12 min ](blog-multi-agent-systems-backoffice.html) [ Engineering May 13, 2026 ### VLMs and the document pipeline collapse: when one model replaces four stages Vision Language Models fold OCR, classification, layout analysis and extraction into a single call. When the collapse pays back, where the classical stack still wins, and the hybrid routing most production programs actually ship. 11 min ](blog-vlm-architecture-document-pipeline.html) [ Engineering May 12, 2026 ### From reactive to predictive: when document AI starts forecasting the work The next IDP shift isn't faster extraction — it's pipelines that forecast backlog, flag the contract clauses legal will redline, and order the queue by expected value. Three workflows rewritten and the parts that quietly break. 11 min ](blog-reactive-to-predictive-document-automation.html) [ Engineering May 11, 2026 ### From extraction to execution: IDP as the engine, not a step Gartner expects 40% of enterprise apps to ship task-specialized agents by end of 2026. What actually changes when IDP stops feeding the workflow and starts running it — three workflows rewritten and the parts that quietly break. 11 min ](blog-extraction-to-execution.html) [ Engineering May 10, 2026 ### Reasoning models: when document AI thinks before it extracts Reasoning models cost 5–10x more per call. They cut human review 50–70% on the right class of documents. What pays back, what doesn't, and the routing pattern that decides whether the program ships. 11 min ](blog-reasoning-models-document-ai.html) [ Engineering May 09, 2026 ### From RPA to agents: the autonomy bar 2026 is asking documents to clear Gartner expects 30% of enterprises to have automated more than half their workflows by end of 2026, up from under 10% in 2023. The architectural difference between RPA and agents — and the four failure modes that come with the upgrade. 11 min ](blog-agentic-ai-documents-2026.html) [ Product May 09, 2026 ### The AI ROI gap: what the 29% do differently Enterprise AI budgets jumped 65% to US$11.6M, but only 29% report meaningful ROI and 42% have abandoned most of their initiatives. Four traits separate programs that pay back from those that quietly stalled. 10 min ](blog-ai-roi-gap-2026.html) [ Engineering May 08, 2026 ### OCR in 2026: from 70% to 98% — and why the pipeline didn't disappear Multimodal LLMs cleared 98% on printed text and 95% on handwriting. The interesting question stopped being whether to adopt them — it became where in the pipeline they actually pay back. 9 min ](blog-ocr-multimodal-2026.html) [ Engineering May 07, 2026 ### From extraction to decision: how IDP stopped copying fields and started thinking 67% of enterprise document programs are evaluating agentic IDP, up from 23% two years ago. What changed in the reference architecture, and the parts that are quietly harder than the slide suggests. 10 min ](blog-idp-extraction-to-decision.html) [ Engineering May 01, 2026 ### Prompt injection in document AI: the threat model nobody scopes Every document your pipeline ingests is untrusted instruction text. The threat model, three real attack patterns, and the four defenses that actually hold. 9 min ](blog-prompt-injection-document-ai.html) [ Engineering Apr 30, 2026 ### SOC 2 Type II for AI startups: what to build in The five architectural commitments that turn the SOC 2 audit from a quarter-long cleanup project into an emergent property of your platform. 9 min ](blog-soc2-ai-startups.html) [ Regulation Apr 29, 2026 ### Sub-processors in AI: what your DPA needs in 2026 AI products invoke 4–7 sub-processors per request. What your DPA needs to say about LLM providers, observability vendors, and zero-retention APIs in 2026. 9 min ](blog-ai-sub-processors.html) [ Regulation Apr 28, 2026 ### GDPR for document AI: a practical guide for operators Article 28 obligations, lawful basis, sub-processor governance, data subject rights, and what your DPA actually needs to say. 9 min ](blog-gdpr-document-ai.html) [ Engineering Apr 26, 2026 ### Tracing agentic document extraction How to make multi-step LLM workflows debuggable. OpenTelemetry span design, sampling strategies, and the structured logs that turn a black box into a flight recorder. 8 min ](blog-tracing-agentic-extraction.html) [ Engineering Apr 25, 2026 ### Audit trails for non-deterministic outputs How to log AI extractions in a way that holds up to reproducibility, regulatory audit, and customer "why did you extract this?" questions — with the actual schema we use at Cogneris. 8 min ](blog-audit-trail-non-deterministic.html) [ Case Apr 03, 2026 ### From 48h to 2.4s: a fintech underwriting case How a fast-growing fintech replaced a team of 12 operators with an API. 5 min ](case-meridian-credit.html) [ Case Apr 02, 2026 ### Prior auth turnaround cut from 5 days to 4 hours How a regional health plan automated PA triage with HIPAA-grade controls. 6 min ](case-northwind-health.html) [ Case Apr 01, 2026 ### Contract intake at 100x volume — same legal team How a US M&A boutique scaled diligence with Cogneris's extraction API. 5 min ](case-cascade-legal.html) ## Source: glossary.html [Skip to content](#main) Glossary # Document AI vocabulary. A practical glossary of intelligent document processing terms — for buyers, builders, and anyone who has to write a vendor evaluation in 2026. Definitions are written to be useful, not to flatter our category. ### IDP Intelligent Document Processing The category of software that ingests unstructured documents and produces structured data, typically combining OCR, classification, extraction, and validation. Twenty years ago this was rule-based template matching. Today it's predominantly LLM-driven, and the category boundary has blurred with general-purpose AI platforms — but the buyer problem (turn this PDF into a row in my database, accurately, auditably) is unchanged. [Read the buyer's guide →](what-is-idp.html) · Related: [OCR](#ocr) · [Schema-driven extraction](#schema-driven-extraction) · [Compare IDP vendors](compare.html) ### OCR Optical Character Recognition The conversion of pixel data (scanned images, photos of documents) into machine-readable text. Foundational input to document AI, increasingly commoditized — modern OCR engines hit 99%+ character accuracy on clean documents, and even mediocre OCR is good enough when the downstream model is an LLM that can correct context. Obsessing over OCR accuracy is the wrong metric in 2026; the differentiator is what you do with the text after. [Read the full page →](term-ocr.html) · Related: [IDP](#idp) · [Schema-driven extraction](#schema-driven-extraction) ### HITL Human-in-the-Loop A workflow pattern where uncertain or low-confidence machine outputs are routed to a human reviewer before downstream use. Done well, HITL is how you get to 99.9% accuracy on documents where 99% would be unacceptable — the human only sees the 1% the model wasn't sure about, and the corrections feed active learning. Done poorly, HITL is a euphemism for "we ship 80% accuracy and call it a feature." [Read the full page →](term-hitl.html) · Related: [Confidence score](#confidence-score) · [Active learning](#active-learning) · [STP](#stp) ### ReAct Reason + Act architecture An agentic LLM pattern where the model alternates between reasoning steps ("the document looks like an invoice; the line items don't add up; I should call the totals tool") and tool-using actions. Distinct from single-pass extraction (one LLM call, hope for the best) and from pure RAG (retrieve documents, generate an answer). ReAct is a natural fit for document workflows where extraction is one step in a longer chain — classify, extract, validate, escalate. Related: [ReAct architecture for documents (deep dive)](pillar-react-architecture.html) ### Schema-driven extraction An extraction approach where the desired output schema (field names, types, validators) is the primary configuration, and the model is instructed to populate that schema rather than train a per-document-type classifier. Faster to iterate than supervised training, and it generalizes to documents you've never seen before. Trades some defensibility for speed: you're auditing the prompt and model version, not a frozen training artifact. Related: [ReAct](#react) · [Audit trail](#audit-trail) ### Audit trail Tamper-evident, time-ordered metadata for supported document operations can include model version, prompt and response hashes, human actions, and lineage. Coverage depends on the service and workflow configuration. [Read the full page →](term-audit-trail.html) · Related: [Audit trails for non-deterministic outputs](blog-audit-trail-non-deterministic.html) · [Trust & security](security.html) ### Confidence score A model-emitted estimate of how likely an extracted value is to be correct. Used to drive HITL routing, auto-approval thresholds, and quality reporting. Worth knowing: LLM confidence scores are notoriously poorly calibrated out of the box. A 0.95 from one model is not the same 0.95 as another. Calibration is a per-platform engineering problem, not a free input. Related: [HITL](#hitl) · [STP](#stp) ### Multi-LLM consensus Running the same extraction request through multiple LLM providers (or multiple models) and reconciling results. Used as an accuracy lever (where two models agree, confidence is high) and as a hedge against single-vendor failure modes (one provider has an outage; another keeps the request flowing). Costs more compute than single-model extraction; worth it on regulated or high-stakes workflows. Related: [Zero-retention API](#zero-retention-api) · [Sub-processor](#sub-processor) ### Zero-retention API An LLM API configuration where the provider commits to not training on, persisting, or logging customer payloads beyond operational requirements (typically 0–30 days for abuse monitoring, with the customer able to opt out of even that on enterprise tiers). Required for regulated workflows — without it, you're sending PHI, PII, or material non-public information to a vendor that retains and possibly trains on it. OpenAI and Anthropic both offer zero-retention configurations on enterprise plans. Related: [Sub-processor](#sub-processor) · [Sub-processors in AI: what your DPA needs in 2026](blog-ai-sub-processors.html) ### Sub-processor Under GDPR Article 28, any third-party processor engaged by a primary processor to handle personal data on behalf of a controller. AI products typically have 4–7 sub-processors per request — LLM provider, vector store, observability vendor, email/SMS sender, identity provider, payments. Your DPA needs to list them, your customer needs to be able to object, and you need a 30-day notice clock when you change them. Related: [DPA](#dpa) · [Sub-processors in AI](blog-ai-sub-processors.html) · [Data protection](data-protection.html) ### DPA Data Processing Agreement The contractual instrument required by GDPR Article 28 between a controller and processor (and, recursively, between processor and sub-processor). Specifies handling, security, breach obligations, sub-processor governance, audit rights, and retention. In 2026 a DPA without an AI-specific addendum is undercooked — the Article 28 obligations apply to LLM-mediated processing the same way they apply to any other. [Read the full page →](term-dpa.html) · Related: [GDPR for document AI](blog-gdpr-document-ai.html) · [Sub-processor](#sub-processor) ### BAA Business Associate Agreement The contractual instrument required by HIPAA between a covered entity (provider, payer, clearinghouse) and a business associate handling protected health information (PHI). Without a BAA, you cannot legally process PHI on behalf of the covered entity. AI vendors that don't offer BAAs are off the table for healthcare workflows by default. Related: [PHI](#phi) · [Healthcare workflows](solutions-healthcare.html) ### PII Personally Identifiable Information Data that can identify a specific individual either directly (name, government ID, biometric) or indirectly (combination of attributes that uniquely identify in context). Subject to data-protection laws across jurisdictions (GDPR, CCPA/CPRA, LGPD, PIPEDA). Document AI workflows almost always touch PII — the controls that matter are encryption, access scoping, retention discipline, and audit lineage. Related: [PHI](#phi) · [DPA](#dpa) ### PHI Protected Health Information Under HIPAA, individually identifiable health information held by covered entities and business associates. Subject to the HIPAA Privacy Rule (use and disclosure) and Security Rule (technical, physical, and administrative safeguards). PHI is a strict subset of PII with sharper handling rules — minimum-necessary access, breach notification thresholds, and BAA-mediated processor relationships. Related: [BAA](#baa) · [PII](#pii) · [Healthcare workflows](solutions-healthcare.html) ### Active learning A training approach where a model selects uncertain or informative examples for human labeling, then incorporates those labels back into the model. In document AI, the loop typically runs: model extracts, low-confidence fields go to HITL review, reviewer corrections are added to a few-shot prompt or fine-tuning dataset. The cycle improves accuracy on edge cases without bulk retraining. Related: [HITL](#hitl) · [Confidence score](#confidence-score) ### STP Straight-Through Processing A workflow pattern where high-confidence machine outputs are auto-approved without human touch. Common in claims, payments, KYC, and AP automation. STP rates (% of documents that bypass human review) are the headline metric most IDP procurement evaluations compare on. Worth pairing with quality-of-bypass metrics: a 95% STP rate at 99% accuracy is good; a 95% STP rate at 90% accuracy is a recall problem dressed as an efficiency win. Related: [HITL](#hitl) · [Confidence score](#confidence-score) ### FNOL First Notice of Loss The initial report of a claim by a policyholder, triggering the claims workflow in insurance. FNOL packages typically arrive as multi-document bundles — claim form, photos, police reports, repair estimates — and the time to triage them is the gating step on every downstream metric (cycle time, loss adjustment expense, customer satisfaction). Related: [Insurance workflows](solutions-insurance.html) · [Insurance claims extraction](extract-insurance-claims.html) ### KYC Know Your Customer The regulatory process of verifying the identity of customers, typically using a combination of identity documents (passport, driver's license, national ID) and supporting records (proof of address, source of funds). Required across financial services, fintech, marketplaces, and increasingly any business handling significant payments. KYC document workflows are bursty, latency-sensitive, and intolerant of errors — three reasons document AI fits well. [Read the full page →](term-kyc.html) · Related: [KYC document extraction](extract-kyc-documents.html) ### Document classification The step of identifying what kind of document is being processed (invoice vs. payslip vs. ID), typically as a precursor to extraction. In modern stacks, classification and extraction can be the same LLM call — the model identifies the type and emits the type-specific schema in one pass — but for high-volume systems, a separate fast classifier still earns its keep on cost and latency. Related: [Schema-driven extraction](#schema-driven-extraction) · [IDP](#idp) ### Field validation Post-extraction checks that confirm extracted values are well-formed, internally consistent, and (where possible) externally verified. Examples: checksum validation on tax IDs and bank routing numbers, date-range sanity (no future invoices), totals reconciliation (line items sum to total), external lookups (NPI valid, address resolves, EIN matches IRS file). Most extraction errors caught here, not in the model. Related: [Schema-driven extraction](#schema-driven-extraction) · [Confidence score](#confidence-score) Term we should add? ### We update this glossary as the category evolves. If a term you care about is missing, or if you think we got a definition wrong, send us a note. We track suggestions and ship updates monthly. [Suggest a term](contact.html) ## Source: release-notes.html [Skip to content](#main) Release notes # Changes you can verify. Approved changes to Cogneris product surfaces, APIs, schemas, models, and documentation, with impact and migration guidance. ## 2026 October 27, 2026 ### Monthly overage cap on Pro and Scale #### Impact Pro and Scale subscriptions now carry a monthly overage cap: 100 in the subscription's currency (€100 in the EU region, $100 in the US region) by default, adjustable by a company admin up to 10,000 in the console. A request that would take the period's overage past the cap is refused with HTTP 402 and the reason `BILLING_OVERAGE_CAP_REACHED`; reaching the cap exactly is allowed. Company admins receive an e-mail at 80% and at 100% of the included credits and of the cap, once per period. #### Migration No change to existing integrations is required. Clients that run high volume on Pro or Scale should handle HTTP 402 with reason `BILLING_OVERAGE_CAP_REACHED` by raising the cap in the console or waiting for the next billing period. #### Compatibility Credit prices, plan allowances and overage rates are unchanged. Free, Unlimited and Enterprise are not affected: Free still stops at zero credits, and the post-paid plans have no cap. #### Deprecations Nothing is deprecated. Uncapped overage on Pro and Scale is replaced by the capped behaviour above. #### Rollout and rollback Shipped in the billing-be release 2026.10.27.001, serving the EU production region since 2026-10-27; the US region is paused. Rollback is a redeploy of the previous billing-be release, which restores uncapped overage; the cap and the alerts are additive and need no data migration back. #### Evidence [Pricing: how credits work](pricing.html#how-credits-work) and the console's Billing screen, which shows the cap and the overage spent in the period. October 23, 2026 ### Document fraud verification routes #### Impact `POST /api/v1/fraud-verify` analyses a document for signs of tampering or reuse and returns a request id; `GET /api/v1/fraud-verify/{requestId}` returns the result. The routes are enabled per tenant on request and cost 8 credits per analysis; the scopes `fraud-verify.write` and `fraud-verify.read` gate the two calls. #### Migration None for existing integrations: the routes are additive. To use them, request enablement for your tenant and create an API key carrying the fraud-verify scopes; a call on a tenant without enablement returns `fraud.not_enabled`. #### Compatibility Existing routes, envelopes and field shapes are unchanged. The published OpenAPI contract moved to version 2026-10-23 to carry the new routes. #### Deprecations Nothing is deprecated. #### Rollout and rollback Shipped in the api-be and billing-be releases 2026.10.27.001 in the EU production region. Rollback is a redeploy of the previous releases: the routes disappear and no stored data needs migration. #### Evidence [OpenAPI 2026-10-23](openapi.yaml), [fraud verification documentation](https://docs.cogneris.ai/docs/api/fraud-verify/) and the [credit table](pricing.html#how-credits-work). October 23, 2026 ### OpenAPI contract 2026-10-23 #### Impact The published OpenAPI document is version 2026-10-23. It adds the fraud verification routes, documents that asynchronous extraction jobs post-process dates, numbers and tables the same way as a synchronous extraction, and announces that API keys created through the API will require an expiry on a date to be published in the documentation release notes. #### Migration None today. To prepare for the key-expiry change, send an expiry and the scopes you need when you create a key through the API; keys created on the Developers screen already require both. #### Compatibility No existing route, envelope or field shape changed. A key created through the API without an expiry still gets full access and never expires until the announced date. #### Deprecations The no-expiry default for API-created keys is scheduled for removal; the date will be announced in the documentation release notes before it takes effect. #### Rollout and rollback The contract ships with the static site and rolls back with the corresponding site revision; the service behind it is the api-be release 2026.10.27.001. #### Evidence [openapi.yaml](openapi.yaml) and the [documentation release notes](https://docs.cogneris.ai/docs/reference/release-notes/). October 20, 2026 ### Developer MCP server in production #### Impact Cogneris runs a Model Context Protocol server at `https://mcp-eu.cogneris.ai/` (stateless streamable HTTP). From VS Code, Cursor, Claude Code or any MCP client, an assistant can search the documentation and the API reference, create API keys and webhook endpoints, train classifiers, run workflows and Build AI, and fill PDFs — on your tenant, with the same metering and audit trail as the REST API. 53 tools across eight areas; a developer token generated in the console (30 or 90 days) or an OAuth 2.1 sign-in authenticates the session. #### Migration None. The server is additive; REST integrations are unaffected. #### Compatibility Tool calls reach the same services as the REST API and spend the same credits. API keys created through the server never carry the `apikeys.manage`, `mcp.read` or `mcp.write` scopes. #### Deprecations Nothing is deprecated. #### Rollout and rollback Serving in the EU production region (mcp-be release 2026.10.27.001; developer tokens and the 401 with `resource_metadata` since 2026-10-20). The US host is provisioned but the region is paused. Rollback is a redeploy of the previous mcp-be release; nothing in the REST API depends on it. #### Evidence [MCP server page](mcp.html), [documentation](https://docs.cogneris.ai/docs/developers/mcp-server/) and the live [server descriptor](https://mcp-eu.cogneris.ai/.well-known/mcp.json). October 2, 2026 ### Build AI: from a described document type to a draft setup #### Impact Build AI turns a described document type into a draft extraction setup: clarifying questions (2 credits per call), market research (10), a build plan (10) and plan revisions (5). It runs from the Vision console and from the MCP server's builder tools. #### Migration None; the feature is additive and opt-in per use. #### Compatibility Credits are spent from the same pool as every other operation; the four prices are on the pricing page. The REST routes under `/api/v1/builder/sessions` exist but are not yet part of the published OpenAPI contract and may change before they are. #### Deprecations Nothing is deprecated. #### Rollout and rollback Shipped in vision-be (2026.10.29.001), api-be and billing-be (2026.10.27.001) in the EU production region. Rollback is a redeploy of the previous releases; draft setups created by Build AI remain ordinary templates and are unaffected. #### Evidence [Credit table](pricing.html#how-credits-work) and the [MCP tool catalog](mcp.html). September 30, 2026 ### Client SDKs on npm, PyPI, NuGet.org and Maven Central #### Impact Client SDKs for TypeScript (`@cogneris-ai/document-ai-sdk`, with the `@cogneris-ai/document-ai-cli` CLI), Python (`cogneris-document-ai-sdk`), .NET (`Cogneris.DocumentAI`, .NET 8) and Java (`ai.cogneris:cogneris-document-ai-sdk`, Java 17) are published: 0.1.0 on all four registries on 2026-09-21, and 0.2.0 on npm and PyPI on 2026-09-30. #### Migration None; the REST API works without an SDK. Projects consuming local artifacts can switch to the registry packages at the same versions. #### Compatibility 0.2.0 is generated from the OpenAPI contract of 2026-09-28 and corrects a compatibility summary; see VERSIONING.md in the repository. The committed source is 1.0.0 and is not published. #### Deprecations Nothing is deprecated. There is no Go or mobile client. #### Rollout and rollback Published through the repository's release workflow (npm, PyPI) and owner-controlled publication (NuGet.org, Maven Central). Rollback of a faulty package is publishing a corrected version: registry versions are immutable and the previous one stays installable. #### Evidence [Language and platform support](api.html#language-support), the [repository README](https://github.com/cogneris-ai/cogneris-api-examples#readme), and the registry pages on [npm](https://www.npmjs.com/package/@cogneris-ai/document-ai-sdk), [PyPI](https://pypi.org/project/cogneris-document-ai-sdk/), [NuGet.org](https://www.nuget.org/packages/Cogneris.DocumentAI) and [Maven Central](https://central.sonatype.com/artifact/ai.cogneris/cogneris-document-ai-sdk). September 29, 2026 ### Scale plan: 40,000 credits for €499 per month #### Impact The Scale plan is available: €499 per month (US region: $499) for 40,000 credits, with overage at €0.01 per credit — the lowest self-serve rate per credit. #### Migration None. Pro subscriptions can upgrade at any time; the prorated difference is charged on the next invoice. #### Compatibility Free, Pro, Unlimited and Enterprise are unchanged. #### Deprecations Nothing is deprecated. #### Rollout and rollback Seeded in billing-be on 2026-09-29 and serving in the EU production region. Rollback is deactivating the plan in the catalogue; existing Scale subscriptions keep their terms until changed. #### Evidence [Pricing](pricing.html). September 16, 2026 ### Release notes and content-review metadata #### Impact Visitors can now find approved changes and verify when critical site content was reviewed. #### Migration No customer migration is required. #### Compatibility No API, schema, model, SDK, or runtime behavior changed. #### Deprecations No capabilities or interfaces were deprecated. #### Rollout and rollback The page ships with the static-site release and rolls back with the corresponding site revision. #### Evidence [Documentation](https://docs.cogneris.ai/), [status evidence](status.html), and the [public sitemap deployment record](sitemap.xml) show the published site surfaces and their currentness. ## Source: security.html [Skip to content](#main) Security & Trust # Engineered for documents that matter. We process documents that decide loans, claims, payouts and onboardings. That responsibility shapes every choice in our stack — from how data is encrypted to how requests are audited. This page documents what we do, what we don't, and where we're going. Last updated: 2026-04-30 · [Open the Trust Center](trust.html) Security pillars ## Four commitments. Built into the platform from day one. Security at Cogneris isn't a checklist bolted on at the end. These four principles are wired into every code path, every deploy, every customer integration. ### Platform encryption controls The Cogneris platform supports TLS in transit and AES-256-GCM encryption at rest with per-tenant keys managed in Google Cloud KMS where the service and workflow are configured for it. Customer-managed keys are a roadmap capability. ### Strict tenant isolation API requests are scoped to a tenant at the database layer, with tenant-aware authorization and query boundaries. Dedicated compute and single-tenant VPC deployment are roadmap capabilities unless explicitly included in a signed deployment scope. ### Least-privilege access Role-based access control. MFA required for every employee account. Production access follows a dual-approval workflow with time-bound permissions and audit logging. Engineer access to customer document content requires a documented business reason and reviewer approval. ### Configurable audit metadata Supported workflows can record tamper-evident metadata with timestamps, requestor identity, model version, and prompt and response hashes. Coverage, export, and retention depend on the configured