The number that started retiring the per-page price
For most of the category's life, the unit of pricing in document AI was the page. It was easy to count, easy to meter, easy to put in a contract, and it had the great virtue of being the thing the system could obviously measure — a page went in, a page came out, you billed the page. That arrangement is quietly ending for the same reason it is ending across software: the buyer's frame of reference moved. The signal sits in the seat-pricing decline that ran through B2B in 2026 — pure per-seat licensing slipping from around a fifth of the market to roughly a seventh inside a year — and the cause underneath it is the agent. When a workflow is run by a person plus a handful of agents, the seat stops describing the work, and a price tied to seats stops describing the value. Vendors followed the value to where it actually lives, which is consumption and outcome.
Document AI is the same story told in pages instead of seats. The per-page price describes the input — how much paper the system chewed through — and the input is precisely the part the buyer has stopped wanting to pay for. A finance team does not have a budget line for "pages processed". It has a budget line for invoices that get paid without a human touching them, suppliers that get onboarded without a clerk keying a form, claims that get settled inside the service level. The page is the vendor's unit of cost. The decision is the customer's unit of value. Per-page pricing bills the first and asks the buyer to care about it, at the moment the buyer has decided to only pay for the second.
Per-page pricing charges for the paper that went in. The buyer stopped buying paper. They are buying the decision that comes out — and a price that meters the input is invisible to the value the customer can actually see.
Why per-page repeats the seat-pricing mistake
The reason this is worth naming as a mistake, and not just a different convention, is that per-page pricing actively misaligns the two sides of the contract. Under a per-page meter, the vendor is paid more when more pages flow through — which means the vendor is, on paper, rewarded for volume regardless of whether that volume produced a usable result. A document that bounces through extraction, fails validation, lands in a review queue and gets reworked by a human still billed for its pages. The customer paid for throughput and got an exception. The incentive points the wrong way: the vendor's revenue does not fall when the automation fails to land, so nothing in the price structure pushes the vendor to make the hard documents actually clear.
Outcome pricing inverts that. When the billable unit is the approved decision rather than the page, the vendor is paid only when the workflow produced the thing the customer was trying to buy — and is paid more, in aggregate, precisely when automation rises and the share of documents that complete without a human grows. That is the alignment the seat model lost and the page model never had: more automation means more revenue, not less, because the unit being billed is the successful result, and there are more successful results when the system is working well. It is the pricing expression of the same shift we wrote about in moving from extraction to decision — once the product is judged on the decision, the price has to be charged on the decision too, or the two halves of the business pull against each other.
What the buyer is actually buying now
The clearest way to see the turn is to put the two pricing frames beside each other on the same workflow. The per-page column is not wrong about anything it measures; it just measures the part of the system the buyer has stopped budgeting for. The per-outcome column measures the part they have started writing the cheque against.
| What the contract prices | Per-page — input era | Per-outcome — value era |
|---|---|---|
| Billable unit | The page processed, success or not. | The approved decision — invoice reconciled, supplier cleared. |
| Who carries exception cost | The customer — reworked pages still bill. | The vendor — only the clean result is charged. |
| What more automation does to revenue | Neutral or falling — fewer pages, less billing. | Rising — more documents complete without a human. |
| What the buyer can check | A page count they can't tie to value. | An outcome count that maps to their own ledger. |
| Who it wins with | The technical buyer comparing throughput. | The economic buyer comparing result per workflow. |
Read down the right column and the difference is not cosmetic: each line is the same workflow viewed through the instrument the buyer now uses. The page count is a number the vendor can produce and the customer cannot reconcile against anything they care about. The outcome count is a number that ties directly to the customer's own books — which is exactly why the economic buyer, the one who controls the budget, has started insisting the price be denominated in it.
How to define the billable outcome unit
The hard part of outcome pricing is not the philosophy; it is the definition. "Charge for the result" is a slogan until the result is a unit precise enough to put in a contract and auditable enough to put on an invoice. The work is in naming the outcome so tightly that both sides agree, before any money changes hands, on exactly what counts.
- Name the unit in the customer's language, not the system's — the billable event is "an approved decision", "a case closed", "a document that ran extraction, validation and write-back with no human touch", not "a page" or "an API call". The test is whether the unit already appears on the customer's own process map. If it does, they can reconcile the bill against their ledger; if it doesn't, you have named a vendor-side metric and dressed it as an outcome.