service, tenant, and contract. Document lifecycle ## Where your documents live, and how long. We're explicit about every stage. No surprise retention, no opaque storage. Defaults are conservative; everything is configurable per contract. Step 01 ### Ingestion TLS 1.3 from your endpoint to our API. Document bytes are never persisted in access logs. Request metadata is logged without document content. Step 02 ### Processing Extraction runs in-memory only during inference. No document data crosses tenant boundaries. Sub-processor LLM calls are configurable per tenant — including no-LLM workflows. Step 03 ### Storage The platform supports encryption at rest with per-tenant keys where configured. Data-retention periods are tenant-configurable and contract-specific, including workflows configured to discard source documents and extracted content after processing. Step 04 ### Deletion Hard deletion at retention expiry. A 30-day soft-delete window protects against accidental loss; on request, we hard-delete immediately. Audit metadata is retained per your contract. Step 05 ### Audit export Configured workflows can produce tamper-evident metadata including prompt and response hashes, model version, reviewer identity, and timestamps. Retention is tenant-configurable and contract-specific. Step 06 ### Subject requests As a processor, Cogneris routes data subject requests (access, deletion, portability under GDPR or CCPA) to the controller — you. We provide tooling and a 30-day SLA for fulfillment. Compliance posture ## Working with regulated industries. We're transparent about where we are on each framework. Where an audit is in progress, we say so. Where it isn't, we say that too. Trust packets — including DPA, security questionnaire responses, and current audit reports — are available on request. SOC 2 Type II ● Audit in progress GDPR (EU) ● DPA available CCPA / CPRA ● DSR-ready HIPAA ● BAA on Enterprise ISO 27001 ● Roadmap PCI DSS ○ Out of scope ### GDPR — for our EU customers Cogneris acts as a data processor under GDPR Art. 28. Our standard Data Processing Addendum incorporates the EU Standard Contractual Clauses for transfers outside the EEA, and binds every sub-processor to equivalent terms. Sign the DPA at [legal@cogneris.ai](mailto:legal@cogneris.ai). ### CCPA / CPRA — for our California customers Cogneris is a service provider under CCPA §1798.140(ag). We do not sell or share personal information. Data subject requests (right to know, delete, correct, opt out of sharing) are routed to the controller; our SLA is 30 days from receipt. ### SOC 2 Type II — for our enterprise customers Our Type II audit is currently underway. Target report date, audit progress, and our current Type I report are available under NDA — email [security@cogneris.ai](mailto:security@cogneris.ai). ### HIPAA — for healthcare workflows Business Associate Agreements are available on Enterprise contracts. PHI handling is configurable per workflow — including no-retention modes and PHI-aware redaction in audit metadata. Sub-processors ## The infrastructure behind Cogneris. We list every sub-processor that handles customer data, what they do, and where they operate. Customers receive at least 30 days' notice before any addition or change, with the right to object under the DPA. | Sub-processor | Purpose | Region | Data accessed | | Google Cloud Platform | Cloud hosting, Cloud Run, Cloud SQL, GCS object storage, KMS, Secret Manager | EU (europe-west1) for EU-region tenants; US (us-central1) for US-region tenants, non-production environments, and shared administrative processing | Documents, extractions, account metadata, logs, secrets | | Google Cloud Vertex AI | AI inference, document extraction, classification | EU or US depending on tenant and model configuration | Document content during inference, prompts, extracted fields | | Amazon Web Services | Biometric face comparison (Amazon Rekognition CompareFaces) for identity verification | EU (eu-west-1) for EU-region tenants; US (us-east-1) for US-region tenants | The selfie and the identity-document portrait submitted for comparison; never any other document content | | OpenAI | AI inference, document extraction, document Q&A | US | Document content during inference, prompts, extracted fields | | Anthropic | AI inference, document Q&A, agentic workflows | US | Document content during inference, prompts, extracted fields | | New Relic | Observability: application performance monitoring, logs, metrics, traces, and browser monitoring | US | Operational telemetry, stack traces, request and database metadata, IP address and browser telemetry for end users who consent to analytics; document bodies are scrubbed before egress | | Fingerprint | Device intelligence for login fraud detection and account takeover prevention | US | Browser and device characteristics, IP address, and a persistent visitor identifier; never document content | | Google reCAPTCHA | Bot and abuse defence on sign-in and sign-up | US | IP address, browser characteristics, and interaction signals on authentication pages; never document content | | Stripe | Payment processing, billing, invoices | US | Billing contact, payment tokens, invoice metadata; never document content | | Twilio SendGrid | Transactional email and email notifications | US | Email address, message metadata, transactional content; never customer documents | | Twilio | SMS, WhatsApp, MFA and notifications | US | Phone number, message metadata, transactional content; never customer documents | | ipapi.co | IP geolocation lookup for authentication security auditing | Unconfirmed; vendor evidence pending | IP address; SSO discards the geolocation response and does not persist returned country, country-code, or city fields | Subscribe to [sub-processor change notifications](mailto:legal@cogneris.ai?subject=Sub-processor+notifications) to receive updates 30 days before any change. For planned private VPC and customer-managed-key capabilities, see the [private cloud roadmap](private-cloud-document-ai.html). Incident response ## If something goes wrong, customers hear from us first. We maintain a documented incident response plan with defined roles, escalation paths, and customer communication commitments. The headline numbers we hold ourselves to: Detection & triage ### ≤ 1 hour From a confirmed alert to incident commander assigned, severity classified, and customer-impact assessment started. Customer notification ### ≤ 48 hours For a confirmed personal-data breach affecting Customer Data: notification to affected customers within 48 hours of confirmation, designed so you can meet your own GDPR Art. 33 72-hour deadline. Post-incident review ### ≤ 10 business days Written post-incident review delivered to affected customers, including root cause, mitigations, and corrective actions. Notifications include — to the extent then known — the nature of the incident, the categories and approximate number of data subjects and records affected, the likely consequences, the measures taken or proposed, and a contact point for follow-up. Subsequent updates are issued as more information becomes available. Notifications are sent to your account-designated security contact and, on request, to a customer-supplied PGP-protected email address. Full breach-notification terms are in our [DPA](data-protection.html#breach-notification). Availability ## Service state is published, not asserted. Use the single canonical status entry point to choose a production region and inspect its current component state, incident history, and 90-day availability strip: [status.cogneris.ai](https://status.cogneris.ai). The regions run independently and are reported independently. Measured availability across 19 monitored components in the EU production region only was 99.78% over 2026-08-05 to 2026-10-29 (86 complete days), with no single component below 98.5477%. That is an observation, not a commitment — the contractual uptime SLA is a separate figure with service credits attached, and it is set out in your agreement. The measurement method is written out on the [status page](status.html). Responsible disclosure ## Found something? Tell us. We respect security researchers and respond to good-faith reports within 1 business day. We commit to working transparently with reporters and crediting their work. ### How to report - Email [security@cogneris.ai](mailto:security@cogneris.ai) with a clear description of the issue, reproduction steps, and impact. - If the issue is sensitive, request our PGP key in your initial email; we'll respond with one within four business hours. - Please give us 90 days before public disclosure. We commit to fixing critical issues within 30 days, high-severity within 60, and to credit you in the changelog (if you'd like). ### In scope - cogneris.ai and all `*.cogneris.ai` subdomains - The Document AI REST API and webhook system - Authentication, session and tenant-isolation logic - Customer dashboards and review queues ### Out of scope - Self-XSS / clickjacking on pages without sensitive actions - Issues in third-party services we sub-process to (report directly to the vendor) - Rate-limiting and brute-force on public endpoints (we have those covered) - Best-practice header reports without a demonstrated exploit Trust packet ## Need DPAs, audit reports or a security questionnaire? Most enterprise security questionnaires take us under 1 business day. Sub-processor notification subscriptions, DPAs, current SOC 2 status and our pen-test summary are all available on request — no NDA gate for the high-level docs. [Email security@cogneris.ai](mailto:security@cogneris.ai) [Talk to our team](contact.html) ## Source: private-cloud-document-ai.html [Skip to content](#main) Deployment roadmap and trust # Private cloud document AI on the roadmap. Private VPC deployment, customer-managed keys, AWS PrivateLink, and Google Cloud Private Service Connect are roadmap capabilities. They are not generally available unless explicitly included in a signed deployment scope. [Discuss deployment](contact.html)[Security posture](security.html) ## Why private cloud is on the roadmap Regulated production workflows may require dedicated infrastructure, VPC peering, customer-managed keys, region pinning, strict data retention, and audit evidence. This page describes evaluation criteria and intended direction, not currently committed availability. ### Private VPC roadmap Planned capability for a dedicated network boundary with tenant-scoped ingress, egress, logs, and secrets. ### Retention controls Set document data retention by workflow, including zero data retention beyond request processing when the use case requires it. ### Audit evidence Export model version, region, reviewer, validation, and webhook evidence for compliance review. ## Current and planned deployment options | Option | Best fit | Controls | | Shared cloud tenant | Standard API and portal workflows | Tenant isolation, encryption, regional residency, configurable retention | | Dedicated tenant | Enterprise procurement and higher-volume workflows | Dedicated keys, rate limits, stricter access controls, custom DPA terms | | Private VPC deployment — roadmap | Regulated buyers with requirements beyond shared SaaS | Planned private networking, BYOK/CMEK, and customer-specific boundaries | | Controlled single-tenant cloud — roadmap | Customers evaluating operational isolation | Planned single-tenant compute and customer-specific observability boundaries | ## Zero data retention and audit retention are different Zero data retention means source documents and extracted content can be discarded after request processing. Audit metadata is a separate record: request ID, timestamps, model version, validation status, reviewer activity, and export decisions can be retained under your contract so compliance teams can still reconstruct what happened. ## Private cloud evaluation checklist Ask every document AI vendor where inference runs, whether embeddings leave the selected region, whether document bodies appear in logs, whether LLM sub-processors use zero-retention APIs, how key revocation works, and whether the audit trail records deployment region and model version. For Cogneris, confirm current availability in a signed deployment scope rather than relying on roadmap language. ## Related pages [Security](security.html)[Environments](https://docs.cogneris.ai/docs/api/getting-started/)[Document extraction API](document-extraction-api.html)[Document parsing API](document-parsing-api.html)[Benchmark guide](document-extraction-benchmark.html)[IDP buyer's guide](what-is-idp.html) ## Source: data-protection.html [Skip to content](#main) Legal # Data Protection. How Cogneris processes customer data on your behalf — Data Processing Addendum, current sub-processors, international transfer mechanisms, and the security measures backing every commitment. Last updated: 14 May 2026 · Sub-processor list version: 2026.05 This page is the central reference for Cogneris customers and prospects evaluating our data protection posture. It links to our binding contractual commitments (the Data Processing Addendum) and discloses the sub-processors we engage to deliver the Service. Read this page alongside our [Privacy Policy](privacy.html) [Security](security.html), and [Trust Center](trust.html) documentation for the full picture. ## 1. Data Processing Addendum (DPA) When Cogneris processes personal data on a customer's behalf in connection with the Service, it does so as a processor (as defined in GDPR Art. 4) on the customer's documented instructions. Our DPA is incorporated by reference into our [Terms of Service](terms.html) and applies automatically to every customer that processes personal data through the Service. A counter-signed copy is available on request to [legal@cogneris.ai](mailto:legal@cogneris.ai). ### 1.1 What the DPA covers (summary) The summary below is provided for convenience. The full DPA controls in case of conflict. Roles Cogneris is a processor; the customer is the controller. Where the customer is itself a processor for an underlying controller, Cogneris acts as a sub-processor. Subject matter, duration, nature, and purpose Cogneris processes Customer Data for the purpose of providing the Service to the customer for the duration of the customer's subscription. Processing operations include receipt, storage, classification, extraction, validation, fraud detection, retrieval, and deletion of documents and the personal data they contain. Categories of data subjects and personal data As determined by the customer when uploading documents. Typical categories include the customer's employees, contractors, vendors, and end users; typical personal data includes identifiers, contact details, financial transaction data, and document metadata. The customer is responsible for not uploading data that requires additional contractual safeguards (e.g. PHI, payment card data in raw form) without first executing the additional terms required by Section 6.3 of the Terms of Service. Processor obligations Cogneris (a) processes Customer Data only on the customer's documented instructions; (b) ensures personnel are bound by confidentiality; (c) implements the security measures described on our Security page; (d) engages sub-processors only on the terms in Section 2 below; (e) assists with data subject requests, DPIAs, and regulator inquiries; (f) notifies the customer of personal data breaches without undue delay (target: within 48 hours of confirmation); and (g) deletes or returns Customer Data on termination as set out in the Terms. No training on Customer Data Cogneris does not use Customer Data, document content, or extracted fields to train or fine-tune generally-available models or any model made available to other customers. Where third-party model providers are used, Cogneris requires contractually that Customer Data is not used to train their models. Audit rights Cogneris makes its independent third-party audit reports (SOC 2 Type II once issued) and ISO security documentation available under NDA. Customers retain the right to audit beyond those reports for cause, on reasonable notice, at the customer's expense. ### 1.2 International transfers and Transfer Impact Assessment Cogneris relies on the European Commission's Standard Contractual Clauses adopted by Implementing Decision (EU) 2021/914 of 4 June 2021 for transfers of personal data from the EEA to third countries that are not covered by an adequacy decision, applying: - Module Two (controller-to-processor) when the customer is a controller and engages Cogneris as processor; - Module Three (processor-to-processor) when the customer is a processor and Cogneris acts as a sub-processor. For UK personal data, Cogneris applies the UK International Data Transfer Addendum (Version A1.0, 21 March 2022) to the EU SCCs. For Swiss personal data, the SCCs are read with the adaptations published by the Swiss FDPIC. Cogneris performs and documents a Transfer Impact Assessment for every destination country that is not covered by an adequacy decision, applying the EDPB Recommendations 01/2020 methodology. The current TIA covers transfers to the United States and concludes that the combination of the SCCs, our supplementary measures (see below), and the practical likelihood of access for the data categories in scope brings the level of protection up to a standard essentially equivalent to EU law. The TIA is reviewed at least annually and on triggering events (e.g., new sub-processor, material change in the destination country's legal regime, or a government access request). A summary of the TIA is available under NDA on request to [legal@cogneris.ai](mailto:legal@cogneris.ai?subject=TIA%20request). Supplementary measures applied to all in-scope transfers: - Technical: The platform supports TLS in transit and AES-256-GCM at rest with per-tenant keys in GCP KMS where configured; pseudonymization of identifiers at the API edge where feasible; minimal-scope payloads for LLM providers; and restricted production egress. Customer-managed keys are a roadmap capability unless included in a signed deployment scope. - Organizational: written government-access policy requiring legal review of any disclosure request; sub-processor due diligence and annual recertification; mandatory privacy and security training; incident response playbook with defined customer notification. - Contractual: EU SCCs with all sub-processors; onward-transfer restrictions; sub-processor obligation to notify Cogneris of any government access request affecting Customer Data (where legally permitted) and to use lawful means to challenge disproportionate requests; audit and termination rights. ### 1.3 Breach notification to customers Where Cogneris becomes aware of a confirmed personal data breach affecting Customer Data (Art. 4(12) GDPR), Cogneris will notify affected customers without undue delay and within 48 hours of confirmation, providing — to the extent then known — the nature of the breach, categories and approximate number of data subjects and records affected, the likely consequences, the measures taken or proposed, and the contact point for follow-up. Subsequent updates are issued as more information becomes available. This commitment is designed to enable the customer (as controller) to comply with its own 72-hour notification obligation under Art. 33 GDPR. Notifications are sent to the customer's account-designated security contact and, on request, to a customer-supplied PGP-protected email address. ### 1.4 California Consumer Privacy Act — service provider terms Where Cogneris processes personal information of California residents on behalf of a customer that is a "business" under the CCPA/CPRA, Cogneris acts as a "service provider" as defined in Cal. Civ. Code §1798.140(ag). The DPA includes the service-provider terms required by §1798.140(ag)(1), namely: - Cogneris is prohibited from selling or sharing personal information (as those terms are defined in the CCPA/CPRA); - Cogneris is prohibited from retaining, using, or disclosing the personal information for any purpose other than the specific business purpose set out in the agreement (the provision of the Service), including a prohibition on retaining, using, or disclosing the personal information outside of the direct business relationship between Cogneris and the customer; - Cogneris is prohibited from combining personal information received from one business with personal information received from any other source, except as permitted under §1798.140(ag)(1)(D); - Cogneris agrees to comply with applicable obligations under the CCPA/CPRA and to provide the same level of privacy protection as required of businesses by the CCPA/CPRA; - the customer has the right to take reasonable and appropriate steps to ensure that Cogneris uses the personal information transferred in a manner consistent with the customer's obligations under the CCPA/CPRA; - Cogneris will notify the customer if it determines it can no longer meet its obligations under the CCPA/CPRA; - the customer has the right, upon notice, to take reasonable and appropriate steps to stop and remediate unauthorized use of personal information; - on termination, Cogneris will delete or return all personal information processed on the customer's behalf in accordance with the DPA and the applicable Order Form, and require the same of its sub-processors. Cogneris does not engage in cross-context behavioral advertising and does not "share" personal information for that purpose. The same restrictions are mirrored in our contracts with sub-processors that process Customer Data. ## 2. Sub-processors Cogneris engages the third-party sub-processors below to deliver, secure, and support the Service. Each is assessed before onboarding for security, privacy, and reliability, and is engaged under a written agreement imposing data protection obligations no less protective than those in our DPA. This list is generated from src/subprocessors.json and verified on every pull request. Registry version 2026.10, last updated 2026-10-30. ### 2.1 Notification of changes Cogneris will provide at least 30 days' prior notice before adding or replacing a sub-processor that processes Customer Data. Notice is delivered by email to the account's primary admin and by update to this page. Customers may object on reasonable data protection grounds during the notice period; if the parties cannot resolve the objection, the customer may terminate the affected portion of the Service for convenience and receive a pro-rata refund of unused, prepaid fees. Subscribe to sub-processor change notifications by emailing [legal@cogneris.ai](mailto:legal@cogneris.ai?subject=Subscribe%20-%20sub-processor%20updates). ### 2.2 Infrastructure and platform sub-processors | Sub-processor | Purpose | Processing location | Transfer mechanism | | Google Cloud Platform (Google LLC / Google Ireland Ltd.) | Cloud hosting, Cloud Run, Cloud SQL, GCS object storage, KMS, Secret Manager | EU (europe-west1) for EU-region tenants; US (us-central1) for US-region tenants, non-production environments, and shared administrative processing | EU SCCs / Google Cloud DPA (evidence pending) | ### 2.3 AI model sub-processors Cogneris orchestrates specialist agents that may invoke third-party large language models. Customer Data sent to these providers is governed by no-training contractual terms. Customers can configure model selection per tenant, including no-LLM workflows, and Enterprise customers can request model allow-lists. | Sub-processor | Purpose | Processing location | Transfer mechanism | | Google Cloud Vertex AI (Google LLC / Google Ireland Ltd.) | AI inference, document extraction, classification | EU or US depending on tenant and model configuration | EU SCCs / Google Cloud DPA (evidence pending) | | Amazon Web Services (Amazon Web Services, Inc. / Amazon Web Services EMEA SARL) | Biometric face comparison (Amazon Rekognition CompareFaces) for identity verification | EU (eu-west-1) for EU-region tenants; US (us-east-1) for US-region tenants | EU SCCs / AWS GDPR Data Processing Addendum (evidence pending) | | OpenAI (OpenAI, L.L.C.) | AI inference, document extraction, document Q&A | US | EU SCCs and OpenAI DPA (evidence pending) | | Anthropic (Anthropic, PBC) | AI inference, document Q&A, agentic workflows | US | EU SCCs and Anthropic DPA (evidence pending) | ### 2.4 Operational sub-processors | Sub-processor | Purpose | Processing location | Transfer mechanism | | New Relic (New Relic, Inc.) | Observability: application performance monitoring, logs, metrics, traces, and browser monitoring | US | EU SCCs and New Relic DPA (evidence pending) | | Fingerprint (FingerprintJS, Inc.) | Device intelligence for login fraud detection and account takeover prevention | US | EU SCCs and Fingerprint DPA (evidence pending) | | Google reCAPTCHA (Google LLC / Google Ireland Ltd.) | Bot and abuse defence on sign-in and sign-up | US | EU SCCs and Google Terms (evidence pending) | | Stripe (Stripe, Inc.) | Payment processing, billing, invoices | US | EU SCCs and Stripe DPA (evidence pending) | | Twilio SendGrid (Twilio SendGrid, Inc. / Twilio Inc.) | Transactional email and email notifications | US | EU SCCs and Twilio DPA (evidence pending) | | Twilio (Twilio Inc.) | SMS, WhatsApp, MFA and notifications | US | EU SCCs and Twilio DPA (evidence pending) | | ipapi.co (Kloudend, Inc.) | IP geolocation lookup for authentication security auditing | Unconfirmed; vendor evidence pending | EU SCCs and Kloudend DPA (evidence pending) | Affiliates of the entities above that perform sub-processing on the same terms are deemed authorized sub-processors and do not require separate notice. Where a sub-processor entry shows multiple processing locations, the actual location depends on the customer's selected region and the routing of the specific request. Entries marked evidence pending are operating under the stated transfer mechanism while Legal completes the counter-signed DPA, SCC and transfer impact assessment file for that vendor. Customers may request the current evidence status for any sub-processor at [legal@cogneris.ai](mailto:legal@cogneris.ai). ## 3. Customer-controlled data residency Document storage and primary processing run in the EU region by default. Customers can request a US-region deployment at the time of onboarding. Enterprise customers can request additional residency options (e.g. UK, Canada, Brazil) subject to a scoping conversation. Selecting a residency option restricts the storage and primary processing location, but routine sub-processor calls (for example, Stripe billing, email delivery, error monitoring) may still occur in the sub-processor's standard region. ## 4. Security measures (Annex II to the DPA) Cogneris maintains a written information security program. The technical and organizational measures we apply are described in detail on our [Security](security.html) page and form Annex II to our DPA. They include, at a minimum: - platform support for encryption in transit and AES-256-GCM encryption at rest with per-tenant keys in Google Cloud KMS where configured; - strict tenant isolation at the database layer; optional dedicated infrastructure for Enterprise; - role-based access control with mandatory MFA for all employee accounts and dual-approval, time-bound production access; - immutable audit logging of every extraction, review, and export, including model and prompt metadata; - vulnerability management, dependency scanning, and quarterly penetration testing; - documented incident response plan with defined breach notification commitments; - employee security training, background checks where permitted, and confidentiality undertakings. ## 5. Data subject requests Where Cogneris processes personal data as a processor, it will assist the customer (the controller) in responding to data subject requests within applicable statutory deadlines. Data subjects who interact directly with Cogneris's website and marketing are covered by the rights described in our [Privacy Policy](privacy.html). Data subjects whose data was uploaded to the Service by a Cogneris customer should submit requests to that customer in the first instance. Direct requests to Cogneris can be filed through our [privacy request form](privacy-request.html); if the online service is unavailable, email the contact shown there. For other data protection queries, our [Data Protection Officer and EU/UK representatives](dpo.html) can be contacted directly. ## 6. How to request the DPA A pre-executed DPA, including the SCCs, is available on request. Email [legal@cogneris.ai](mailto:legal@cogneris.ai?subject=DPA%20request) with your legal entity name, the entity name of the Cogneris contracting party (if known), and any specific schedules you require (e.g. UK Addendum, Swiss adaptations, customer-specific Annex II). ## 7. AI transparency (EU AI Act Article 50) The Cogneris Service uses AI to classify documents, extract structured fields, answer questions about document content, and generate advisory fraud-risk signals. This section sets out how Cogneris meets the transparency obligations of Article 50 of Regulation (EU) 2024/1689 (the "EU AI Act") for the relevant parts of the Service. ### 7.1 Direct interaction with an AI system (Art. 50(1)) Where a natural person interacts directly with the Service in a way that involves an AI system (for example, the Document Q&A feature), the Cogneris user interface and API responses make clear that the response is produced by an AI system. AI-mediated responses are labelled as such and accompanied by source citations to the underlying document. This obligation is on Cogneris as provider of the AI system. ### 7.2 AI-generated or AI-modified content (Art. 50(2)) Outputs generated by the Service that constitute synthetic text — including summaries, AI-drafted explanations, and answers in Document Q&A — are marked as AI-generated in the response payload, both in the user interface and in the API response (for example, via a dedicated `generated_by_ai: true` field and a machine-readable provenance label). Field-extraction results (where the AI identifies a value already present in the document) are not "synthetic" content but are still flagged with confidence scores and the model version that produced them. ### 7.3 Emotion recognition and biometric categorization (Art. 50(3)) Cogneris does not operate an emotion-recognition system and does not provide biometric categorization. Where a customer's documents include facial images (for example, ID-document workflows), the Service performs document field extraction; it does not infer emotion or biometric categories. If a customer plans to use any such functionality, additional contractual terms and a separate impact assessment are required before onboarding. ### 7.4 Deep fakes and AI-generated images, audio, or video (Art. 50(4)) The Service does not generate deep-fake images, audio, or video. The Service is a structured-data extraction and Q&A platform; it does not produce synthetic media. ### 7.5 Customer (deployer) obligations When a Cogneris customer deploys the Service in a context governed by the EU AI Act — particularly where the customer's overall workflow may itself qualify as a high-risk AI system under Annex III (e.g., creditworthiness assessment, life and health insurance pricing, certain employment uses) — the customer (as deployer) is responsible for the deployer-side obligations, including notifying affected natural persons under Art. 50, conducting a Fundamental Rights Impact Assessment under Art. 27 where required, and ensuring meaningful human oversight before any consequential decision. Cogneris provides the documentation, logging, and configuration needed to support those obligations and references them in our [Terms of Service](terms.html) §7 (acceptable use) and the DPA. ### 7.6 Model documentation and inventory Cogneris publishes, and updates on material change, an inventory of the AI models used in the Service, including model family, version, provider, deployment region, and intended use. The current inventory is available under NDA on request to [legal@cogneris.ai](mailto:legal@cogneris.ai?subject=Model%20inventory%20request). Where a customer's procurement requires a Provider Card or System Card meeting the EU AI Act Art. 53 (general-purpose AI) or Art. 13 (high-risk AI) documentation format, Cogneris will provide one for the relevant components. ## 8. Changes to this page We may update this page to reflect new sub-processors, regional availability, or changes to our processing operations. The "Last updated" date and the sub-processor list version at the top of this page indicate the current revision. Material changes to sub-processors are notified separately as described in Section 2.1. ## Source: trust.html [Skip to content](#main) Trust Center # Security, privacy, and compliance for document AI. This page summarizes Cogneris security controls, data protection commitments, subprocessors, availability references, customer report requests, and the compliance roadmap. Last updated: 30 October 2026 ## Security Overview [Read the security overview](security.html) ## Data Protection [Read data protection details](data-protection.html) ## Availability Use the single canonical status entry point to choose a production region and inspect its current component state, incident history, and 90-day availability strip: [status.cogneris.ai](https://status.cogneris.ai). Measured availability 99.78% across 19 monitored components in the EU production region only, 2026-08-05 to 2026-10-29 (86 complete days). Lowest single component in that window: 98.5477%. Measured availability is history, not a commitment. The contractual uptime SLA is a separate figure with service credits attached and is set out in your agreement. [How this number is measured](status.html) ## Report Request Enterprise customers can request the customer evidence package, DPA template, security overview, subprocessor list, TIA summary, and incident history summary. [Request trust materials](contact.html?topic=trust-report) ## Subprocessors | Name | Purpose | Processing location | Legal entities and transfer mechanisms for each sub-processor are recorded on the [Data Protection](data-protection.html) page. ## Compliance Roadmap | Item | Status | ## Controls ## Data collected ## Source: status.html [Skip to content](#main) Status # What our platform actually did. Each Cogneris production region publishes a live status page; a paused region is named as such below. The number below is what those pages measured — not what a contract promises. We keep the two apart on purpose. Measured availability · 2026-08-05 to 2026-10-29 · 86 days 99.78% What is measured 19 monitored components across the EU production region only, 1634 component-days, each weighted equally. Worst single component 98.5477% — tickets-be in the EU region. No component in the window fell below this. Source The public `status.json` feed behind each regional status page. Anyone can read it back. ### This is not the SLA Measured uptime is history: what the monitors recorded over a stated window. It has no contractual force and it will change every time we re-measure. The uptime SLA is a commitment with service credits attached — 99.9% monthly on Professional, 99.99% on Enterprise. The binding wording lives in your contract, and the tiers are summarised on [pricing](pricing.html). A measured number above an SLA is not evidence the SLA will be met, and a measured number below one is not automatically a breach. Read them separately. ## Measured response time The probes behind the availability figure also record how long each component takes to answer. Those samples are published here, because "low latency" is not a claim until there is a number under it — and the number has to say what it measured. Median response time · 2026-10-29T14:35:00Z to 2026-10-30T14:30:00Z 150 ms What is measured 19 monitored components across the EU production region only, 5472 probe samples taken every 300 seconds. Tail p95 300 ms · p99 396 ms. Nearest-rank over every sample, so each figure is an observed response, not an interpolation. Source The same public `status.json` feed behind the availability number. Anyone can read it back. ### What this number is not It is not document-processing time. The probes measure how fast an endpoint answers, not how long an extraction takes. Extraction time moves with document size and profile, and a 40-page PDF has nothing to do with the figure above. When we publish that number it will be measured separately and labelled as such. It is not a contractual commitment. Like measured availability, this is history over a stated window. It changes every time we re-measure and it carries no service credits. The slowest samples are kept, not trimmed. The probe cuts off at 10000 ms; 0 samples in this window reached it, so for those the true value is "at least that". Dropping them would flatter the tail. | Service group | p50 | p95 | p99 | | Internal Tools | 144 ms | 297 ms | 396 ms | | Public API & Billing | 161 ms | 321 ms | 444 ms | | Authentication | 162 ms | 324 ms | 402 ms | | Content Management | 143 ms | 258 ms | 361 ms | | E-Signature | 147 ms | 276 ms | 354 ms | | Document Extraction | 153 ms | 296 ms | 402 ms | | Workflow | 144 ms | 300 ms | 396 ms | Region by region — US median not reported, EU median 150 ms. The figure above covers the EU production region only. A region reads not reported when its status feed carried no latency series for this window; it then contributes nothing to the figure, and the scope line above says so. Where both report, the gap between them is the network path from the probe to each estate as much as the estate itself, so read it as observed round-trip rather than one region being slower at its work. Each region's feed is generated independently and one can lag the other by a day, so each contributes its own most recent 24 hours. ## Live status, per region Cogneris runs separate production estates in the United States and the European Union. They fail independently, so they are reported independently. Each page shows current component state, a 24-hour response-time chart, a 90-day history strip, and incident records. A region whose status feed has stopped updating is paused: it is marked below, contributes nothing to the measured number, and its page keeps the last published history. [ United States ### US production status status-us.cogneris.ai State: paused; status feed last updated 2026-09-12. Over the window above: not measured across 0 components; lowest component not measured at not measured. ](https://status-us.cogneris.ai) [ European Union ### EU production status status-eu.cogneris.ai State: measured; status feed last updated 2026-10-30. Over the window above: 99.7847% across 19 components; lowest component tickets-be at 98.5477%. ](https://status-eu.cogneris.ai) ## How the number is produced Every claim on this page comes from the same feed the status pages render. The measurement is scripted, the output is committed, and a check fails the build if the page and the measurement disagree. | Step | What happens | | Collect | A collector polls each component through Cloud Monitoring uptime checks and writes a daily availability figure per component, per region. | | Window | Complete UTC days from the first day on which every monitored component in both production regions had been in service for a full day. That rule exists because the raw 90-day history reaches back into the rollout, when components that were not deployed yet read as 0% and drag the fleet average to roughly 58%. Publishing that would be measuring the build-out, not the service. | | Aggregate | The mean over component-days — every monitored component weighted equally on every complete day in the window. The worst single component is published alongside it, because an average on its own hides an outlier. | | Publish | The result is committed to `src/uptime-evidence.json` and rendered here. CI re-checks that this page, [pricing](pricing.html), [security](security.html), and the [Trust Center](trust.html) all render the committed figure. | Window length — 86 days is short, and we would rather say so than stretch it. Every component in the EU production region only was in service on 2026-08-05; the window grows from there, and this page is re-measured as it does. Last measured 2026-10-29. ## Incidents and notifications Incidents are posted to the status page for the affected region as they are worked, with follow-up notes when the cause is understood. Customer notification obligations, response targets and the escalation path are contractual and set out in your agreement — see [Security](security.html) for how incidents are handled and [Trust Center](trust.html) for the evidence package. For environment endpoints and health checks, see [environments](https://docs.cogneris.ai/docs/api/getting-started/). [Release notes](release-notes.html) ## Source: privacy.html [Skip to content](#main) Legal # Privacy Policy. How we collect, use, share, and protect personal information when you use the Cogneris document intelligence platform, our website, and related services. Last updated: 27 October 2026 · Effective date: 7 May 2026 This Privacy Policy explains how Cogneris, Inc. ("Cogneris", "we", "us", or "our") processes personal information in connection with the Cogneris document intelligence platform, the cogneris.ai website (the "Site"), our APIs, and related services (together, the "Service"). It applies to visitors to our Site, prospective customers, customers, end users authorized by our customers, and other individuals who interact with us. Cogneris acts as the controller of personal information we collect directly from website visitors and account administrators of our customers. When our customers use the Service to process documents that contain personal data, Cogneris acts as a processor on the customer's behalf, and our processing is governed by the customer's agreement with us, including our Data Processing Addendum ("DPA"). This Policy describes our practices in our capacity as controller. For processing as processor, please contact your Cogneris account administrator. ## Who we are Cogneris is the trading name of Cogneris, Inc., a corporation incorporated in the State of Delaware, United States (state registration no. 10640571). - Registered entity: Cogneris, Inc. - Principal place of business: 7345 W Sand Lake Rd, Ste 210, Office 3732, Orlando, FL 32819, United States - Telephone: [+1 407 934-0064](tel:+14079340064) - Email: [legal@cogneris.ai](mailto:legal@cogneris.ai) - Website: cogneris.ai ## 1. Information we collect ### 1.1 Information you provide - Account information: name, work email, role, company, phone number, and login credentials when you register, request a demo, or contact us. - Billing information: billing contact, billing address, and tax identifiers. Card data is collected by our payment processor (Stripe); we do not store full card numbers. - Communications: the contents of messages you send us through email, contact forms, support tickets, and meetings. - Marketing preferences: opt-ins, event registrations, and content you request. ### 1.2 Information we collect automatically - Usage data: pages viewed, features used, API calls, timestamps, error reports, and performance metrics. - Device and connection: IP address, browser type and version, operating system, language, referring URL, and approximate location derived from IP. - Cookies and similar technologies: see Section 11. ### 1.3 Customer Data processed on behalf of customers Our customers upload documents and configure extraction templates through the Service ("Customer Data"). Customer Data may contain personal data of the customer's employees, contractors, vendors, or end users. When Cogneris processes Customer Data, it does so as a processor on documented instructions from the customer under the customer's agreement and the DPA. We do not determine the purposes of processing Customer Data, and we do not use Customer Data to train or improve generally-available models (see Section 3). ### 1.4 Information from third parties - Authentication providers (e.g. Google, Microsoft) when you sign in with single sign-on. - Business data providers for sales prospecting (e.g. company size, industry, public profile information). - Service providers who help us operate the Services (analytics, customer support, payments, communications). ## 2. How we use personal information We use personal information to: - provide, operate, secure, and improve the Services; - create and administer accounts, authenticate users, and manage subscriptions; - process payments, invoicing, and tax; - respond to inquiries, provide customer support, and communicate service-related notices; - send marketing communications about Cogneris products, events, and content (subject to your preferences and applicable law); - monitor, troubleshoot, prevent fraud, abuse, and security incidents, and enforce our terms; - comply with legal obligations, respond to lawful requests, and protect our rights and the rights of others; - maintain audit logs as described in our security documentation; and - conduct internal analytics and product research using aggregated and de-identified data. ## 3. AI models and training Cogneris does not use Customer Data, document content, or extracted fields to train, fine-tune, or evaluate generally-available foundation models or any model made available to other customers. We may use de-identified, aggregated metadata about system performance (such as latency, error rates, and feature usage) to improve the Services. Where we use third-party model providers (such as OpenAI, Anthropic, or Google Vertex), we contractually require that Customer Data sent to those providers is not used to train their models. ## 4. Legal bases (EEA, UK, Switzerland) If you are in the European Economic Area, the United Kingdom, or Switzerland, our legal bases for processing personal information are: - Performance of a contract — to provide the Services to you or your organization, manage accounts, and provide support. - Legitimate interests — to operate, secure, and improve the Services, prevent fraud and abuse, conduct sales and marketing to business contacts, and develop our business, where those interests are not overridden by your rights. - Consent — for cookies that are not strictly necessary, certain marketing communications, and where required by applicable law. You can withdraw consent at any time. - Legal obligation — to comply with tax, accounting, regulatory, and other legal requirements. ## 5. How we share personal information We share personal information with: - Sub-processors and service providers who provide infrastructure (Google Cloud), payments (Stripe), email and communications (SendGrid, Twilio), analytics, customer support, and AI model inference (OpenAI, Anthropic, Google Vertex). A current list of sub-processors is available on our [Data Protection](data-protection.html) page. - Customers — if you are an authorized user of a customer account, we share your account and usage information with the customer that administers that account. - Professional advisors such as auditors, lawyers, and accountants under confidentiality obligations. - Authorities and other parties when required by law, in response to a lawful request, or to protect our rights, the safety of users, or the integrity of the Services. - Successors in connection with a merger, acquisition, financing, reorganization, or sale of assets, subject to confidentiality obligations. We do not sell personal information for monetary consideration. We do not "share" personal information for cross-context behavioral advertising as those terms are defined under California law. ## 6. International data transfers Cogneris operates internationally. Personal information may be transferred to, processed in, and stored in countries other than the country in which it was collected, including the United States and the European Union. Where we transfer personal information from the EEA, the UK, or Switzerland to a country that has not been recognized as providing an adequate level of data protection, we rely on appropriate safeguards, including the European Commission's Standard Contractual Clauses (and the UK International Data Transfer Addendum where applicable). You can request a copy of the safeguards we apply by contacting [legal@cogneris.ai](mailto:legal@cogneris.ai). ## 7. Retention We retain personal information for as long as necessary to provide the Services and for the purposes described in this Policy, including to comply with legal, accounting, and regulatory obligations, resolve disputes, enforce agreements, and maintain audit trails. Customer Data is retained according to the customer's contract and configured retention settings; on termination, Customer Data is deleted or returned in line with the DPA. Our default audit-log retention is seven (7) years. Marketing contact information is retained until you unsubscribe or otherwise object. ## 8. Your rights Depending on where you live, you may have the following rights with respect to your personal information: - Access — request a copy of the personal information we hold about you. - Rectification — request correction of inaccurate or incomplete information. - Erasure / deletion — request that we delete personal information, subject to legal exceptions. - Restriction — request that we restrict processing in certain circumstances. - Objection — object to processing based on legitimate interests or for direct marketing. - Portability — receive personal information in a structured, commonly used, machine-readable format. - Withdraw consent — where processing is based on consent. - Opt out of sale or sharing for cross-context behavioral advertising (California) — Cogneris does not engage in either, but you may submit a request to confirm. - Limit use of sensitive personal information (California) — Cogneris does not use sensitive personal information for purposes that require this right, but you may submit a request. - Non-discrimination — we will not discriminate against you for exercising your rights. - Lodge a complaint with your supervisory authority. For the EU, that is the data protection authority of your country of residence; for the UK, that is the Information Commissioner's Office (ICO). ### How to exercise your rights Use our [privacy request form](privacy-request.html), which sends the request only to Cogneris's explicitly configured privacy-request API. If online submission is unavailable, email [dpo@cogneris.ai](mailto:dpo@cogneris.ai). We will verify your identity using information already associated with your account and will respond within the time frame required by applicable law (generally 30 days under GDPR/UK GDPR; 45 days under CCPA/CPRA, extendable as permitted). If your information is processed by Cogneris on behalf of one of our customers (Customer Data), please contact that customer directly; we will support the customer in responding where required. When we log your request, we email you a receipt with a tracking code. Look up the status and chain-position proof at [cogneris.ai/privacy/request/track/](privacy/request/track/). ## 9. Children The Services are intended for businesses and are not directed to children. We do not knowingly collect personal information from anyone under the age of 16. If you believe a child has provided us with personal information, please contact us and we will take appropriate steps to delete it. ## 10. Security We implement technical and organizational measures designed to protect personal information, including encryption in transit (TLS 1.3) and at rest (AES-256-GCM), strict tenant isolation, role-based access controls, multi-factor authentication for employees, and continuous monitoring. A summary is available on our [Security](security.html) page. No system is perfectly secure; we encourage you to use a strong, unique password and to notify us of any suspected unauthorized access. ## 11. Cookies and similar technologies We use cookies and similar technologies to operate the Site and Platform, remember your preferences, analyze usage, and (where you consent, in jurisdictions that require consent) measure the effectiveness of marketing. You can manage cookie preferences through our cookie banner where shown and through your browser settings. Strictly necessary cookies cannot be disabled. We honor Global Privacy Control (GPC) signals where required. On cogneris.ai, analytics cookies come from Google Analytics and marketing cookies from the Meta Pixel (Meta Platforms Ireland Limited). Neither is loaded until you accept that category in the cookie banner, and neither is loaded at all when your browser sends a GPC signal. You can change your choice at any time from the "Cookie preferences" link in the footer. ## 12. Third-party links The Site may contain links to third-party websites or services. We are not responsible for the privacy practices of those third parties. We encourage you to review their privacy notices. ## 13. Changes to this Policy We may update this Policy from time to time. If we make material changes, we will notify you by email (where we have your address) or by prominent notice on the Site before the changes take effect. The "Last updated" date at the top of this Policy indicates when it was last revised. ## 14. How to contact us For questions about this Policy, to exercise your rights, or to raise a concern, contact: - Email: [legal@cogneris.ai](mailto:legal@cogneris.ai) - Telephone: [+1 407 934-0064](tel:+14079340064) - Postal: Cogneris, Inc. — Privacy, 7345 W Sand Lake Rd, Ste 210, Office 3732, Orlando, FL 32819, United States ### EU representative (Article 27 GDPR) Cogneris does not currently maintain an establishment in the European Economic Area, so we are designating a representative in the EU under Article 27 GDPR. The representative's name and postal address will be published here once the appointment is signed. In the interim, individuals in the European Economic Area can raise questions or complaints about our processing at: - Email: [eu-rep@cogneris.ai](mailto:eu-rep@cogneris.ai) ### UK representative (Article 27 UK GDPR) For individuals in the United Kingdom, a separate representative is being appointed under Article 27 of the UK GDPR. Its name and postal address will be published here once the appointment is signed. In the interim, contact: - Email: [uk-rep@cogneris.ai](mailto:uk-rep@cogneris.ai) Appointing an Article 27 representative does not affect Cogneris's own responsibility under the GDPR or UK GDPR. Designating a representative is required for controllers and processors that are not established in the EU/UK and are subject to those laws under Article 3(2). If Cogneris maintains an establishment in the EU or the UK, the relevant lead supervisory authority is identified in our records of processing and available on request. ### Brazil (LGPD) If you are in Brazil, Cogneris's LGPD privacy contact is [legal@cogneris.ai](mailto:legal@cogneris.ai). You have the right to confirm processing, access your data, request correction or anonymization, request portability, request deletion, obtain information about shared parties, and revoke consent. You may also lodge a complaint with the Autoridade Nacional de Proteção de Dados (ANPD). Encarregado / Data Protection Officer: the designation under LGPD Art. 41 is in progress, and the Encarregado's name will be published here once it completes. In the interim, write to [dpo@cogneris.ai](mailto:dpo@cogneris.ai). ## Source: terms.html [Skip to content](#main) Legal # Terms of Service. These Terms govern your use of the Cogneris document intelligence platform. Please read them carefully — they include important provisions about liability, dispute resolution, and how we treat your data. Last updated: 7 May 2026 · Effective date: 7 May 2026 Quick summary (not a substitute for the Terms below): Cogneris provides a document intelligence platform. You own your documents and the data extracted from them. We do not use your documents to train generally-available AI models. We provide the Service "as available" with limited warranties; our liability is capped. A separate Data Processing Addendum governs personal data processing. If you have a signed Master Services Agreement with Cogneris, that agreement controls and these Terms are subordinate. ## 1. Acceptance of these Terms These Terms of Service (the "Terms") form a binding agreement between you, or the entity you represent ("Customer", "you"), and Cogneris, Inc., a Delaware corporation with its principal place of business at 7345 W Sand Lake Rd, Ste 210, Office 3732, Orlando, FL 32819, United States ("Cogneris", "we", "us", or "our"), and govern your access to and use of the Cogneris document intelligence platform, our APIs, our website cogneris.ai, and related services (collectively, the "Service"). By creating an account, accessing the Service, signing an Order Form, or clicking to accept, you confirm that you have read and agree to these Terms and that you have authority to bind your organization. If you and Cogneris have entered into a separate written Master Services Agreement, Order Form, or similar contract that addresses the Service (a "Negotiated Agreement"), the Negotiated Agreement controls in the event of any conflict with these Terms. ## 2. Definitions - Authorized User — an individual employee, contractor, or agent of Customer who is authorized by Customer to use the Service under Customer's account. - Customer Data — all data, content, documents, and files that Customer or an Authorized User uploads to, generates within, or transmits through the Service, including extracted fields and outputs derived from the foregoing. - Documentation — the user guides, API references, and technical documentation that Cogneris makes available for the Service. - Order Form — an ordering document or online subscription flow specifying the Service plan, fees, and term. - Subscription Term — the period during which Customer is authorized to use the Service, as set out in the applicable Order Form or plan. ## 3. Eligibility and accounts The Service is intended for use by businesses. You must be at least 18 years old and legally able to enter into contracts to use the Service. You are responsible for the accuracy and completeness of registration information, for keeping account credentials confidential, for the activities of your Authorized Users, and for notifying us promptly of any suspected unauthorized access. ## 4. The Service Subject to these Terms and payment of applicable fees, Cogneris grants Customer a non-exclusive, non-transferable, non-sublicensable, worldwide right during the Subscription Term to access and use the Service for Customer's internal business purposes. Cogneris may update the Service from time to time, provided that material reductions in core functionality during a paid Subscription Term will be subject to reasonable notice and, where applicable, the customer's contractual remedies. ## 5. Subscriptions, fees, and taxes Fees are set out in the applicable Order Form, plan page, or other ordering process. Unless otherwise stated, fees are payable in advance, are non-refundable, and are exclusive of taxes, duties, and similar governmental charges, all of which Customer is responsible for. We may suspend access for non-payment after providing written notice and a reasonable opportunity to cure. We may change pricing for renewal terms on at least 30 days' prior notice before the renewal date. ## 6. Customer Data ### 6.1 Ownership and license to Cogneris As between the parties, Customer owns and retains all rights, title, and interest in and to Customer Data. Customer grants Cogneris a limited, worldwide, royalty-free license to host, copy, transmit, display, process, and otherwise use Customer Data solely as necessary to provide, secure, and support the Service, to comply with legal obligations, and to enforce these Terms. ### 6.2 No training on Customer Data Cogneris does not use Customer Data to train, fine-tune, or evaluate generally-available foundation models or any model made available to other customers. Cogneris may use de-identified, aggregated metadata about system performance (such as latency, error rates, and feature usage) to operate and improve the Service. Where the Service uses third-party model providers, Cogneris contractually requires that Customer Data sent to such providers is not used to train their models. ### 6.3 Customer responsibilities Customer represents and warrants that: - Customer has all rights, consents, and authority necessary to upload Customer Data to the Service and to grant the license in Section 6.1; - Customer Data and Customer's use of the Service comply with all applicable laws, including data protection, export control, and intellectual property laws; - Customer is responsible for the accuracy, quality, and legality of Customer Data; - Customer will not upload to the Service any data subject to special regulatory regimes (such as protected health information under HIPAA, payment card data subject to PCI-DSS in raw form, classified information, or material protected under export control regulations) unless the parties have agreed in writing on additional terms permitting such use, including any required Business Associate Agreement. ## 7. Acceptable use Customer and Authorized Users will not, and will not permit any third party to: - access or use the Service in violation of any law or third-party right; - upload content that is unlawful, defamatory, infringing, or that contains malware; - use the Service to compete with Cogneris, to build a competing product, or to benchmark for publication without our written consent; - reverse engineer, decompile, or disassemble the Service, except to the extent expressly permitted by applicable law; - circumvent technical protection measures, rate limits, or access controls; - interfere with the integrity, performance, or security of the Service or the data of any third party; - use the Service to make decisions that produce legal or similarly significant effects on individuals without appropriate human review; - resell, sublicense, or make the Service available to third parties except as expressly permitted in an Order Form. ## 8. Privacy and data protection Cogneris processes personal information in accordance with our [Privacy Policy](privacy.html). Where Cogneris processes personal data on behalf of Customer in providing the Service, the processing is governed by our Data Processing Addendum (DPA), which is incorporated into these Terms by reference and is available at [data-protection.html](data-protection.html) or on request to [legal@cogneris.ai](mailto:legal@cogneris.ai). The DPA includes the Standard Contractual Clauses where applicable to international transfers. ## 9. Security Cogneris maintains a written information security program with administrative, technical, and physical safeguards designed to protect Customer Data, as further described on our [Security](security.html) page. Cogneris will notify Customer without undue delay after becoming aware of a confirmed unauthorized acquisition of, access to, or disclosure of Customer Data. ## 10. Service levels and support Cogneris provides standard support during business hours through documented channels. Service-level commitments (uptime, response times, and credits), if any, are set out in the applicable plan or Order Form. Beta or preview features may be made available from time to time and are provided "as is" without service-level commitments. ## 11. Intellectual property Cogneris and its licensors retain all right, title, and interest in and to the Service, including all software, models, prompts, templates, documentation, and trademarks, and any improvements, derivatives, or modifications thereof. No rights are granted except as expressly set out in these Terms. ### 11.1 Feedback If Customer or any Authorized User provides feedback, suggestions, or ideas about the Service ("Feedback"), Customer grants Cogneris a perpetual, irrevocable, worldwide, royalty-free license to use, modify, and incorporate the Feedback into the Service without restriction or obligation. ## 12. Confidentiality Each party may disclose confidential information to the other in connection with the Service. The receiving party will use confidential information only to exercise rights and perform obligations under these Terms, will protect it with the same care it uses to protect its own confidential information of similar importance (and in no event less than reasonable care), and will not disclose it except to its representatives bound by confidentiality obligations. These obligations do not apply to information that is or becomes public through no fault of the receiving party, was lawfully known before disclosure, is independently developed without use of the confidential information, or is rightfully obtained from a third party. The receiving party may disclose confidential information to comply with law or legal process, provided that, where permitted, it gives the disclosing party prompt notice. ## 13. Third-party services The Service may interoperate with third-party services that Customer chooses to enable. Cogneris is not responsible for those third-party services, and Customer's use of them is governed by the third party's terms. ## 14. Term, suspension, and termination These Terms apply from the date of acceptance and continue for the Subscription Term, renewing as set out in the Order Form or plan. Either party may terminate for the other's material breach not cured within 30 days of written notice. Cogneris may suspend access immediately if Customer's use poses a security or legal risk to Cogneris, the Service, or other customers, or for non-payment after notice. On termination, Customer's access to the Service will cease, and Cogneris will delete or return Customer Data in accordance with the DPA and the applicable Order Form. Provisions that by their nature should survive termination will survive. ## 15. Warranty disclaimer Except as expressly set out in these Terms or in a Negotiated Agreement, the Service is provided "AS IS" and "AS AVAILABLE". To the maximum extent permitted by law, Cogneris disclaims all warranties, whether express, implied, statutory, or otherwise, including warranties of merchantability, fitness for a particular purpose, title, non-infringement, and any warranty arising out of course of dealing or usage of trade. Cogneris does not warrant that the Service will be uninterrupted, error-free, or that extraction outputs will be free from inaccuracies. Customer is responsible for reviewing outputs before relying on them for any significant decision. ## 16. Limitation of liability To the maximum extent permitted by law: - Neither party will be liable for any indirect, incidental, special, consequential, exemplary, or punitive damages, or for lost profits, lost revenue, lost data, or business interruption, arising out of or related to these Terms, even if advised of the possibility of such damages. - Each party's aggregate liability arising out of or related to these Terms will not exceed the amounts paid or payable by Customer to Cogneris for the Service during the twelve (12) months preceding the event giving rise to the liability. The above limitations do not apply to: (a) Customer's payment obligations; (b) either party's breach of confidentiality obligations; (c) Customer's breach of Section 7 (Acceptable use); (d) either party's indemnification obligations; or (e) liability that cannot be limited under applicable law (such as fraud, gross negligence, willful misconduct, or, in some jurisdictions, personal injury or death). ## 17. Indemnification By Cogneris. Cogneris will defend Customer against any third-party claim alleging that the Service, used in accordance with these Terms, infringes the third party's intellectual property right, and will pay damages and costs finally awarded against Customer or agreed in settlement, provided that Customer promptly notifies Cogneris, gives Cogneris sole control of the defense, and reasonably cooperates. Cogneris has no liability for claims arising from Customer Data, modifications not made by Cogneris, combinations with non-Cogneris products, or use after Cogneris has provided a non-infringing alternative. By Customer. Customer will defend Cogneris against any third-party claim arising from (i) Customer Data; (ii) Customer's breach of Section 6.3 or Section 7; or (iii) Customer's violation of law in connection with the Service, and will pay damages and costs finally awarded against Cogneris or agreed in settlement, subject to the same procedural conditions as above. ## 18. Publicity Cogneris may identify Customer as a Cogneris customer and use Customer's name and logo on our website and in marketing materials, subject to Customer's brand guidelines. Customer may opt out by notice to [legal@cogneris.ai](mailto:legal@cogneris.ai). Other public announcements require the parties' written agreement. ## 19. Modifications Cogneris may modify these Terms from time to time. If we make material changes, we will provide reasonable notice (for example, by email or in-product notice) before the changes take effect. Continued use of the Service after the effective date of the updated Terms constitutes acceptance of the changes. If Customer does not accept the changes, Customer's sole remedy is to stop using the Service and, where applicable, to terminate any unexpired paid Subscription Term and receive a pro-rata refund of unused, prepaid fees. ## 20. Governing law and disputes These Terms are governed by the laws of the State of Delaware, United States, excluding its conflict-of-laws rules and the United Nations Convention on Contracts for the International Sale of Goods. The parties submit to the exclusive jurisdiction of the state and federal courts located in Delaware for any dispute that is not subject to arbitration, except that either party may seek injunctive or equitable relief in any court of competent jurisdiction to protect its intellectual property or confidential information. Where required by local law, consumers retain the protections and forum rights granted to them by the law of their country of residence. ## 21. General - Entire agreement. These Terms, together with any Order Form, the DPA, and the Privacy Policy, are the entire agreement between the parties regarding the Service and supersede all prior or contemporaneous understandings on that subject. - Order of precedence. In a conflict: (1) a Negotiated Agreement signed by both parties; (2) an Order Form; (3) the DPA; (4) these Terms; (5) other policies referenced herein. - Assignment. Neither party may assign these Terms without the other's prior written consent, except that either party may assign to an affiliate or successor in connection with a merger, reorganization, or sale of substantially all assets, with notice to the other party. - Force majeure. Neither party is liable for failure or delay caused by events beyond its reasonable control. - No waiver. Failure to enforce a provision is not a waiver of the right to do so later. - Severability. If any provision is held unenforceable, the remaining provisions remain in effect. - Independent contractors. The parties are independent contractors. These Terms do not create a partnership, joint venture, or agency. - Notices. Notices to Cogneris must be sent to [legal@cogneris.ai](mailto:legal@cogneris.ai). Notices to Customer may be sent to the email associated with Customer's account. - Export and sanctions. Customer will comply with applicable export control and sanctions laws and will not access the Service from a jurisdiction subject to comprehensive sanctions. - U.S. government end users. The Service is "commercial computer software" as defined in the FAR and DFARS, licensed under these Terms. ## 22. Contact Questions about these Terms? Contact us at [legal@cogneris.ai](mailto:legal@cogneris.ai) or by post: Cogneris, Inc. — Legal, 7345 W Sand Lake Rd, Ste 210, Office 3732, Orlando, FL 32819, United States. ## Source: accounts-payable-automation-api.html [Skip to content](#main) Finance automation # Accounts payable automation with an API. Cogneris turns invoice packets, receipts, purchase orders, vendor forms, and payment evidence into validated data for AP workflows and finance systems. [Talk to our team](contact.html)[See invoice extraction](extract-invoices.html) ## From invoice inbox to ERP-ready JSON AP automation is not only invoice OCR. The workflow needs invoice capture, vendor identification, line-item extraction, PO matching, duplicate checks, tax and total validation, approval routing, and an audit trail for exceptions. ### Extract Invoices, receipts, purchase orders, statements, and vendor forms. ### Validate Reconcile totals, tax, currency, dates, PO numbers, and duplicate invoices. ### Route Send clean invoices to ERP systems, flag uncertain fields with their source citation, and queue documents for review when the workflow calls for it. ## Where it fits Use Cogneris when your finance team needs structured AP data but your product or operations team also needs API control, webhooks, tenant boundaries, and evidence for audits. ## AP workflow schema A production AP automation API should return more than header fields. Cogneris can return invoice lines, tax and freight values, PO references, payment terms, vendor identifiers, duplicate signals, validation errors, review state, and source citations in the same response. | Workflow step | Fields and checks | Handoff | | Invoice capture | Vendor, invoice number, dates, totals, tax, currency, line items, remittance details. | ERP draft, AP inbox, or review queue. | | PO matching | PO number, buyer, seller, quantities, unit prices, received goods, tolerance rules. | Matched approval or exception task. | | Exception review | Low confidence fields, duplicate invoices, mismatched totals, missing tax IDs, policy failures. | Reviewer decision with audit record. | | ERP export | Normalized JSON, webhook event, validation status, source evidence, retry metadata. | SAP, NetSuite, Oracle, Workday, Coupa, or custom finance systems. | ## ERP and e-invoicing coverage AP teams increasingly need both unstructured document extraction and structured e-invoicing workflows. Cogneris can handle email PDFs, scanned invoices, attachments, purchase orders, PEPPOL exceptions, and vendor documents before sending approved data downstream. ## Related pages [Invoice matching automation](invoice-matching-automation-api.html)[E-invoicing automation](e-invoicing-automation.html)[PEPPOL document AI](peppol-e-invoicing-document-ai.html)[Invoice extraction](extract-invoices.html)[Purchase order OCR API](purchase-order-ocr-api.html)[SAP document extraction](sap-document-extraction.html)[Accounting firms](solutions-accounting.html)[Finance automation](solutions-finance-automation.html)[Rossum comparison](compare-rossum.html)[Docsumo comparison](compare-docsumo.html) ## Source: bank-statement-converter-api.html [Skip to content](#main) Statements to JSON # Bank statements to JSON in one API call. Send a bank or credit-card statement, PDF or scan, and get back normalized transactions, balances and account metadata as JSON, with every value linked to the page it came from. 400 free credits, no card. [Get 400 free credits](https://login.cogneris.ai/signup)[Read the quickstart](https://docs.cogneris.ai/docs/getting-started/api-quickstart/) No credit card. 400 credits for 14 days. ## Convert PDFs into usable transaction data Bank statement conversion is more than OCR. Cogneris extracts opening and closing balances, account holder, institution, account number, transaction dates, descriptions, amounts, fees, recurring deposits, and page-level evidence. ### PDF to JSON Return nested statement data for lending, finance automation, and customer portals. ### PDF to CSV Export normalized transactions for spreadsheet review or reconciliation workflows. ### Validation checks Check balance math, duplicate pages, date ranges, and low-confidence line items. ## Use cases Use the converter for cash-flow underwriting, mortgage intake, bank statement reconciliation, tenant screening, merchant onboarding, accounting workflows, and embedded finance applications that need reliable transaction-level data from PDFs or scans. ## What it costs A 5-page statement, classified and extracted, uses about 9 credits, so the 400 free credits cover roughly 40 statements. The Free plan lasts 14 days, needs no card and stops at zero credits: nothing is billed. ### Free €0 for 14 days. 400 credits, no credit card. Processing stops when the credits run out. ### Pro €99/month. 5,000 credits a month, then €0.03 per extra credit. Billed in euros for EU accounts. [Get 400 free credits](https://login.cogneris.ai/signup)[See all plans](pricing.html) ## Related pages [Bank statement extraction](extract-bank-statements.html)[Financial spreading](financial-spreading-document-ai.html)[Lending portal](lending-document-portal.html)[PDF data extraction](pdf-data-extraction-api.html)[Documents to JSON](convert-documents-to-json-api.html) ## Source: check-ocr-api.html [Skip to content](#main) Payments # Check OCR API for payment ops. Extract structured data from check images, remittance documents, and payment packets with confidence scores, validation state, and audit metadata. [Talk to our team](contact.html)[OCR API](ocr-api.html) ## From check image to usable payment data Check processing needs more than text recognition. Cogneris can extract payer, payee, numeric amount, written amount, date, memo, endorsement context, routing hints, remittance references, and source evidence for review. ### Field extraction Return check fields as typed JSON with confidence and document-region evidence. ### Validation Compare amount fields, detect missing dates, flag unreadable regions, and route exceptions. ### Payment workflow Use webhooks to connect check OCR to AR, treasury, lockbox, or customer-service workflows. ## Remote deposit capture and MICR extraction Remote deposit capture workflows need more than the written amount. Cogneris can extract MICR context, routing and account hints, check number, payer and payee, date, front/back image evidence, and endorsement signals so payment teams can validate the deposit before posting. ## Duplicate checks and bank validation terms For lockbox, AR, treasury, and customer portals, check OCR should flag duplicate check numbers, mismatched written and numeric amounts, missing endorsements, stale dates, unreadable MICR regions, and bank-validation fields that need a human review. ## Where it fits Use check OCR for remote deposit capture workflows, remittance processing, insurance payments, lending operations, nonprofit donations, property-management checks, bank validation queues, and customer payment portals. ## Check OCR benchmark fields Evaluate check OCR APIs on MICR readability, routing number, account hint, check number, payer, payee, numeric amount, written amount, memo, date, signature and endorsement evidence, duplicate detection, and amount reconciliation before routing results into payment operations. ## Related pages [OCR API](ocr-api.html)[Extraction API](document-extraction-api.html)[Bank statements](extract-bank-statements.html)[Banking](solutions-banking.html)[Finance automation](solutions-finance-automation.html) ## Source: claims-document-processing-api.html [Skip to content](#main) Insurance document AI # Claims document processing for review teams. Cogneris extracts insurance claims packets into structured data with document classification, field confidence, evidence links, and reviewer routing. [Talk to our team](contact.html)[See insurance workflows](solutions-insurance.html) ## Claims arrive as messy packets A single claim can include forms, ACORD documents, invoices, estimates, medical attachments, photos, policy pages, EOBs, emails, and proof documents. Cogneris classifies the packet before extracting the fields that matter. ## Extraction with audit trail Claim number, policy number, claimant, loss date, incident details, service lines, invoice totals, coverage references, repair estimates, and missing-document status can all be returned with source evidence. ## Where it fits Use Cogneris to triage incoming claims, prepare reviewer work queues, validate supporting evidence, and push clean data into claims or policy systems. ### Extract Turn PDFs, scans, photos, and document packets into structured records. ### Validate Apply confidence thresholds, business rules, and cross-document checks. ### Route Send clean results downstream flag uncertain fields with their source citation, and queue documents for review when the workflow calls for it. ## Related pages [EOB OCR API](eob-ocr-api.html)[Insurance solutions](solutions-insurance.html)[ACORD extraction](extract-acord-forms.html)[Invoice extraction](extract-invoices.html)[Fraud docs](https://docs.cogneris.ai/docs/api/fraud-verify/) ## Source: document-anonymization-api.html [Skip to content](#main) Privacy and controls # Document anonymization with audit trail. Reduce sensitive document exposure by redacting, masking, or extracting only the fields a workflow is allowed to use. [Data protection](data-protection.html)[Security](security.html) ## Least-data extraction for sensitive workflows Document automation often handles IDs, bank statements, claims, medical invoices, addresses, tax forms, and contracts. Cogneris can limit outputs to approved fields, route sensitive values through review, and keep an audit trail of what was extracted or masked. ### PII redaction Mask names, addresses, account numbers, tax IDs, health identifiers, or tenant-defined sensitive fields before sharing outputs. ### Policy controls Apply tenant-specific retention, access, validation, and review rules for GDPR, HIPAA, and internal data-handling policies. ### Audit evidence Log source pages, export decisions, and model or rule versions for compliance review. ## Document anonymization is not just black boxes Some workflows need true redaction in a derived file. Others need data anonymization through selective extraction so downstream systems never receive unnecessary personal data. The stronger pattern is to define the minimum field set, validate it, and keep sensitive evidence behind access controls. ## Verification and fraud workflows Anonymization often sits next to document verification. A tenant can mask health, tax, or identity fields for most users while still allowing a reviewer to inspect the original evidence when a fraud signal or validation rule requires escalation. ## Related pages [Data protection](data-protection.html)[Security](security.html)[Privacy](privacy.html)[Validation docs](https://docs.cogneris.ai/docs/product/core-concepts/)[Document verification](document-verification-api.html)[Source citations](document-ai-source-citations.html) ## Source: document-conversion-api.html [Skip to content](#main) API · conversion # Document conversion API for workflow output. Convert incoming PDFs, scans, images, and packets into structured or searchable outputs your downstream systems can actually use. [Open API reference](api.html)[Read extraction docs](https://docs.cogneris.ai/docs/api/document-extraction/) ## Convert documents without losing evidence File conversion alone produces another file to manage. Cogneris keeps conversion connected to extraction, validation, source citations, confidence scores, and webhook delivery so converted output can move into a workflow with context intact. ### Structured formats Return JSON for APIs, CSV for operations exports, XML for legacy systems, XLSX for review, and UBL for e-invoicing workflows. ### Searchable files Create searchable PDF output while preserving page references and extraction evidence for reviewers. ### ERP handoff Send converted invoice, purchase order, receipt, and vendor data to SAP, NetSuite, Dynamics, or custom ERP queues. ## When to use conversion vs extraction Use conversion when another system requires a specific file or interchange format. Use extraction when your application needs typed fields, validation status, or review decisions. Most production teams need both: converted artifacts for records and structured JSON for automation. ## Related pages [Documents to JSON API](convert-documents-to-json-api.html)[PDF data extraction](pdf-data-extraction-api.html)[SAP extraction](sap-document-extraction.html)[PEPPOL e-invoicing](peppol-e-invoicing-document-ai.html)[Extraction API](document-extraction-api.html) ## Source: document-fraud-detection-api.html [Skip to content](#main) Risk and verification # Document fraud detection with evidence. Score uploaded documents for tampering, synthetic evidence, inconsistent values, duplicates, and risky fields before extracted data reaches the workflow. [Open fraud docs](https://docs.cogneris.ai/docs/api/fraud-verify/)[Talk to our team](contact.html) ## Fraud checks need source evidence A fraud score is only useful when reviewers can see why it changed the workflow. Cogneris attaches evidence to risk signals so teams can inspect suspicious edits, mismatched fields, duplicate documents, and values that do not agree across a packet. Fraud verification is opt-in per tenant; extraction also requires template enablement. Results are advisory, and broken or unverifiable PDF signatures set a minimum review level. ### Signals with a reason Invalid CPF/CNPJ, impossible dates, PDF re-saves and metadata anomalies, error-level analysis, duplicate files, and C2PA and PDF signatures verified on the original upload. ### Cross-document checks Compare names, addresses, account numbers, dates, totals, and IDs across files. ### Review thresholds Route only high-risk documents or fields to human review. ## Where fraud detection fits Use document fraud detection in onboarding, lending, insurance claims, AP automation, tenant screening, KYC, vendor setup, and customer portals where a forged file can create downstream financial or compliance risk. ## Verification before approval Pair fraud scoring with document verification rules: expected document type, required fields, cross-document consistency, duplicate checks, expiry dates, and review thresholds. The output should explain whether a document is approved, rejected, or waiting on a reviewer. ## Related pages [Fraud detection docs](https://docs.cogneris.ai/docs/api/fraud-verify/)[Document verification API](document-verification-api.html)[KYC extraction](extract-kyc-documents.html)[Claims API](claims-document-processing-api.html)[Mortgage extraction](mortgage-document-extraction-api.html)[Source citations](document-ai-source-citations.html) ## Source: document-verification-api.html [Skip to content](#main) Validation and risk # Verify documents with evidence. Classify, extract, and verify uploaded documents before onboarding, lending, claims, AP, or portal workflows continue. [Talk to our team](contact.html)[Fraud detection](document-fraud-detection-api.html) ## Verification combines extraction and checks Document verification is the step between upload and action. Cogneris checks whether the file is the expected type, whether required fields exist, whether extracted values agree across documents, and whether the evidence should be approved, rejected, or reviewed. ### Identity and packet checks Compare names, addresses, dates, IDs, account numbers, and document types across uploads. ### Fraud and tampering signals Flag suspicious edits, duplicates, synthetic evidence, and inconsistent values. ### Review thresholds Set confidence and risk thresholds by tenant, workflow, document type, and field. ## Verification use cases Use verification for KYC document collection, proof of address, income verification, borrower packets, tenant screening, vendor onboarding, healthcare intake, and claims evidence. ## Related pages [Fraud detection API](document-fraud-detection-api.html)[Proof of address](proof-of-address-extraction-api.html)[KYC extraction](extract-kyc-documents.html)[Document anonymization](document-anonymization-api.html)[Validation docs](https://docs.cogneris.ai/docs/product/core-concepts/) ## Source: eob-ocr-api.html [Skip to content](#main) Healthcare document AI # EOB OCR API for claim review. Extract explanation-of-benefits details from payer PDFs, scans, and claim packets with field evidence, validation, and review routing. [Talk to our team](contact.html)[Healthcare workflows](solutions-healthcare.html) ## Turn EOBs into review-ready data Explanation-of-benefits documents carry dense service lines, payer decisions, member details, provider references, allowed amounts, adjustments, denial reasons, and payment responsibility. Cogneris extracts those values into a structured record with source citations. ### Claim fields Patient, provider, payer, claim number, member ID, dates of service, CPT context, diagnosis context, and policy references. ### Service lines Extract charges, allowed amounts, paid amounts, adjustments, coinsurance, deductible, denial codes, and explanation text. ### Review routing Flag low-confidence fields, missing lines, mismatched totals, or sensitive values for your healthcare operations team. ## Where it fits Use EOB OCR for claims review, patient billing support, benefits verification, revenue-cycle operations, insurance portals, and document intake workflows that need HIPAA-aware handling. ## EOB extraction fields to benchmark Compare EOB OCR APIs on patient and member fields, payer and provider details, claim number, service dates, CPT or procedure context, billed amount, allowed amount, paid amount, denial code, adjustment reason, deductible, coinsurance, and field-level citations for each service line. ## Related pages [Healthcare](solutions-healthcare.html)[Claims API](claims-document-processing-api.html)[Insurance](solutions-insurance.html)[Document anonymization](document-anonymization-api.html)[Data protection](data-protection.html) ## Source: health-insurance-card-ocr-api.html [Skip to content](#main) Healthcare document AI # Health card OCR for intake. Extract insurance-card data from uploads and scans, then validate coverage fields before onboarding, eligibility, billing, or claims workflows continue. [Talk to our team](contact.html)[Healthcare solutions](solutions-healthcare.html) ## Fields extracted from insurance cards Health insurance cards vary by payer and plan. Cogneris can extract member name, member ID, payer, plan name, group number, Rx BIN, PCN, copay values, phone numbers, effective dates, and page evidence for every field. ### Front and back cards Process photo uploads, scans, PDFs, and multi-image submissions from patient portals. ### Coverage evidence Return citations and confidence scores so staff can verify sensitive fields quickly. ### PHI controls Apply tenant access, retention, redaction, and audit controls for healthcare workflows. ## Where it fits Use health insurance card OCR during patient onboarding, benefits verification, claims intake, prior authorization support, billing setup, and healthcare customer portals. For complete medical packets, pair card extraction with EOB OCR and medical-record routing rules. ## Related pages [Healthcare document AI](solutions-healthcare.html)[EOB OCR API](eob-ocr-api.html)[Document anonymization](document-anonymization-api.html)[Data protection](data-protection.html)[Document verification](document-verification-api.html) ## Source: id-card-ocr-api.html [Skip to content](#main) KYC document AI # ID card OCR API for KYC workflows. Cogneris turns identity cards, driver licenses, passports, and proof documents into normalized fields with confidence, citations, and review state. [Talk to our team](contact.html)[See KYC extraction](extract-kyc-documents.html) ## Identity extraction without brittle templates KYC teams need consistent identity data from varied scans, mobile photos, and PDFs. Cogneris extracts names, dates, numbers, addresses, issuing authority, expiration, and MRZ data when present. ## Review-ready output Every sensitive field can carry page evidence, bounding-box location, confidence, and validation status so teams can route uncertain identity records to a reviewer. ## Where it fits Use it for customer onboarding, marketplace seller verification, embedded finance, lending portals, tenant onboarding, and regulated account opening flows. ### Extract Turn PDFs, scans, photos, and document packets into structured records. ### Validate Apply confidence thresholds, business rules, and cross-document checks. ### Route Send clean results downstream flag uncertain fields with their source citation, and queue documents for review when the workflow calls for it. ## Related pages [KYC extraction](extract-kyc-documents.html)[Banking](solutions-banking.html)[Marketplace onboarding](solutions-marketplaces.html)[Security](security.html) ## Source: invoice-matching-automation-api.html [Skip to content](#main) AP automation # Invoice matching by API. Extract invoices, purchase orders, receipts, and line items, then validate 2-way or 3-way matches before posting to the ERP. [Plan AP automation](contact.html)[AP automation API](accounts-payable-automation-api.html) ## 2-way and 3-way matching Cogneris extracts invoice headers, taxes, payment terms, PO numbers, line items, quantities, unit prices, goods receipts, and delivery evidence. Validation rules compare invoice data against purchase orders and receipts before the workflow approves or routes an exception. ### 2-way match Compare invoice lines against the purchase order, including quantities, prices, vendor, and tax rules. ### 3-way match Add goods receipt, proof of delivery, or service acceptance evidence before approval. ### Exception routing Route mismatches, missing POs, duplicate invoices, and low-confidence fields to review. ## ERP-ready output Return normalized JSON with citations, validation outcomes, reviewer state, and webhook delivery for SAP, NetSuite, QuickBooks, Dynamics, and custom AP queues. ## Related pages [Accounts payable API](accounts-payable-automation-api.html)[Invoice extraction](extract-invoices.html)[Purchase order OCR](purchase-order-ocr-api.html)[Proof of delivery OCR](proof-of-delivery-ocr-api.html)[SAP extraction](sap-document-extraction.html) ## Source: mortgage-document-extraction-api.html [Skip to content](#main) Lending automation # Mortgage document extraction for loan packets. Extract borrower, income, asset, identity, property, and condition data from mixed mortgage packets with field citations, validation checks, and review routing. [Talk to our team](contact.html)[Banking solutions](solutions-banking.html) ## One packet, many document types Mortgage workflows rarely receive a clean single file. Cogneris classifies and extracts from paystubs, W-2s, tax returns, bank statements, IDs, proof of address, purchase contracts, appraisal documents, insurance forms, and trailing conditions. ### Classify the packet Split mixed uploads into document types before extraction. ### Validate across files Compare names, dates, account holders, addresses, totals, and missing conditions. ### Route exceptions Send only uncertain or inconsistent fields to a reviewer. ## Built for underwriting checks Use structured output for income verification, asset review, KYC checks, condition clearing, fraud signals, and document completeness. Each field can include confidence, page evidence, and source citations so reviewers can inspect the value before it enters the loan system. ## Related pages [Banking automation](solutions-banking.html)[Bank statement extraction](extract-bank-statements.html)[Tax return extraction](extract-tax-returns.html)[KYC extraction](extract-kyc-documents.html)[Proof of address](proof-of-address-extraction-api.html) ## Source: proof-of-address-extraction-api.html [Skip to content](#main) Verification workflows # Proof of address extraction with evidence. Cogneris extracts and validates address evidence from utility bills, bank statements, telecom bills, lease documents, tax forms, and onboarding packets. [Talk to our team](contact.html)[See portal automation](document-automation-portal.html) ## Address verification needs more than OCR Proof-of-address workflows need the name, normalized address, issuer, statement date, document type, and enough evidence to decide whether the document is current and tied to the applicant. ## Built for portals and onboarding Use Cogneris inside upload portals to collect missing proof, classify the document, extract fields, validate freshness, and route exceptions to human review. ## Common documents Utility bills, bank statements, credit-card statements, insurance letters, lease agreements, tax notices, employer letters, and government correspondence. ## Proof-of-address fields to benchmark Compare proof-of-address extraction APIs on applicant name matching, normalized address, issuer, statement date, freshness window, document type, country-specific address formats, low-confidence routing, and cross-document validation against KYC or onboarding records. ### Extract Turn PDFs, scans, photos, and document packets into structured records. ### Validate Apply confidence thresholds, business rules, and cross-document checks. ### Route Send clean results downstream flag uncertain fields with their source citation, and queue documents for review when the workflow calls for it. ## Related pages [Document portal](document-automation-portal.html)[KYC extraction](extract-kyc-documents.html)[Bank statements](extract-bank-statements.html)[Validation docs](https://docs.cogneris.ai/docs/product/core-concepts/) ## Source: proof-of-delivery-ocr-api.html [Skip to content](#main) Logistics documents # Proof of delivery OCR with evidence. Extract delivery evidence from scanned POD forms, mobile photos, carrier PDFs, and logistics packets for reconciliation and claims workflows. [Talk to our team](contact.html)[See BOL OCR](extract-bill-of-lading.html) ## Delivery data needs proof, not just text Proof-of-delivery documents confirm what was delivered, when, where, by whom, and whether exceptions were recorded. Cogneris extracts the fields and keeps source evidence attached for carrier disputes, freight audit, retail receiving, and cargo claims. ### POD fields POD number, delivery date, shipper, consignee, address, carrier, driver, item counts, condition notes, and exception codes. ### Signature evidence Detect signature presence, signer name, stamps, handwritten notes, and low-confidence regions that need review. ### Workflow handoff Send validated JSON to TMS, ERP, claims, carrier portals, or reconciliation queues with webhook delivery. ## Where it fits Use proof-of-delivery OCR for delivery reconciliation, retail receiving, freight audit, carrier chargebacks, insurance claims, and logistics document portals where a missing or unclear POD can delay payment. ## POD extraction fields to benchmark Evaluate proof-of-delivery OCR APIs on POD number, shipment reference, delivery date, consignee, delivery address, carrier, driver, signature presence, signer name, exception notes, item counts, damaged-goods notes, and citation quality for handwritten or stamped regions. ## Related pages [Bill of lading OCR](extract-bill-of-lading.html)[Claims API](claims-document-processing-api.html)[Retail](solutions-retail.html)[Extraction API](document-extraction-api.html)[Document types](document-types.html) ## Source: purchase-order-ocr-api.html [Skip to content](#main) Procurement automation # Purchase order OCR API for clean PO data. Cogneris extracts purchase orders into ERP-ready JSON with line items, buyer and vendor details, approval status, delivery terms, and source evidence. [Talk to our team](contact.html)[See AP automation](accounts-payable-automation-api.html) ## From PO PDF to validated JSON Purchase orders are rarely isolated. They connect to invoices, receipts, vendor records, approvals, and ERP postings. Cogneris keeps those relationships visible so teams can match the purchase order against downstream documents instead of keying fields by hand. ## Fields teams usually extract PO number, buyer, vendor, bill-to and ship-to addresses, delivery date, currency, tax, totals, approval status, line items, quantities, unit prices, SKUs, GL hints, and payment terms. ## Why this page matters Competitor pages often rank for exact phrases like purchase order OCR and purchase order extraction. This page gives Cogneris a focused landing page for procurement, AP automation, ERP integration, and PO matching searches. ### Extract Turn PDFs, scans, photos, and document packets into structured records. ### Validate Apply confidence thresholds, business rules, and cross-document checks. ### Route Send clean results downstream flag uncertain fields with their source citation, and queue documents for review when the workflow calls for it. ## Related pages [Invoice extraction](extract-invoices.html)[AP automation](accounts-payable-automation-api.html)[SAP extraction](sap-document-extraction.html)[Document extraction API](document-extraction-api.html) ## Source: searchable-pdf-api.html [Skip to content](#main) OCR conversion # Searchable PDFs from scans. Convert scanned files into searchable PDFs, OCR text, Markdown, layout-aware chunks, and structured JSON from the same API pipeline. [Talk to our team](contact.html)[Document conversion](document-conversion-api.html) ## Searchable PDF plus structured output Some workflows need an OCR text layer for search and archiving. Others need JSON, citations, tables, and validation. Cogneris can produce both from one document job so teams do not run separate OCR, parsing, and extraction pipelines. ### OCR text layer Make scanned PDFs searchable while preserving the original visual file. ### Markdown and chunks Produce layout-aware text and metadata for RAG and internal search. ### JSON extraction Return typed fields, tables, barcodes, signatures, and citations when workflow automation needs data. ## Use cases Use the searchable PDF API for document archives, discovery, claims files, contracts, invoices, customer portals, scanned records, and RAG pipelines that need both human-readable files and machine-readable evidence. ## Related pages [Document conversion API](document-conversion-api.html)[OCR API](ocr-api.html)[Document AI for RAG](document-ai-for-rag.html)[Document parsing API](document-parsing-api.html)[Documents to JSON](convert-documents-to-json-api.html) ## Source: extract-acord-forms.html [Skip to content](#main) Insurance documents # ACORD forms extraction for insurance ops. Extract structured insurance data from ACORD forms, policy packets, endorsements, and supporting claim documents with citations and validation. [Start your 14-day trial](https://login.cogneris.ai/signup)[Insurance solutions](solutions-insurance.html) ## Extraction for underwriting and claims packets Insurance operations need consistent data from inconsistent packets. Cogneris can classify ACORD forms, policy declarations, endorsements, loss runs, FNOL documents, repair estimates, and medical bills before extracting fields into a single reviewable record. ### Policy fields Applicant, named insured, limits, deductibles, coverages, premiums, dates, vehicles, properties, and carriers. ### Claims context Loss dates, claim numbers, parties, damages, medical-bill context, police reports, and supporting evidence. ### Review state Route uncertain fields to reviewers and keep source citations for underwriting, claims, and audit checks. ## Where it fits Use ACORD form extraction for broker intake, MGA workflows, policy servicing, carrier submissions, claims triage, and insurance document portals. ## Related pages [Insurance](solutions-insurance.html)[Claims extraction](extract-insurance-claims.html)[Document portal](document-automation-portal.html)[Extraction API](document-extraction-api.html)[Hyperscience comparison](compare-hyperscience.html) ## Source: extract-bank-statements.html [Skip to content](#main) Use case · Bank statements # Bank statement extraction. Transaction by transaction. Cash-flow underwriting, AML, expense automation, and small-business lending all depend on bank statements being parsed accurately, transaction by transaction. Cogneris extracts every transaction, balance, and account-metadata field from US bank and credit-card statements in under three seconds — with category inference and per-line confidence. [Book a demo](contact.html) [API reference](api.html) ## Why bank statements are hard Every US bank formats statements differently. Chase, Bank of America, Wells Fargo, Capital One, Citi, US Bank, and the regional banks each have their own layout, transaction formatting, and balance presentation. Add credit-card statements and the format space doubles. Cash-flow underwriting needs every transaction parsed; OCR-based parsers hit 70-80% transaction accuracy and miss balances on multi-page statements. ## How Cogneris does it Cogneris parses statements semantically: it identifies the institution, finds the statement period, walks every transaction, normalizes amounts and dates, and reconciles ending balance against the transaction sum. Multi-page statements are stitched automatically; credit-card statements are handled with the same schema as bank statements. Transaction categories (deposit, withdrawal, ACH, wire, fee, interest) are inferred from descriptors. ### Sample extraction output doc_typeBank statement institutionJPMorgan Chase account_holderJordan T. Hall account_number****6789 statement_period2026-03-01 to 2026-03-31 opening_balanceUS$ 18,420.55 total_depositsUS$ 12,450.00 total_withdrawalsUS$ 9,234.18 ending_balanceUS$ 21,636.37 transactions47 transactions parsed balance_check✓ deposits − withdrawals = balance change avg_daily_balanceUS$ 19,820.40 nsf_count0 confidence0.98 → auto-approved ## What you get out of the box ### Every US bank, every credit card Chase, BoA, Wells, Capital One, Citi, US Bank, regionals, plus Visa / Mastercard / Amex statements. ### Transaction-level confidence Each transaction is returned with a confidence score, source page reference, and bounding box. ### Multi-page stitching 30-page statements are stitched into one logical document with transaction continuity preserved. ### Balance reconciliation Opening + deposits − withdrawals = ending. Inconsistencies are flagged before they hit your underwriting model. ## Integration patterns For lenders, fintechs, and expense-automation platforms, Cogneris provides REST APIs and signed webhooks suitable for custom integrations with cash-flow models, lender CRMs, and other downstream systems. The [bank statement converter API](bank-statement-converter-api.html) can also produce CSV-style transaction exports. ## Bank statement OCR API evaluation Benchmark bank statement extraction APIs on transaction row recall, balance reconciliation, duplicate transaction handling, statement-period detection, account-holder normalization, source citations, and whether the API can parse scanned PDFs as well as native PDFs. For lending and underwriting, auto-approval rate matters more than raw OCR accuracy. ## Compliance & trust Bank statements contain account numbers and transaction patterns. The Cogneris platform supports encryption at rest with per-tenant keys where configured. Document and audit retention are tenant-configurable and contract-specific. See our [security page](security.html) for the current control boundaries. ## Get started Pay-per-page pricing means you can [start an evaluation today](contact.html) without an annual commit. Most teams ship their first bank-statement extraction into production within a week. ## Related extractors Cogneris extracts dozens of structured document types. The closest neighbors to bank-statement extraction: - [KYC document extraction](extract-kyc-documents.html) — government IDs, proof of address, and beneficial ownership certifications for onboarding and CIP/CDD. - [Bank statement converter](bank-statement-converter-api.html) — turn statement PDFs into transaction JSON, balances, and CSV-ready exports. - [Financial spreading document AI](financial-spreading-document-ai.html) — normalize bank statements, tax returns, and borrower financials for credit workflows. - [Payroll extraction](extract-payroll.html) — pay stubs, W-2s, and 1099s for income verification and YTD reconciliation. - [Tax return extraction](extract-tax-returns.html) — 1040s, K-1s, and schedules pulled into AGI, deductions, and refund workflows. For broader context, see [the IDP buyer's guide](what-is-idp.html), the [2026 State of Document AI report](state-of-document-ai-2026.html), or [estimate ROI](roi-calculator.html) at your volume. ## Source: extract-bill-of-lading.html [Skip to content](#main) Logistics documents # Bill of lading OCR with validation. Extract shipment fields from bills of lading, freight invoices, packing lists, and delivery packets, then send clean logistics data to TMS, ERP, or claims workflows. [Talk to our team](contact.html)[Extraction API](document-extraction-api.html) ## Structured freight data from messy packets Bills of lading rarely arrive alone. Cogneris can classify multi-document logistics packets, split pages by document type, extract shipment values, and keep the source evidence attached to each field. ### Shipment fields BOL number, carrier, consignee, shipper, origin, destination, dates, weights, quantities, and references. ### Line items Extract freight classes, commodities, package counts, charges, accessorials, and table rows. ### Workflow handoff Deliver JSON over REST or webhooks with confidence, page citations, and review status. ## Use cases Use bill of lading extraction for freight audit, carrier onboarding, customs packets, cargo claims, delivery reconciliation, invoice matching, and supply-chain document portals. ## Fields logistics teams benchmark High-intent bill of lading OCR evaluations usually test BOL number, SCAC, PRO number, container number, consignee, shipper, notify party, pickup and delivery addresses, weight, dimensions, freight class, accessorial charges, and exception notes. Cogneris keeps those fields linked to page citations and review status. ## Related pages [Proof of delivery OCR](proof-of-delivery-ocr-api.html)[Document extraction API](document-extraction-api.html)[Document parsing API](document-parsing-api.html)[OCR API](ocr-api.html)[Document portal](document-automation-portal.html)[Invoice extraction](extract-invoices.html) ## Source: extract-contracts.html [Skip to content](#main) Use case · Contracts # Contract extraction. Clause by clause. Contract review often starts with recurring fields such as parties, term, renewal, governing law, liability cap, indemnity, and termination triggers. Cogneris extracts the fields configured for each contract type with source-clause citations and optional playbook comparison. [Start your 14-day trial](https://login.cogneris.ai/signup) [API reference](api.html) ## Why contracts are hard Contract extraction requires careful review. A missed liability cap, renewal date, or governing-law field can affect a legal workflow, and counterparties draft similar obligations in different language and structures. Compound sentences, definitions, and cross-references make a representative evaluation set and source-clause review essential. ## How Cogneris does it Cogneris parses contracts semantically. The classifier identifies the contract type (MSA, SOW, NDA, DPA, employment, lease, vendor agreement, amendment) and routes to a type-specific schema. The extractor walks the document end-to-end — reading definitions, resolving cross-references, distinguishing operative clauses from boilerplate, and preserving the source-clause citation for every extracted field. The validator runs cross-clause checks (e.g., is the term-length consistent across the term clause and the renewal clause) and flags deviations from your playbook with severity ranking. ### Sample extraction output doc_typeMaster Services Agreement partiesCustomer Inc. — Vendor Co. effective_date2026-04-15 initial_term3 years (cite §2.1, p. 3) renewalAuto-renew 1 year — 60-day notice (cite §2.2) governing_lawDelaware (cite §17.1) forumDelaware Court of Chancery (§17.2) liability_cap12 months fees — playbook prefers 24 (§14.3) indemnityMutual; IP carveout, broad exception (§13) termination_for_cause30-day cure (§16.1) termination_for_convenienceCustomer only, 60-day notice (§16.2) data_processingDPA referenced — Exhibit B (§10) most_favored_nationnone change_of_controltermination right — review (§19.4) deviations_vs_playbook3 medium · 1 high — see review queue confidence0.94 → ready for legal review ## What you get out of the box ### Configured contract types Workflows can be configured for MSA, SOW, NDA, DPA, sub-processor agreement, employment, lease, vendor, supply, and amendment documents, including multilingual fields. ### Citation per field Configured fields can include a page reference, section anchor, and bounding box. Click-through review presents the source clause alongside the extracted value. ### Playbook deviation flags Configure your standard positions and fallbacks per contract type; Cogneris presents identified deviations with severity and the relevant counterparty language. ### Cross-clause validation Term length, renewal trigger, notice period, governing law — Cogneris checks consistency across clauses and flags drafting errors before signature. ## Integration patterns Cogneris provides [REST APIs](api.html) and signed webhooks suitable for custom integrations with CLM, CRM, and homegrown legal systems. Vendor-specific field mapping, attachments, reviewer routing, and record write-back are scoped as customer integration work. ## Compliance & trust Contracts are among the most sensitive documents an organization handles. The Cogneris platform supports encryption at rest with per-tenant keys where configured; customer-managed keys are on the roadmap. Document and audit retention are tenant-configurable and contract-specific. Configured workflows can record tamper-evident audit metadata for extraction, review, and export events. See our [security page](security.html) for the current control boundaries. ## Get started Use the trial to evaluate contract extraction against representative agreements, your configured schema, and your legal review policy before connecting an operational workflow. ## Related extractors Cogneris extracts dozens of structured document types. The closest neighbors to contract extraction: - [Invoice extraction](extract-invoices.html) — vendor, line items, GL-code hints, and totals for AP automation. - [KYC document extraction](extract-kyc-documents.html) — government IDs, proof of address, and beneficial ownership certifications used for counterparty due diligence. - [Insurance claim extraction](extract-insurance-claims.html) — carriers, dates of loss, coverage codes, and amounts from claim packets that often reference underlying contracts. For broader context, see [the IDP buyer's guide](what-is-idp.html), the [2026 State of Document AI report](state-of-document-ai-2026.html), or [estimate ROI](roi-calculator.html) at your volume. ## Source: extract-insurance-claims.html [Skip to content](#main) Use case · Insurance claims # Insurance claim extraction. Hours, not weeks. FNOL forms, medical bills, repair estimates, police reports — claim packages arrive as multi-document bundles, each with its own format. Cogneris extracts and normalizes every supporting document into a single claim schema in under five seconds per document, ready to trigger straight-through processing or route to a claims handler. [Book a demo](contact.html) [API reference](api.html) ## Why claim packages stall A typical auto-claim package contains a FNOL form, a police report, repair estimates from one or two body shops, photos with damage descriptions, and possibly medical bills. A property claim might include the FNOL, a contractor estimate, photos, and a public adjuster's report. Each document type has a different format; manual handling means a claim sits in a queue for days. OCR-based parsers handle structured forms but break on free-text reports and unusual estimate formats. ## How Cogneris does it Cogneris parses each document type with a specialized schema and merges them into a single claim record. FNOL forms produce structured incident data. Repair estimates produce line items, parts, and totals. Medical bills produce CPT codes, charges, and provider info. Police reports produce parties, citations, and narrative sections. The merged claim record is then ready for straight-through processing rules (auto-approve, refer to handler, refer to SIU). ### Sample extraction output claim_typeAuto — collision claim_numberAC-2026-04-0188 incident_date2026-04-18 claimantRobert M. Taylor policy_numberPOL-1029384 vehicle_regABC-1234 (NY) documentsFNOL, police report, 2× repair estimates, 6 photos repair_estimate_lowUS$ 8,420.00 (Acme Auto Body) repair_estimate_highUS$ 9,180.00 (City Collision) estimated_lossUS$ 8,800.00 coverage_check✓ Comprehensive — confirmed fault_indicatorAt-fault: third party stp_eligible✓ — within auto-approve threshold confidence0.96 → straight-through processing ## What you get out of the box ### Multi-document claims FNOL, police reports, medical bills, repair estimates, photos with annotations — merged into one claim record. ### Line-item parts & labor Repair estimates parsed to part numbers, labor hours, and totals — ready for negotiation or comparison. ### Medical bill normalization CPT codes, ICD codes, charges, allowed amounts, provider info — normalized across hospital, clinic, and provider invoices. ### Straight-through-processing flags Configurable auto-approval rules (loss threshold, fraud score, document completeness) trigger STP or route to a handler. ## Integration patterns Cogneris integrates with major claims systems (Guidewire, Duck Creek, Insurity, Sapiens) via REST + webhook. The [async API](api.html) handles 50-document claim packages in under 90 seconds end-to-end and pushes the normalized claim record into your loss-adjustment workflow. ## Compliance & trust Claim documents can include PII, medical information, and accident-scene data. The Cogneris platform supports encryption at rest with per-tenant keys where configured. Document and audit retention are tenant-configurable and contract-specific. See our [security page](security.html) for the current control boundaries. ## Get started Pay-per-page pricing means you can [start an evaluation today](contact.html) without an annual commit. Most teams ship their first claim-document extraction into production within a week. ## Related extractors Cogneris extracts dozens of structured document types. The closest neighbors to insurance-claim extraction: - [KYC document extraction](extract-kyc-documents.html) — government IDs and proof of address used for claimant verification and fraud-control workflows. - [Invoice extraction](extract-invoices.html) — vendor invoices and medical bills that frequently accompany claim packets. - [Contract extraction](extract-contracts.html) — policy contracts, endorsements, and binding agreements referenced during claim adjudication. For broader context, see [the IDP buyer's guide](what-is-idp.html), the [2026 State of Document AI report](state-of-document-ai-2026.html), or [estimate ROI](roi-calculator.html) at your volume. ## Source: extract-invoices.html [Skip to content](#main) Use case · Invoices # Invoice extraction. Line by line. AP automation, three-way match, and GL-coding all start with the same problem: vendor invoices use different layouts. Cogneris extracts configured fields such as vendor, customer, line items, taxes, totals, and payment terms with per-line confidence and source references. [Start your 14-day trial](https://login.cogneris.ai/signup) [API reference](api.html) ## Why invoices are hard Every vendor designs their invoice differently. The header might call the customer "Bill To", "Sold To", or "Customer"; the line-item table can have 4 columns or 12; tax can be a single line, a per-row column, or a separate breakdown. EU VAT, US sales tax, GST, and reverse-charge invoices add more layout and validation cases that should be tested against a representative document set. ## How Cogneris does it Cogneris parses invoices semantically, not by template. The classifier recognizes invoice variants (standard, credit memo, recurring, prepayment, statement, EU VAT-style, freight bill); the extractor walks the document to find header, line items, taxes, totals, and payment terms; the validator runs arithmetic checks (line-item sum + tax = total) and surfaces discrepancies before the data hits your ERP. Multi-page invoices are stitched automatically; bundled vendor statements are split into individual invoices. ### Sample extraction output doc_typeInvoice vendorAcme Logistics LLC vendor_tax_idEIN 12-3456789 customerAcme Manufacturing invoice_numberACL-2026-04812 invoice_date2026-04-22 due_date2026-05-22 (Net 30) po_numberPO-58213 line_items9 lines parsed subtotalUS$ 18,420.00 taxUS$ 1,565.70 (8.5% NY state) totalUS$ 19,985.70 arithmetic_check✓ subtotal + tax = total three_way_match✓ PO-58213 + GRN-9914 (1 line variance < 0.5%) gl_hint5210 — Freight & Logistics duplicate_checkno duplicate found confidence0.97 → auto-approved ## What you get out of the box ### Every invoice format PDF, scan, photo, email attachment, EDI 810. US, EU VAT, UK, GST, multi-language. Standard, credit memo, recurring, prepayment, statement. ### Line-item confidence Each line returned with confidence, source page, and bounding box. Low-confidence rows route to AP review; the rest auto-post. ### Three-way match Pass PO and GRN with the invoice; Cogneris returns the match decision and any line-level discrepancies with configurable tolerance bands. ### GL-code suggestion Vendor + line description + history maps to your chart of accounts. Suggestions ship with confidence; AP keeps final approval. ## Invoice fields and JSON shape Veryfi and Mindee-style buyer searches usually ask which fields are returned. Cogneris returns the standard AP fields plus workflow fields: vendor identity, remittance data, invoice numbers, purchase orders, tax details, line items, duplicate status, match status, validation errors, citations, and reviewer state. ``` { "document_type": "invoice", "fields": { "vendor_name": { "value": "Acme Logistics LLC", "confidence": 0.99 }, "invoice_number": { "value": "ACL-2026-04812", "confidence": 0.98 }, "po_number": { "value": "PO-58213", "confidence": 0.96 }, "total": { "value": 19985.70, "currency": "USD", "confidence": 0.99 } }, "line_items": [{ "description": "Freight", "quantity": 12, "amount": 18420.00 }], "validation": { "status": "passed", "three_way_match": "partial_match" } } ``` ## API samples for developers Start with the [Node.js API sample](https://docs.cogneris.ai/docs/api/examples/), [Python API sample](https://docs.cogneris.ai/docs/api/examples/), or [REST API reference](api.html). For platform teams, the same invoice schema can be used in a document portal, monitored inbox, or AP workflow webhook. ## Integration patterns Cogneris provides [REST APIs](api.html) and signed webhooks suitable for custom integrations with ERP and AP systems. Vendor-specific field mapping, master-data lookups, attachments, reconciliation, and record write-back are scoped as customer integration work. ## Compliance & trust Invoices contain vendor banking details, customer information, and tax identifiers. The Cogneris platform supports encryption at rest with per-tenant keys where configured. Document and audit retention are tenant-configurable and contract-specific. See our [security page](security.html) for the current control boundaries. ## Get started Use the trial to evaluate invoice extraction against your own representative documents, schema, confidence thresholds, and review policy before deciding whether to connect an operational workflow. ## Related extractors Cogneris extracts dozens of structured document types. The closest neighbors to invoice extraction: - [Receipt extraction](extract-receipts.html) — itemized receipts for expense reports, T&E reconciliation, and reimbursement workflows. - [Contract extraction](extract-contracts.html) — parties, term, governing law, liability caps, and named clauses from MSAs, NDAs, and SaaS agreements. - [Bank statement extraction](extract-bank-statements.html) — cash-flow parsing and transaction-level extraction from US and Canadian statement PDFs. For broader context, see [the IDP buyer's guide](what-is-idp.html), the [2026 State of Document AI report](state-of-document-ai-2026.html), or [estimate ROI](roi-calculator.html) at your volume. ## Source: extract-kyc-documents.html [Skip to content](#main) Use case · KYC verification # KYC document verification. In seconds. Onboarding, account opening, AML refresh, age verification — KYC documents need to be verified fast and right. Cogneris extracts and verifies US driver's licenses, passports, utility bills, bank letters, and selfies in under three seconds — with MRZ checksum validation, face-match, and an immutable audit trail that satisfies regulator scrutiny. [Book a demo](contact.html) [API reference](api.html) ## Why KYC is the conversion bottleneck Onboarding flows lose 20-30% of users at the document-upload step when verification is slow, rejects valid documents, or breaks on unusual lighting. Manual review queues add hours of latency that crypto, neobank, and fintech onboarding can't afford. Regulators meanwhile expect a defensible audit trail proving every check was performed and every decision recorded — something OCR-based parsers don't provide out of the box. ## How Cogneris does it Cogneris extracts the document data (name, DOB, document number, issuing authority, expiry), validates structural integrity (MRZ checksum on passports, hologram patterns on IDs, layout integrity on US driver's licenses by state), and optionally runs a face-match against an uploaded selfie. Each step is logged with timestamps and confidence scores. The result is a single decision (pass / refer / fail) backed by a complete evidence record. ### Sample extraction output doc_typeUS Passport mrz_line_1P6 months mrz_check✓ all check digits valid face_match✓ 0.94 against uploaded selfie liveness✓ passive liveness pass decisionPASS — onboarding unlocked ## What you get out of the box ### All US ID document types US passports, driver's licenses (all 50 states), state IDs, utility bills, bank letters, selfies for face-match. ### Structural integrity checks MRZ checksum on passports, layout fingerprint match for state IDs, hologram pattern detection on enhanced IDs. ### Face-match & liveness Optional face-match against uploaded selfie + passive liveness detection. Bound by the same audit trail as extraction. ### Regulator-ready audit Every step (extraction, integrity, face-match, decision) logged with confidence scores and timestamps. Exportable for regulator review. ## Integration patterns Cogneris provides REST APIs and signed webhooks suitable for custom integrations with onboarding and verification workflows. API and webhook samples are available for common development stacks; vendor-specific mobile or KYC integration work must be scoped separately. ## Compliance & trust KYC documents can include identifiers and biometric data. The Cogneris platform supports encryption at rest with per-tenant keys where configured. Document, biometric, and audit retention must be defined for the tenant and contract; do not assume a default period. See our [security page](security.html) for the current control boundaries. ## Get started Pay-per-page pricing means you can [start an evaluation today](contact.html) without an annual commit. Most teams ship their first KYC extraction into production within a week. ## Related extractors Cogneris extracts dozens of structured document types. The closest neighbors to KYC document extraction: - [ID card OCR API](id-card-ocr-api.html) — driver licenses, passports, and identity fields with source evidence for onboarding and KYC portals. - [Proof of address extraction](proof-of-address-extraction-api.html) — utility bills, statements, and address evidence for CIP/CDD checks. - [Bank statement extraction](extract-bank-statements.html) — transaction-level parsing of US and Canadian statements for cash-flow underwriting and source-of-funds checks. - [Insurance claim extraction](extract-insurance-claims.html) — carriers, dates of loss, and amounts from claim packets where claimant identity verification overlaps with KYC controls. - [Contract extraction](extract-contracts.html) — beneficial-ownership-related agreements and entity formation documents that arrive alongside CIP/CDD packages. For broader context, see [the IDP buyer's guide](what-is-idp.html), the [2026 State of Document AI report](state-of-document-ai-2026.html), or [estimate ROI](roi-calculator.html) at your volume. ## Source: extract-payroll.html [Skip to content](#main) Use case · Payroll & W-2 # Payroll & W-2 extraction. In seconds. Income verification, employment proof, KYC — the workflows depend on payroll docs being structured fast and right. Cogneris extracts gross pay, net pay, employer, employment dates, withholdings, and YTD figures from US payslips, W-2s, paystubs, and 1099s in under three seconds, with a complete audit trail behind every extraction. [Book a demo](contact.html) [API reference](api.html) ## Why payroll docs are uniquely hard US payroll documents come in dozens of layouts because every payroll provider — ADP, Paychex, Gusto, Workday, Rippling, Justworks — formats them slightly differently. The fields that matter for downstream workflows (gross, net, federal/state withholdings, FICA, YTD, employer EIN) live in different places across formats. Manual data entry is slow and error-prone; rigid template-based OCR breaks the moment a customer's payroll provider updates their layout. ## How Cogneris does it Cogneris uses a multi-LLM ReAct architecture that recognizes payroll documents semantically rather than positionally. The platform identifies the document type (W-2 vs 1099 vs paystub vs verification letter), extracts the canonical fields into a normalized schema, and validates them against expected ranges and consistency rules. The result is the same JSON shape regardless of which payroll provider produced the original PDF. ### Sample extraction output doc_typeW-2 (Wage and Tax Statement) tax_year2025 employee_nameJordan T. Hall ssn***-**-1234 employer_ein12-3456789 employer_nameAcme Industries, Inc. wages_box_1US$ 124,580.00 federal_tax_withheldUS$ 22,420.00 ss_wages_box_3US$ 124,580.00 ss_tax_withheld_box_4US$ 7,724.00 medicare_wages_box_5US$ 124,580.00 medicare_tax_withheldUS$ 1,806.00 stateNY state_wagesUS$ 124,580.00 state_tax_withheldUS$ 7,815.00 confidence0.99 → auto-approved ## What you get out of the box ### Provider-agnostic Same schema for ADP, Paychex, Gusto, Workday, Rippling, Justworks — and any provider you haven't seen yet. ### All US payroll docs W-2, 1099-NEC, 1099-MISC, 1099-K, paystubs, employment verification letters, offer letters. ### YTD & period parity Cross-checks YTD figures against period totals and flags inconsistencies before they hit your underwriting model. ### PII-aware audit trail SSN and tax-ID values are redacted in audit metadata by default. Full extraction is retained per your retention policy. ## Integration patterns The [REST API](api.html) accepts document uploads and supports async processing. API and signed-webhook samples are available for common development stacks, and the same interfaces are suitable for custom integrations with HRIS, CRM, and workflow systems. ## Compliance & trust Payroll documents contain SSN, employer ID, and other regulated identifiers. The Cogneris platform supports encryption at rest with per-tenant keys where configured. Document and audit retention are tenant-configurable and contract-specific. See our [security page](security.html) for the current control boundaries. ## Get started Pay-per-page pricing means you can [start an evaluation today](contact.html) without an annual commit. Most teams ship their first payroll extraction into production within a week. ## Related extractors Cogneris extracts dozens of structured document types. The closest neighbors to payroll extraction: - [Bank statement extraction](extract-bank-statements.html) — transaction parsing that pairs with payroll for cash-flow underwriting and stated-income reconciliation. - [Tax return extraction](extract-tax-returns.html) — 1040s, K-1s, and schedules used alongside W-2/1099 data for income verification. - [KYC document extraction](extract-kyc-documents.html) — government IDs and proof of address that complete the onboarding packet around payroll-based income proof. For broader context, see [the IDP buyer's guide](what-is-idp.html), the [2026 State of Document AI report](state-of-document-ai-2026.html), or [estimate ROI](roi-calculator.html) at your volume. ## Source: extract-receipts.html [Skip to content](#main) Use case · Receipts # Receipt extraction. From any photo. Expense reports break on the same edge cases every time: faded thermal print, glare on a smartphone photo, foreign currencies, and itemized restaurant tabs that nobody wants to type in. Cogneris extracts the merchant, line items, taxes, and totals from any receipt format in under 2 seconds — and pushes the result straight to Concur, Brex, or Ramp. [Book a demo](contact.html) [API reference](api.html) ## Why receipts are hard Receipts are the worst-quality input in the document-AI pipeline. Thermal paper fades within months. Smartphone photos arrive crumpled, rotated, or with a finger in the frame. Restaurant receipts split lines across tax, tip, and total in unpredictable orders. International receipts add currencies, languages, and tax regimes most parsers don't handle. OCR-only systems sit at 75–80% field accuracy on receipts and collapse on damaged images — which means employees spend their post-trip evenings re-typing data into expense forms. ## How Cogneris does it Cogneris's classifier identifies the receipt type (retail, restaurant, taxi, hotel folio, fuel, parking, freight) and routes to a specialist extractor. Image preprocessing handles glare, rotation, perspective skew, and partial folds. The extractor parses merchant identity, line items, tax breakdown by rate, payment method, and timestamps with per-field confidence. The validator runs arithmetic checks (line items + tax + tip = total) and flags receipts that don't reconcile before they hit your expense system. ### Sample extraction output doc_typeReceipt — Restaurant merchantTrattoria del Borgo merchant_addressVia Roma 42, 50125 Firenze, IT vat_idIT 03152470485 transaction_date2026-04-18 20:42 line_items6 items parsed subtotalEUR 84.00 taxEUR 8.40 (10% IVA) tipEUR 7.60 totalEUR 100.00 currencyEUR payment_methodVisa ****4421 arithmetic_check✓ subtotal + tax + tip = total duplicate_checkno duplicate found policy_flagsnone — within meal limit confidence0.96 → auto-categorized ## What you get out of the box ### Every receipt type Retail, restaurant, taxi, ride-share, hotel folio, fuel, parking, freight. Thermal print, smartphone photo, email PDF, screenshot. ### Damaged-image resilience Faded thermal paper, glare, rotation, perspective skew, partial folds. Field-level confidence drops on damage so AP can route for review. ### Multi-language & multi-currency 30+ languages with the original text preserved alongside an English-normalized field set. Currency stays original — your policy engine converts. ### Duplicate detection Merchant + amount + timestamp + last-4 of card. Catches resubmissions and the same-meal-two-attendees pattern with the differences surfaced. ## Receipt fields and JSON shape Receipt OCR API comparisons usually focus on field coverage and mobile-photo resilience. Cogneris returns merchant details, transaction date/time, line items, tax breakdown, tip, total, payment method, duplicate status, policy flags, confidence, and citations. ``` { "document_type": "receipt", "fields": { "merchant_name": { "value": "Trattoria del Borgo", "confidence": 0.98 }, "transaction_date": { "value": "2026-04-18T20:42:00", "confidence": 0.96 }, "total": { "value": 100.00, "currency": "EUR", "confidence": 0.99 }, "payment_last4": { "value": "4421", "confidence": 0.94 } }, "line_items": [{ "description": "Dinner", "amount": 84.00 }], "policy_flags": [] } ``` ## API samples for developers Use the [Node.js API sample](https://docs.cogneris.ai/docs/api/examples/), [Python API sample](https://docs.cogneris.ai/docs/api/examples/), or [REST API reference](api.html) to integrate receipt data with expense apps, card matching, or a custom reimbursement portal. ## Integration patterns Cogneris provides [REST APIs](api.html) and signed webhooks suitable for custom integrations with expense platforms, card workflows, and ERPs. Vendor-specific mapping, attachment handling, reconciliation, and write-back behavior are scoped as customer integration work. ## Compliance & trust Receipts contain payment-card data and personal expense patterns. The Cogneris platform supports encryption at rest with per-tenant keys where configured. Document and audit retention are tenant-configurable and contract-specific. See our [security page](security.html) for the current control boundaries. ## Get started Pay-per-page pricing means you can [start an evaluation today](contact.html) without an annual commit. Most teams ship their first receipt extraction into production within a week and reach steady-state accuracy on their employee-receipt mix in under 30 days. ## Related extractors Cogneris extracts dozens of structured document types. The closest neighbors to receipt extraction: - [Invoice extraction](extract-invoices.html) — vendor, line items, GL-code hints, and totals for AP automation. - [Bank statement extraction](extract-bank-statements.html) — transaction-level parsing used to reconcile expense reports against card and account statements. - [Payroll extraction](extract-payroll.html) — pay stubs and W-2s often grouped with receipts in employee reimbursement and onboarding workflows. For broader context, see [the IDP buyer's guide](what-is-idp.html), the [2026 State of Document AI report](state-of-document-ai-2026.html), or [estimate ROI](roi-calculator.html) at your volume. ## Source: extract-tax-returns.html [Skip to content](#main) Use case · US tax returns # US tax return extraction. Line-level fidelity. Lending, mortgage, audit, and high-net-worth underwriting all depend on US tax returns being parsed accurately, line by line. Cogneris extracts every numeric line of a 1040, every form 1099, and the standard schedules (C, E, K-1) into a normalized schema in under five seconds — with a confidence score per field and a complete audit trail. [Book a demo](contact.html) [API reference](api.html) ## Why US tax returns are a pain A complete US tax return is rarely one form. It's a 1040 plus an unpredictable mix of schedules (A, B, C, D, E, K-1) and 1099s (NEC, MISC, INT, DIV, B, R) — sometimes 30+ pages of forms with field codes that humans can read at a glance and OCR systems consistently misread. Lenders and underwriters need every line accurate; a misread Schedule C net profit is a denial. Manual data entry takes hours per return; legacy OCR systems hit acceptable accuracy on the 1040 itself but fall over on schedules. ## How Cogneris does it Cogneris extracts US tax forms into a flat normalized schema keyed by IRS line number — so `line_22_total_income` on a 1040 is the same field whether the customer used TurboTax, H&R Block, or paper-filed. Every line is returned with a confidence score, a bounding box reference back to the source page, and a validation result against IRS arithmetic rules. Schedule attachments are parsed and linked to their parent form automatically. ### Sample extraction output doc_typeForm 1040 (2024) filing_statusMarried filing jointly primary_taxpayerJordan T. Hall ssn***-**-1234 agi_line_11US$ 287,420.00 taxable_income_line_15US$ 256,140.00 total_tax_line_24US$ 51,228.00 federal_tax_withheld_line_25aUS$ 48,150.00 refund_line_34US$ 0.00 amount_owed_line_37US$ 3,078.00 schedules_attachedSchedule A, Schedule B, Schedule D, Schedule E forms_10991099-INT, 1099-DIV, 1099-B (×2) arithmetic_check✓ all lines balance confidence0.97 → auto-approved ## What you get out of the box ### Every line, every schedule Full 1040, Schedules A/B/C/D/E/K-1, and the 1099 family. Returned as a flat JSON keyed by IRS line numbers. ### Arithmetic validation IRS line arithmetic rules are checked automatically. Inconsistencies are flagged before the document reaches your model. ### Multi-year stitching Process 2 or 3 years of returns at once and get a stitched view ready for underwriting (income trend, AGI history, deduction patterns). ### PII-aware audit trail Retention is tenant-configurable and contract-specific; confirm redaction and retention behavior for the configured workflow. ## Integration patterns For mortgage lenders and underwriters, Cogneris provides async processing, REST APIs, and signed webhooks suitable for custom integrations with underwriting models and lender CRMs. ## Compliance & trust Tax returns are highly sensitive. The Cogneris platform supports encryption at rest with per-tenant keys where configured. Document and audit retention are tenant-configurable and contract-specific. See our [security page](security.html) for the current control boundaries. ## Get started Pay-per-page pricing means you can [start an evaluation today](contact.html) without an annual commit. Most teams ship their first tax-return extraction into production within a week. ## Related extractors Cogneris extracts dozens of structured document types. The closest neighbors to tax-return extraction: - [Payroll extraction](extract-payroll.html) — W-2s, 1099s, and pay stubs that reconcile against 1040 and schedule data. - [Bank statement extraction](extract-bank-statements.html) — transaction-level parsing for interest income, dividend, and cash-flow validation. - [KYC document extraction](extract-kyc-documents.html) — government IDs and proof of address used alongside tax returns for full client onboarding. For broader context, see [the IDP buyer's guide](what-is-idp.html), the [2026 State of Document AI report](state-of-document-ai-2026.html), or [estimate ROI](roi-calculator.html) at your volume. ## Source: solutions-accounting.html [Skip to content](#main) Accounting firms # Document AI for accounting firms. Close more clients. Tax season throughput is gated by how fast your team keys W-2s, 1099s, K-1s, and brokerage statements into Lacerte, UltraTax, ProSeries, or Drake. Every transposition error is a downstream amendment risk. Cogneris extracts the full client packet the moment it arrives — form-by-form, line-by-line — with field-level provenance back to the source PDF. [Book a demo ](contact.html) [Security & compliance](security.html) The challenge ## Tax-season volume crushes manual data entry — and every keyed figure is an amendment waiting to happen. ### Tax season document deluge Clients arrive with bundles of W-2s, 1099s, K-1s, brokerage statements, and receipts. Manual sorting and data entry caps your throughput and squeezes margins on every 1040. ### Manual entry into tax software Lacerte, UltraTax, ProSeries, Drake. Each form keyed by hand. Each transposition error a downstream amendment risk. ### Audit-ready workpaper indexing Review and audit engagements demand traceable source documentation. Building that index from a shared drive of PDFs costs billable hours that don't bill. How Cogneris helps ## Client packets, structured. Tax software, populated. Cogneris sorts inbound 1040 packets the moment they hit the portal, extracts every form to the line, and pushes the result directly into Lacerte, UltraTax, ProSeries, Drake, or CCH Axcess — with a click-through audit trail from every figure on the return back to the page it came from. - 1040 source-document classification — inbound client packets sorted automatically — W-2s, 1099-NEC, 1099-INT/DIV/B/R, K-1s, mortgage interest statements, brokerage 1099-B activity — each tagged, named, and routed to the right return. - Line-level extraction to tax software — form-level data extracted to Lacerte, UltraTax, ProSeries, Drake, or CCH Axcess via published integrations, with field-level provenance preserved for review. - Bank statement transaction categorization — sub-3-second extraction with smart-merchant categorization for Schedule C and rental Schedule E preparation. - Audit-ready workpaper indexing — source documents linked to financial-statement line items, with reviewer signoff and an audit-defensible chain of custody. 1040 source documents — extraction result client_idC-RT-204871 (PII redacted in log) tax_year2025 w2_forms3 detected · US$ 187,420 total wages 1099_int2 detected · US$ 1,847 interest 1099_div1 detected · US$ 4,920 (US$ 2,840 qualified) k1_forms1 detected · partnership: US$ 12,400 ordinary income schedule_b_required✓ yes — interest + dividends > US$ 1,500 routingLacerte template: form-1040 Extracted in 6.3s · all fields linked to source PDFs 99.6% Field-level accuracy on W-2, 1099, K-1 extraction <6s Per-page extraction across mixed-form bundles 25+ Tax forms supported out of the box API REST and signed-webhook integration Accounting specifics ## Built for the documents that drive the return. Client packets arrive in every imaginable shape — emailed PDFs, portal uploads, scanned mail-ins, mobile photos of W-2s. Cogneris's accounting configuration handles the full bundle: W-2s, 1099-NEC/MISC/INT/DIV/B/R/K, K-1s from 1065 and 1120-S filings, Schedule C/E/F supporting docs, bank and brokerage statements, payroll registers, fixed asset and depreciation schedules, and prior-year returns for roll-forward. ### W-2 batch ingestion Drop a folder of W-2s; receive a normalized CSV ready for tax software, with cross-form reconciliation against the prior year's wages. ### K-1 schedule parsing Partnership, S-corp, and trust K-1s parsed line by line with state apportionment schedules detected and routed. ### Brokerage 1099-B activity Composite 1099-B with hundreds of lots parsed to Form 8949 with cost-basis flags for non-covered securities. ### Source-document audit trail Every figure in the return traces back to a specific page in a specific source document — defensible under IRS examination. Tax & bookkeeping integrations ## Lives where your preparation and close systems already run. Cogneris integrates with the major tax-preparation platforms and bookkeeping systems through each vendor's published data import APIs — so extracted form data flows directly into the return file or the general ledger without an intermediate data-entry step. - Lacerte & ProSeries (Intuit) — extracted form data flows directly to the tax preparation file via Intuit's data import APIs. - UltraTax CS (Thomson Reuters) — source document data populates the input screens with field-level mapping preserved for review. - Drake Tax — imported via Drake's data integration layer, with workpaper PDFs linked to the e-file submission package. - CCH Axcess Tax (Wolters Kluwer) — multi-entity returns supported, with K-1 reconciliation across partners and shareholders. - QuickBooks, Xero & NetSuite — bookkeeping integrations for monthly close — source documents parsed and matched to GL transactions automatically. Documents we extract ## The document types behind tax season and bookkeeping. AP intake, source-document gathering for 1040 prep, monthly reconciliations — they all run on the same set of inputs. Cogneris extracts each into structured objects ready for your tax-prep or accounting platform. For category context, see [the IDP buyer's guide](what-is-idp.html), the [2026 State of Document AI report](state-of-document-ai-2026.html), or [estimate ROI](roi-calculator.html) at your volume. - [Invoice extraction](extract-invoices.html) — vendor, line items, GL-code hints, and totals for AP automation across NetSuite, QuickBooks, Xero, and Sage. - [Receipt extraction](extract-receipts.html) — itemized employee receipts with tip handling and policy-flag detection for T&E reconciliation. - [Tax return extraction](extract-tax-returns.html) — prior-year 1040s, K-1s, and schedules with AGI, deductions, and partner-level allocations parsed for current-year prep. - [Bank statement extraction](extract-bank-statements.html) — transaction-level extraction from client statements for monthly reconciliation, cash-flow review, and source-of-funds documentation. - [Payroll extraction](extract-payroll.html) — W-2s, 1099s, and pay stubs with wage, withholding, and YTD reconciliation tied directly to 1040 line items. FAQ ## Common questions. Which tax software does Cogneris integrate with? Cogneris ships published integrations with Lacerte and ProSeries (Intuit), UltraTax CS (Thomson Reuters), Drake Tax, and CCH Axcess Tax (Wolters Kluwer). Extracted form data — W-2 boxes, 1099 amounts, K-1 partner allocations, Schedule C line items — flows directly into the preparation file via each vendor's data import API. Field-level provenance is preserved so reviewers can click any imported figure and jump straight to the source PDF page. QuickBooks, Xero, and NetSuite are also supported for bookkeeping engagements. Can Cogneris handle multi-entity engagements? Yes. Partnership (1065), S-corp (1120-S), C-corp (1120), and trust (1041) returns are supported alongside individual 1040s. K-1s issued by pass-through entities are reconciled across partners and shareholders, and state apportionment schedules are detected and routed. CCH Axcess Tax is the primary integration target for multi-entity workflows; Lacerte and UltraTax also support entity returns. The tenant model isolates each client engagement, so a single firm can run consolidated workpapers across hundreds of related entities without cross-leakage. How does Cogneris stay current with IRS form revisions? Form templates are version-pinned per tax year. When the IRS publishes a revision — a new box on the W-2, a renamed line on the 1099-NEC, a changed K-1 footnote schedule — we ship an updated extractor before the form enters general circulation. Tax year 2025 forms are supported out of the box; tax year 2026 drafts are added as IRS releases them. Customers see a per-form version stamp on every extraction so they can prove which template revision produced a given figure during peer review or audit. Is Cogneris SOC 2 Type II audited? Yes — Cogneris carries an annual SOC 2 Type II report covering the Security, Availability, and Confidentiality trust services criteria, audited under AICPA SSAE 18. The current report is available under NDA. Controls align with the safeguards CPA firms expect under the IRS Pub 4557 safeguarding guidance and the Gramm-Leach-Bliley Act safeguards rule. Access controls, encryption in transit (TLS 1.3) and at rest (AES-256-GCM via GCP KMS), and immutable audit logs are built into the platform — not bolted on. What happens with client documents after a return is filed? Document, extraction, and audit retention are tenant-configurable and contract-specific. Firms must define periods that match their engagement, tax, litigation-hold, and operational requirements. How does pay-per-page pricing work for tax season volume? Cogneris prices per page extracted, not per seat or per return — so a 200-return practice and a 2,000-return practice both pay only for what they process. Tax season volume tiers kick in automatically once monthly throughput crosses set thresholds; the unit rate drops as your processed-page count rises. There's no minimum commit and no surge pricing in March and April. Firms typically run a paid pilot on 50–100 client packets before season starts to model their effective cost per return. Accounting evaluation ### Send us 50 client packets. We'll show you what comes out. Send us 50 client packets (de-identified). We'll run them through Cogneris and return a structured-extraction report with form coverage, accuracy, and time per packet. No commitment. [Book an evaluation](contact.html) [Security & compliance](security.html) ## Source: solutions-banking.html [Skip to content](#main) Banking # Document AI for banks. Audit-ready by default. Loan applications, KYC packages, bank and credit-card statements, mortgage closing files, and court orders pile up in every banking workflow. Cogneris structures every document the moment it arrives — with SOC 2 Type II controls (audit in progress), GLBA-aligned PII handling, zero-retention LLM access, and an audit trail your examiners and internal audit can defend. [Book a demo ](contact.html) [Security & compliance](security.html) The challenge ## Banking documents are PII-heavy, regulator-scrutinized, and arrive in every imaginable format. ### Loan/credit document chaos Income docs, statements, ID, and employment verification arrive in different formats from every applicant. Manual review delays decisions and breeds inconsistency across underwriting teams. ### KYC/AML at regulator-defensible scale BSA, CIP, and CDD/EDD obligations require documented, auditable verification on every customer. Manual review caps throughput and creates audit gaps that surface in the next examination. ### Underwriting cycle time Days, not minutes, from application to decision. Every hour of latency is a deal lost to a faster lender — and every manual hand-off adds another point where the file can stall or get re-keyed. How Cogneris helps ## Banking documents, structured. Compliance, built in. Cogneris extracts loan applications, KYC packages, bank statements, and court orders — with SOC 2 Type II controls (audit in progress), GLBA-aligned PII handling, and zero-retention LLM access on the critical path. - Loan application bundle parsing — pay stubs, W-2s, 1099s, bank statements, and ID parsed from a single drop-in submission with field-level confidence scoring on income, employment, and assets. - KYC/AML document verification — government ID, proof of address, and beneficial ownership extracted and cross-validated with sanctions and PEP screening; full CIP/CDD audit record preserved per regulator request. - Bank statement transaction parsing — sub-3-second extraction of transactions, balances, and recurring-deposit patterns from US and Canadian bank/credit-card statements for cash-flow underwriting. - Configurable audit metadata — supported workflows can record tamper-evident model version, prompt and response hashes, validation outcomes, and reviewer decisions. Loan application bundle — extraction result applicant_idA-LN-204771 (PII redacted in log) ssn_match✓ verified monthly_income_grossUS$ 8,420.00 employer_verification✓ via The Work Number bank_statements3 months, 247 transactions recurring_deposits✓ matches stated income kyc_status✓ CIP/CDD complete decision_payloadready for LOS Extracted in 4.1s · zero-retention LLM path 99.4% Field-level extraction accuracy on income docs <3s Per-page extraction time on bank statements SOC 2 Type II audit in progress GLBA Privacy & Safeguards-aligned controls Banking specifics ## Built for the documents that drive lending and compliance. Banking documents arrive in every imaginable format — uploaded loan packages, faxed court orders, scanned IDs, multi-month statement PDFs from a thousand institutions, and beneficial ownership certifications captured at the branch. Cogneris's banking configuration handles the full bundle: every document parsed, every field cross-validated, every PII element tracked through the audit trail. ### Income verification W-2, 1099, and pay stubs parsed to line-item granularity. YTD reconciliation with stated income, with confidence flags for human review on low-confidence fields. ### AML transaction screening Bank statement transactions matched against sanctions, PEP, and structured-deposit patterns flagged for SAR review. Outputs feed your existing AML case-management tooling. ### Mortgage document indexing TRID-required disclosures, appraisals, title commitments, and homeowners insurance extracted as a closing package with cross-document field reconciliation. ### Court order processing Levies, garnishments, and subpoenas parsed with key dates and obligated-account identifiers surfaced for legal review and timely response. LOS, core & fintech integrations ## Lives where your lending and core systems already run. Cogneris integrates with the loan origination, core banking, and decisioning platforms that already run your bank — so extracted document data flows directly into the LOS and customer record without an intermediate data-entry step. - nCino & Encompass — loan origination data flows directly into LOS workflows with full document provenance preserved. - Salesforce Financial Services Cloud — extracted applicant data lands as structured records on Person Account and Financial Account objects. - Plaid & MX — bank statement extraction complements (and reconciles against) ACH-feed data when statements arrive as PDFs. - Jack Henry, Fiserv & FIS cores — account opening packages and beneficial ownership certifications routed to the core via existing fintech connectors. - Mambu, Q2 & Alloy — extracted data integrates with the modern loan-origination and decisioning stack. Documents we extract ## The document types behind lending and KYC. Every loan, account opening, and beneficial-ownership review pulls from the same handful of document types. Cogneris extracts each as a structured object — fields, confidence scores, full audit trail. For category context, see [the IDP buyer's guide](what-is-idp.html), the [2026 State of Document AI report](state-of-document-ai-2026.html), or [estimate ROI](roi-calculator.html) at your volume. - [Bank statement extraction](extract-bank-statements.html) — cash-flow underwriting from US and Canadian statements, with transactions, balances, and recurring-deposit detection in under 3 seconds per page. - [Check OCR API](check-ocr-api.html) — payer, payee, amount, date, MICR context, remittance data, and validation evidence for payment workflows. - [KYC document extraction](extract-kyc-documents.html) — government IDs, proof of address, and beneficial-ownership certifications cross-validated against sanctions and PEP screening, with full CIP/CDD audit record. - [Payroll extraction](extract-payroll.html) — pay stubs, W-2s, and 1099s with gross, deductions, and YTD reconciliation against stated income for loan underwriting. - [Tax return extraction](extract-tax-returns.html) — 1040s, schedules, and K-1s with AGI, deductions, and refund/owed amounts pulled to support credit decisioning. FAQ ## Common questions. What is Cogneris's SOC 2 status? Our SOC 2 Type II audit is currently underway. Target report date, audit progress, and our current Type I report are available under NDA — email [security@cogneris.ai](mailto:security@cogneris.ai). Our control environment is aligned with the NIST 800-53 Moderate baseline. We map SOC 2 controls to FFIEC IT examination handbook expectations so vendor risk teams at banks and credit unions can complete OCC bulletin 2013-29 third-party reviews without bespoke evidence collection. How does Cogneris handle PII and customer data under GLBA? Cogneris's controls are aligned with the GLBA Privacy Rule and Safeguards Rule. Customer non-public personal information (NPI) is encrypted in transit (TLS 1.3) and at rest (AES-256-GCM with GCP KMS), access is gated by role-based authorization with audit logging, and all LLM API calls run with zero-retention configured — no training, no logging beyond operational requirements. Sub-processors are listed on our data-protection page, and a written information security program (WISP) is provided to Enterprise customers on request. What document types are supported out of the box for banking? Loan applications, pay stubs, W-2s, 1099s, tax returns, employment verification letters, bank and credit-card statements, KYC packages (government ID + proof of address), beneficial ownership certifications, business formation documents (Articles of Incorporation, EIN letters), mortgage TRID disclosures, appraisals, title commitments, trust documents, FinCEN/CTR/SAR filings, and court orders (levies, garnishments, subpoenas). Custom document types are available on Enterprise with few-shot configuration — typically 1 business day from sample documents to a working extractor. Can Cogneris run inside our VPC? Private cloud, customer-managed keys, AWS PrivateLink, and Google Cloud Private Service Connect are roadmap capabilities, not generally available services. Confirm current availability in a signed deployment scope. How long is customer document data retained? Document, extraction, and audit retention are tenant-configurable and contract-specific. Customers must define periods that match their legal, regulatory, litigation-hold, and operational requirements. Does Cogneris integrate with our LOS/core banking platform? Yes — nCino, Encompass, Mambu, Alloy, and Q2 are supported for loan origination and decisioning. Salesforce Financial Services Cloud lands extracted applicant data on Person Account and Financial Account objects. Jack Henry, Fiserv, and FIS cores are supported via existing fintech connectors for account opening packages and beneficial ownership certifications. Plaid and MX bank-feed data is reconciled against statement-PDF extractions when both are present. Custom integrations are available on Enterprise. Banking evaluation ### Send us a sample bundle. We'll show you what comes out. Send us 50–100 sample documents (de-identified or under NDA). We'll run them through Cogneris and return a structured-extraction report with field-level accuracy, processing time, and compliance posture. No commitment. [Book an evaluation](contact.html) [Security & compliance](security.html) ## Source: solutions-embedded-finance.html [Skip to content](#main) Embedded finance # Document verification inside your product. Neobanks and lenders need document checks at onboarding but can't afford the latency or UX disruption of redirecting users to a third-party portal. Cogneris embeds directly in your flow — sub-3-second results, no context switch. [Book a demo ](contact.html) [API reference](api.html) The challenge ## Document friction costs you customers. ### Onboarding drop-off Redirecting users to a third-party verification portal during account opening breaks the flow. Every additional step loses a percentage of users you'll never recover. ### Review queue latency Manual document review means applicants wait hours for a decision. In a market where competitors approve in minutes, that wait destroys conversion. ### Compliance requirements Lending and account-opening regulations require documented proof of income, identity, and address — all of which must be verified, stored, and auditable on demand. How Cogneris helps ## Embedded, white-label, sub-3-second. Cogneris's API integrates directly into your onboarding flow. Users upload documents inside your product, and extraction results arrive before they have time to lose patience — returning normalised income, address, and identity fields ready to use. - White-label API — users never see a third-party interface. Document upload and feedback happen entirely within your product's UX. - Pre-built models for finance documents — payslips return net/gross income, employment status, and employer name. Bank statements return transaction history and average balance. Utility bills return address and date. - Confidence-gated auto-approval — high-confidence extractions approve instantly. Edge cases route to your review queue with pre-filled data, not a blank form. - Regulatory-ready storage — documents and extraction results stored with configurable retention periods and GDPR-compliant data-processing agreements. Payslip extraction — onboarding employerTech Corp Ltda. gross_salaryUS$ 12,480.00 net_salaryUS$ 9,856.00 employment_typeFull-time — permanent ref_month2026-03 confidence0.97 → auto-approved Decision returned in 2.4s Supported: Payslips · Bank statements · Utility bills · Tax returns · National IDs <3s End-to-end document verification 100% White-label — your brand, your UX 5 Core finance document types pre-built 99.2% Accuracy on payslips and statements Embedded finance specifics ## Built for in-flow KYC. Embedded finance products live or die on conversion at the identity-verification step. Three-second response times aren't a stretch goal; they're the difference between a user finishing onboarding and dropping off. Every architectural choice in Cogneris's embedded-finance configuration optimizes for that latency budget without compromising the regulator-facing audit trail underneath. ### Sub-3-second extraction ID, payslip, bank-statement extraction in under 3 seconds at the 95th percentile. Headless API — no third-party redirect, no UX context-switch. ### Income verification stack Cross-references payslips, bank statements, and verification letters. Returns reconciled monthly income with audit lineage. Works as a Plaid alternative for pay-stub income. ### State-aware compliance FFIEC, BSA/AML, and state-level KYC requirements (NY DFS, CA DFPI, TX DOB) configurable per workflow. ### Regulator-ready evidence Every verification produces an immutable evidence record: extraction inputs, decision rationale, model version, reviewer if any. Ready for examiner sampling. The embedded-finance stack ## Plays well with your existing rails. Cogneris slots into the toolchain neobanks, lenders, and embedded-finance platforms already run — augmenting income and identity data, not replacing it. - Plaid — Use Cogneris for income docs that Plaid can't access (1099s, employer letters, pay-stubs from non-Plaid-connected payrolls). Same schema, complementary coverage. - Argyle / Pinwheel — Cross-validate Argyle/Pinwheel payroll data against extracted pay-stubs for fraud detection. - Alloy / Persona — Cogneris handles the document-AI side of the verification stack; Alloy/Persona orchestrate the rules. We integrate as a verification step in the orchestration. - Card issuers — Marqeta, Lithic, Stripe Issuing — extracted KYC artifacts pushed directly into card-issuance workflows. - Core banking — REST APIs and signed webhooks suitable for custom integrations with embedded-banking platforms. Documents we extract ## The document types behind onboarding and underwriting at the partner layer. Embedded finance products run KYC, bank statement underwriting, and counterparty risk reviews at scale — across hundreds of partner brands. Cogneris extracts each input into structured fields your decision engine can consume. For category context, see [the IDP buyer's guide](what-is-idp.html), the [2026 State of Document AI report](state-of-document-ai-2026.html), or [estimate ROI](roi-calculator.html) at your volume. - [Bank statement extraction](extract-bank-statements.html) — cash-flow underwriting for small-business borrowers with transactions, balances, NSF detection, and recurring-deposit patterns from US/CA statements. - [KYC document extraction](extract-kyc-documents.html) — government IDs, proof of address, and beneficial-ownership certifications for CIP/CDD on every onboarded merchant. - [Invoice extraction](extract-invoices.html) — counterparty invoices for revenue verification, with vendor, line items, and totals used to corroborate volume claims during onboarding. FAQ ## Common questions. Why Cogneris instead of Plaid for income verification? Plaid is excellent when the user's payroll provider is in Plaid's network. Cogneris covers the gap — 1099 contractors, gig workers, employer verification letters, payslips from smaller payroll providers. Most lenders use both. Can we use Cogneris without redirecting users to a third-party portal? Yes. The API is fully headless. Your app captures the document upload, sends it to Cogneris, gets structured data back. No Cogneris branding visible to your users. Does Cogneris handle state-level compliance requirements? We support workflow-level configuration for NY DFS Part 500, CA DFPI MTL, TX DOB, and the major state lender frameworks. Each verification produces evidence aligned with the framework's documentation requirements. What happens to documents after verification? Document, extraction, and audit retention are tenant-configurable and contract-specific. Customers must define periods that match applicable AML, privacy, and operational requirements. Can Cogneris extract from international IDs? Yes — major OECD passports and IDs are supported with MRZ checksum and structural integrity validation. State-specific compliance rules apply per the applicable jurisdiction. Embedded intelligence ## Document verification that feels like part of your product. 30 minutes to see it running inside a sample onboarding flow. We bring the demo; you decide if it fits. [Book a demo ](contact.html) [API reference](https://docs.cogneris.ai/) ## Source: solutions-enterprises.html [Skip to content](#main) Enterprise solutions # Document scale without the headcount. From complex AP workflows to compliance-critical document processing, Cogneris delivers the throughput, security, and auditability that large organizations demand — without adding staff. [Book a demo ](contact.html) [See the product](products.html) The challenge ## Why enterprise document processing is hard. ### Volume without throughput Thousands of documents per day across departments — invoices, HR files, contracts — with no unified pipeline to route, prioritize, and track them end to end. ### Security & compliance mandates InfoSec teams block cloud-first tools without SOC 2 and ISO 27001 certifications, or without the option for dedicated VPC or on-premises deployment. ### ERP & system integration Extracted data needs to flow automatically into SAP, Salesforce, Oracle, or NetSuite — not require manual re-entry that defeats the purpose of automation. How Cogneris helps ## Built for the enterprise from day one. Not bolt-on. Cogneris is built for regulated, high-volume environments. Every feature — from confidence scoring to audit logs — was designed for enterprise requirements, not added as an afterthought. - Human-in-the-loop queues — confidence scoring surfaces exceptions automatically; reviewers only touch documents that need them. - Private deployment roadmap — Private cloud, BYOC, and on-premises deployment are not generally available. Any contract-dependent evaluation requires explicit technical and operational confirmation in a signed scope. - Custom system integration — REST APIs and signed webhooks suitable for connecting ERP, CRM, and finance systems. - Full audit trail — every extraction, review, and approval logged with timestamps and user identity, ready for SOX or external audit. Invoice extraction — live result vendorAcme, Inc. invoice_noINV-001234 totalUS$ 48,200.00 tax_id12-3456789 due_date2026-05-15 confidence0.98 → auto-approved Routed to SAP in 1.1s Security baseline: SOC 2 Type II (audit in progress) · ISO 27001 on roadmap · TLS 1.3 in transit · AES-256 at rest · Dedicated KMS per tenant 10M+ Documents processed per month 99.2% Extraction accuracy on structured docs <1.2s Average extraction latency 48h Typical enterprise onboarding time Enterprise platform ## Built for the enterprise stack. Enterprise document processing isn't about OCR accuracy on one document — it's about throughput, dual-approval workflows, audit traceability, and integration into the systems you already run. Cogneris Enterprise is the configuration we ship to teams running ERP migrations, SOX-relevant AP processes, and regulator-facing extraction workflows. ### Single-tenant VPC roadmap Planned dedicated-project and network-boundary capability. Not generally available unless included in a signed deployment scope. ### Customer-managed keys roadmap Planned key-ownership and revocation capability. Not generally available unless included in a signed deployment scope. ### SOX-ready audit trail Configured workflows can produce tamper-evident metadata with model version, prompt and response hashes, and post-processor decisions. Export and retention are contract-specific. ### Named TAM & 24/7 escalation Dedicated technical account manager, a documented escalation path, and a 4-hour critical-incident response SLA. Integrations ## Lives where your data already is. Cogneris pushes structured output directly into the systems your teams already run. No middleware, no batch jobs, no overnight syncs. - ERP — REST APIs and signed webhooks suitable for custom integrations with ERP systems. - ITSM / workflow — Bidirectional sync with Salesforce, ServiceNow, and Jira for exception routing and approval workflows. - Data warehouse — Daily exports to Snowflake, Databricks, and BigQuery. CDC patterns supported on request. - Identity — SSO via Okta, Azure AD, Ping. SCIM provisioning for user lifecycle. Just-in-time access for support staff. - Observability — Audit-trail bridge to Splunk, Datadog, and Sumo Logic — without sending document content to the observability layer. Documents we extract ## The document types behind procurement, legal review, and onboarding at scale. Large organizations run thousands of contract reviews, invoice ingests, vendor onboardings, and HR document moves per quarter. Cogneris extracts each input into structured records flowing into your ERP, CLM, and HCM. For category context, see [the IDP buyer's guide](what-is-idp.html), the [2026 State of Document AI report](state-of-document-ai-2026.html), or [estimate ROI](roi-calculator.html) at your volume. - [Contract extraction](extract-contracts.html) — MSAs, NDAs, SaaS agreements, and vendor contracts with parties, term, governing law, liability caps, and named-clause extraction for CLM and legal review. - [Invoice extraction](extract-invoices.html) — vendor invoices at scale, with line items, GL-code hints, totals, and 3-way matching ready for SAP, Oracle, Workday, and NetSuite. - [KYC document extraction](extract-kyc-documents.html) — government IDs, proof of address, and beneficial-ownership certifications for vendor onboarding and counterparty due diligence. - [Payroll extraction](extract-payroll.html) — pay stubs, W-2s, 1099s, and tax forms with wage, withholding, and YTD data flowing into Workday HCM and ADP. FAQ ## Common questions. Do you offer true on-premise deployment? Single-tenant VPC, private cloud, and customer-managed keys are roadmap capabilities, not generally available services. Confirm any dedicated deployment in a signed scope before relying on it. What's your enterprise SLA? 99.95% uptime, 4-hour critical-incident response, 24-hour resolution target on production-blocking issues. SLA credits apply for breaches and are negotiated into the contract. Can Cogneris meet our SOX 404 testing requirements? Configured workflows can produce tamper-evident audit metadata including model version, prompt and response hashes, and post-processor decisions. Availability, export, and retention depend on the service, tenant configuration, and contract. How does Enterprise pricing scale? Annual commits with volume discounts that scale into eight figures of documents per year. Pay-per-page (no commit) is also available — Enterprise customers typically move to commit pricing once monthly volume stabilizes. Can our internal audit team review Cogneris directly? Yes. Enterprise contracts include audit rights. In practice most customers prefer the SOC 2 Type II report (under NDA) plus a security questionnaire response — both turn around in under a business day. Enterprise ready ## See Cogneris running on your documents. In 30 minutes we'll show you the platform processing your own document types, with your compliance requirements in scope from day one. [Book an enterprise demo ](contact.html) [See pricing](pricing.html) ## Source: solutions-finance-automation.html [Skip to content](#main) Finance automation # Automate the paper trail. Keep the audit trail. AP/AR teams spend 40% of their month re-keying data from invoices, receipts, and statements into ERPs — then produce reconciliation errors anyway. Cogneris extracts, validates, GL-codes, and routes every document automatically so your team focuses on exceptions, not data entry. [Book a demo ](contact.html) [See the product](products.html) The challenge ## Why month-end close still takes a week. ### Manual data entry at scale AP teams re-key vendor names, invoice numbers, amounts, and GL codes for every invoice — compounding error rates across thousands of documents per month. ### Reconciliation errors at close Discrepancies between invoices, POs, and receipts aren't discovered until month-end reconciliation — too late to dispute without damaging supplier relationships. ### Audit evidence gaps SOX and external audits require full evidence of every transaction decision. Manually assembled audit packages take days to produce and are prone to gaps. How Cogneris helps ## From invoice inbox to ERP without touching it. Cogneris extracts every field from every invoice, suggests GL codes and cost centres, validates against your PO register, and pushes results directly to your ERP — with a complete audit trail included. - GL-code suggestion and cost-centre tagging — extracted invoice data maps to your chart of accounts automatically, with override capability for edge cases. - Custom ERP integration — REST APIs and signed webhooks suitable for connecting finance systems. - SOX-ready audit trail — every document decision (extracted value, confidence score, reviewer, timestamp) is permanently logged and exportable for any audit without manual evidence gathering. - Exception-only human review — low-confidence extractions and matched-against-PO discrepancies surface to reviewers. Everything else flows straight through. Invoice → ERP pipeline result vendorAcme Supplies, Inc. amountUS$ 23,450.00 taxUS$ 2,110.50 (sales tax) gl_code6200 — Services cost_centreMarketing — NY po_match✓ PO-2026-0311 statusauto-approved → NetSuite 40% Reduction in AP team processing time 99.5% Extraction accuracy on invoices 100% Decisions logged — SOX-ready on day one 5 ERP connectors — SAP, NetSuite, Xero + AP/AR specifics ## Three-way match. Done. Finance automation isn't OCR — it's a workflow. Extraction is one step. Three-way match against POs and goods receipt. GL coding by category. Variance detection at the line-item level. Approval routing by amount. Exception handler queues for the 5-10% of invoices that need human eyes. Cogneris bundles all of it into one platform; the alternative is gluing five vendors together. ### Three-way match engine Invoice ↔ PO ↔ goods receipt note. Line-item granularity. Configurable tolerance thresholds. Exceptions routed automatically. ### GL coding rules engine Vendor-specific GL accounts, cost centers, and tax codes. Learnable per-vendor coding from your historical entries. Auditable rule application. ### Approval routing Configurable workflows by amount, GL account, or vendor. Slack / email / in-app notifications. Escalation if approver SLA misses. ### Exception queue 5-10% of invoices need human review. The exception queue surfaces them with the failing rule highlighted, ready for a single-click resolution. ERP integrations ## Pushes data where it already lives. Cogneris provides REST APIs and signed webhooks suitable for custom integrations between document workflows and finance systems. Scope, mapping, and delivery timelines depend on the target system. - ERP systems — custom mapping can use REST APIs and signed webhooks; bi-directional sync is not a packaged connector. - SAP S/4HANA — OData-based integration. Supports company codes, cost centers, profit centers, tax codes. Works with both on-prem ECC and cloud S/4. - Oracle Cloud ERP — REST + OAuth integration. Supplier hierarchy, segment-based GL coding, and Oracle's approval workflows mirrored in Cogneris's review UI. - Workday Financials — Customer Central integration. Supplier invoice records created with full line-item detail and matching attachments. - Microsoft Dynamics 365 — Common Data Service integration. Vendor invoices created with line-level detail; PowerAutomate flows can extend the workflow. Documents we extract ## The document types behind AP, AR, and reconciliation. Finance ops runs on inbound invoices, expense receipts, statements that need matching to the GL, and payroll documents that touch tax filings. Cogneris extracts each into structured records ready for ERP ingestion. For category context, see [the IDP buyer's guide](what-is-idp.html), the [2026 State of Document AI report](state-of-document-ai-2026.html), or [estimate ROI](roi-calculator.html) at your volume. - [Invoice extraction](extract-invoices.html) — vendor invoices with line items, GL-code hints, totals, and 3-way matching ready for NetSuite, SAP, Oracle, Workday, QuickBooks, and Xero. - [E-invoicing automation](e-invoicing-automation.html) — invoice intake, tax and total validation, PO matching, review routing, ERP handoff, and audit trails for AP compliance. - [SAP document extraction](sap-document-extraction.html) — extract AP packets before sending clean or reviewed invoice, vendor, PO, and receipt data into SAP-centered workflows. - [Receipt extraction](extract-receipts.html) — employee receipts for T&E with itemized purchases, policy-flag detection, and routing for reimbursement workflows. - [Bank statement extraction](extract-bank-statements.html) — bank and credit-card statements with transaction-level parsing for monthly reconciliation, fraud review, and cash-position reporting. - [Payroll extraction](extract-payroll.html) — pay stubs, W-2s, and 1099s with wage and withholding data tied to general-ledger and tax-filing workflows. FAQ ## Common questions. How does Cogneris handle the 5% of invoices with weird formats? The exception queue. Cogneris's confidence score per field flags low-confidence extractions; they go into a queue with the document, the extracted fields, and the failing validation rule highlighted. Most exceptions resolve in under 30 seconds — much faster than re-keying the whole invoice. Do we have to replace our existing ERP? No. Cogneris sits in front of the ERP as a document-to-data layer. The ERP stays the system of record; Cogneris eliminates the data-entry workflow between AP and the ERP. Most customers' ERPs are unchanged after Cogneris deployment. How does GL coding actually work? Cogneris learns from your historical bills and invoices. After roughly 200 documents per vendor, the system suggests GL accounts, cost centers, and tax codes with high confidence. The learning is per-tenant — your patterns aren't shared with other customers. Can we audit GL coding decisions? Yes. Every coding decision is logged with the rule that produced it (system, learned, manual override). Auditors can pull a sample and trace each line back to source. What's the typical deployment timeline? Mid-market: 4-6 weeks including ERP integration, vendor master sync, and a 4-week parallel run with the existing process. Enterprise: 8-12 weeks with custom workflow configurations. Finance automation ## Close faster. With confidence. 30 minutes to see Cogneris processing your actual invoice formats — with GL-code mapping and ERP push configured to your chart of accounts. [Book a demo ](contact.html) [See pricing](pricing.html) ## Source: solutions-healthcare.html [Skip to content](#main) Healthcare # Document AI for clinical workflows. HIPAA-aware by default. Medical bills, prior authorizations, patient intake forms, EOBs, and clinical notes pile up in every healthcare workflow. Cogneris structures every document the moment it arrives — under a HIPAA BAA, with PHI-aware retention, zero-retention LLM access, and an audit trail your compliance team can defend. [Book a demo ](contact.html) [Security & compliance](security.html) The challenge ## Healthcare documents are PHI-laden, code-heavy, and arrive in every imaginable format. ### Medical billing accuracy UB-04, CMS-1500, and itemized bills carry CPT, ICD-10, HCPCS, and modifier codes that drive payment. Manual code entry produces errors that snowball into denials, audits, and rework. ### Prior auth turnaround Prior authorization requests pile up while patients wait. Manual review of clinical documentation and policy criteria delays care decisions, frustrates providers, and increases administrative cost per request. ### PHI handling under HIPAA Every document touching a patient's chart is PHI. Most general-purpose AI platforms aren't configured for HIPAA — no BAA, unclear LLM data flow, default-on logging that captures protected health information. How Cogneris helps ## Clinical documents, structured. Compliance, built in. Cogneris extracts medical bills, prior auth packages, patient intake, and clinical notes — under a HIPAA BAA, with code validation, code-set normalization, and zero-retention LLM access on the critical path. - Medical bill line-item parsing — CPT, ICD-10-CM, ICD-10-PCS, HCPCS, modifiers, charges, allowed amounts, and provider taxonomy extracted as structured arrays, code-validated against current CMS code sets. - Prior authorization triage — clinical documentation, member info, and policy criteria extracted from PA packages with auto-routing to clinical reviewers when criteria are ambiguous. - Patient intake digitization — handwritten and faxed intake forms parsed to discrete fields ready to populate the EHR, with confidence flags for human review on low-confidence fields. - HIPAA BAA on Enterprise — Business Associate Agreement, PHI-aware retention, zero-retention LLM API, and an immutable per-request audit trail your compliance team can show during a HIPAA Security Rule audit. UB-04 medical bill — extraction result patient_idP-RT-948112 (PHI redacted in log) provider_npi✓ 1234567890 service_date2026-04-22 primary_dxICD-10: J45.40 (asthma, uncomplicated) cpt_codes99214, 94640, 94060 total_chargesUS$ 487.00 allowed_amountUS$ 312.46 code_validation✓ All codes valid (CMS code set 2026) Extracted in 2.4s · zero-retention LLM path 99.7% CPT/ICD-10 code-validation accuracy <3s Per medical bill extraction 90d Default raw-document retention BAA HIPAA Business Associate Agreement Healthcare specifics ## Built for the documents that drive payment and care. Healthcare documents arrive in every imaginable format — UB-04 hospital bills, CMS-1500 professional claims, EOBs from payers, prior authorization packages, faxed referrals, handwritten intake forms, and clinical notes. Cogneris's healthcare configuration handles the full bundle: every document parsed, every code validated, every PHI field tracked through the audit trail. ### Code-set validation CPT, ICD-10-CM, ICD-10-PCS, HCPCS Level II validated against current CMS code sets. Mismatches and retired codes flagged for review before downstream adjudication. ### EOB & remittance parsing Explanation of Benefits, remittance advice (835), and denial codes (CARC/RARC) parsed to structured form for revenue cycle management workflows. ### Prior auth packages Multi-document PA packages — clinical notes, lab reports, member info, prescription details — extracted as a unit, with criteria matching against your medical policy. ### PHI redaction in logs Operational logs and audit trails carry pseudonymized identifiers, not PHI in plain text. Field-level lineage preserved without re-exposing protected information. EHR & clearinghouse integrations ## Lives where your clinical and billing systems already run. Cogneris integrates with the major EHR platforms and claims clearinghouses through FHIR APIs, HL7 v2 interfaces, and 835/837 EDI — so extracted document data flows directly into patient charts and claims workflows without an intermediate data-entry step. - Epic — FHIR R4 integration with App Orchard published connector. Extracted bill data populates patient accounts and claims with full document provenance. - Oracle Health (Cerner) — FHIR + Cerner Open Developer Experience. Multi-document bundles arrive as structured chart documents, ready for clinical review. - Athenahealth — Marketplace integration via athena.io APIs. Patient intake forms and clinical documents flow into athenaClinicals automatically. - eClinicalWorks & Meditech — HL7 v2 ADT/DFT/MDM messages and FHIR endpoints supported for both platforms. - Clearinghouses — Change Healthcare, Availity, Waystar — 837 claims and 835 remits ingested and parsed for revenue cycle workflows. Documents we extract ## The document types behind claims, prior auth, and patient onboarding. Healthcare back office runs on faxed referrals, prior-auth packets, insurance claims, and patient intake forms. Cogneris extracts each input into structured fields with HIPAA-aligned controls and a full audit trail. For category context, see [the IDP buyer's guide](what-is-idp.html), the [2026 State of Document AI report](state-of-document-ai-2026.html), or [estimate ROI](roi-calculator.html) at your volume. - [Insurance claim extraction](extract-insurance-claims.html) — claim forms (CMS-1500, UB-04), EOBs, and remittance advices with payer, dates of service, codes, and amounts pulled to a single claim object. - [Health insurance card OCR](health-insurance-card-ocr-api.html) — member IDs, payer names, plan details, group numbers, copays, and coverage evidence from front and back card images. - [KYC document extraction](extract-kyc-documents.html) — government IDs, proof of address, and insurance cards for patient identity verification and coverage validation at intake. - [Contract extraction](extract-contracts.html) — provider contracts, payer agreements, and BAAs with parties, term, reimbursement rates, and HIPAA flow-down terms extracted. FAQ ## Common questions. Is Cogneris HIPAA compliant? Cogneris offers a HIPAA Business Associate Agreement (BAA) on the Enterprise plan. The platform is configured with PHI-aware retention, zero-retention LLM API access, and access controls aligned with HIPAA Security Rule requirements. The Free, Pro, Scale and Unlimited plans are not configured for PHI processing — Enterprise is required for any workflow that handles protected health information. How does Cogneris handle PHI sent to LLM providers? All LLM API calls run with zero-retention configured (no training, no logging beyond operational requirements). OpenAI and Anthropic both offer BAAs covered through Cogneris's platform agreement. PHI is encrypted in transit (TLS 1.3) and at rest (AES-256-GCM with GCP KMS). Customer payloads are processed in memory and not persisted in vendor logs. Can Cogneris parse CPT, ICD-10, and HCPCS codes? Yes. Medical bills are parsed to CPT-code, ICD-10-CM, ICD-10-PCS, and HCPCS Level II level with charges, allowed amounts, modifiers, and provider taxonomy. Code validation against current CMS code sets is run on every extraction with mismatches flagged for review. How long is PHI retained? Document, extraction, and audit retention are tenant-configurable and contract-specific. Customers must define periods that match applicable privacy, clinical, litigation-hold, and operational requirements. Does Cogneris integrate with EHR systems? Yes — Epic, Cerner (Oracle Health), Athenahealth, eClinicalWorks, and Meditech via FHIR APIs and HL7 v2 interfaces. Extracted document data populates patient charts and claims with full document provenance and audit lineage preserved. Healthcare evaluation ### Send us a sample bundle. We'll show you what comes out. Send 50–100 sample documents (de-identified or under NDA + a draft BAA). We'll run them through Cogneris and return a structured-extraction report with code-validation results, confidence distribution, and time-to-process. No commitment. [Book an evaluation](contact.html) [Security & compliance](security.html) ## Source: solutions-hospitality.html [Skip to content](#main) Hospitality, travel & leisure # Frictionless check-in. Zero data entry. Guest check-in, visa verification, expense management, and supplier invoicing all involve documents. Hotel front desks re-key passport data; finance teams key in receipts. Cogneris eliminates every manual keystroke across the property operation. [Book a demo ](contact.html) [See the product](products.html) The challenge ## Documents slow down every part of the guest experience. ### Front-desk re-keying at check-in Staff manually transcribing passport data slows check-in, introduces errors into PMS records, and frustrates guests — especially at high-occupancy peaks when queues form at reception. ### Expense receipt processing volume High-volume hospitality businesses process thousands of receipts per month. Manual data entry delays financial close, misclassifies spend, and creates reconciliation errors at period end. ### Supplier invoice format fragmentation Linen, food, and facility suppliers use entirely different invoice formats. AP teams spend hours normalising before entry — every new supplier adds another format to handle manually. How Cogneris helps ## Passport to PMS in under 2 seconds. Cogneris extracts guest identity documents, expense receipts, and supplier invoices — pushing structured data directly to your PMS, ERP, or expense system without manual intervention. - Passport and ID to PMS in <2s — extract all passport fields including MRZ data and push directly to your property management system via webhook or REST. - Multi-format receipt classification — receipt photos, emailed PDFs, and scanned paper all normalised to a single schema: vendor, amount, date, and spend category. - Supplier invoice normalisation — line items, totals, and tax extracted across all supplier formats and mapped to your chart of accounts automatically. - MRZ validation — machine-readable zone parsed and cross-validated against biographical data, eliminating transcription errors on nationality, DOB, and document number. Guest check-in — passport extraction result doc_typePassport — P