- Align the incentive so automation pays the vendor more — the structure has to make a higher straight-through rate produce more revenue, not less, or the model quietly recreates the per-page conflict in a new costume. The simplest form is a price per completed outcome that the vendor wants to maximise and the customer is happy to pay because each one replaced manual work they were budgeting for anyway.
- Set a floor and a collar to protect margin — outcome pricing exposes the vendor to the document mix, and a quarter where the hard, low-yield documents spike can turn a healthy price into a loss-making one. A price floor per period and a collar on the exception-heavy mix keep the model solvent when the input gets adversarial, without abandoning the outcome frame. This is the same margin discipline the token tax forces, applied to the revenue side instead of the cost side.
- Instrument the outcome before you bill it — you cannot charge for a result you cannot measure auditably, and an outcome bill the customer cannot verify is a dispute waiting to happen. Per-outcome telemetry — which documents completed without a human, which were touched, which were reworked, all tied to a record the customer can inspect — is the precondition, not the nice-to-have. The trace that makes agentic extraction auditable is the same instrument that makes outcome billing defensible.
What these have in common is that they convert a marketing posture into a billing system. The vendor who says "we charge for outcomes" but cannot define the unit, cannot align the incentive, cannot protect the margin and cannot prove the count is not pricing on outcome — they are pricing on hope and will discover the difference the first time a customer audits the invoice or the document mix turns against them.
"We charge for outcomes" is a sentence. An auditable outcome unit, an aligned incentive, a margin floor and per-outcome telemetry are a pricing model. The gap between the two is where the deals — and the margins — are lost.
The four ways outcome pricing goes wrong
Outcome pricing fails in recognisable ways, and every one of them is a version of moving to the new frame without building the machinery the frame requires. The failures are worth naming because each looks reasonable right up to the point it doesn't:
- Charging for an outcome you can't measure auditably — committing to bill per approved decision without the telemetry to prove, to the customer's satisfaction, which decisions were actually approved cleanly. The bill becomes a negotiation every month, the customer disputes the count, and the model that was meant to build trust spends it instead.
- Exposing yourself to negative margin on the hard mix — pricing per outcome with no floor and no collar, then watching a quarter of adversarial layouts, multi-document reconciliations and exception-heavy inputs drive the cost per completed outcome above the price. The frame is right; the absence of a margin guard is what sinks it.
- Keeping per-page because it's easy to meter — staying on the input price because it is simple to count and defensible in a spreadsheet, and losing the deal to the vendor who priced the result. Easy to measure is not the same as easy to sell, and the economic buyer has stopped accepting "but it's how the meter works" as a reason to pay for throughput.
- No clause separating outcome from effort on the edge cases — failing to write the contract term that handles the documents which always go to human review by design, so that genuinely ambiguous inputs either get billed as outcomes they aren't or get done for free. The edge cases are not the exception to the model; they are the part the contract has to name explicitly or the model leaks at both ends.
The trade-off, stated honestly
There is a real tension here, and pretending outcome pricing is free of cost is its own failure mode. The thing per-page and per-seat pricing buy you is predictability: a meter that ticks on input gives both sides a revenue line they can forecast, and a CFO on either side of the table can plan against it. Outcome pricing trades some of that predictability for alignment. Revenue now moves with the automation rate and the document mix, which means it is harder to forecast quarter to quarter and more exposed to a bad mix — and a vendor who moves to it without a floor, a collar and clean telemetry has swapped a predictable margin for an unpredictable one and called it progress.
The honest framing is not "outcome pricing always wins" or "per-page is obsolete", both of which are slogans. It is that the buyer's willingness to pay has moved to the outcome, and a price that does not follow it gets beaten on value even when it wins on simplicity. The discipline is to take the outcome frame and add back the predictability the input meter gave you for free — through the floor, the collar and the telemetry — rather than choosing between alignment and solvency as though you could only have one. High-volume, low-ambiguity workflows can still rationally meter close to the input, because the outcome and the input nearly coincide. The further a workflow sits toward hard, judgement-heavy documents, the more the value concentrates in the decision and the more the price has to move with it.
Why the economic buyer drives this, not the vendor
The turn is not happening because vendors woke up generous. It is happening because the buyer who signs changed. For most of the category's history the document AI purchase was made by a technical buyer comparing accuracy and throughput, and to that buyer a per-page price was legible and fair. The purchase has moved up the org chart to the economic buyer — the finance, operations or shared-services leader who owns the budget the workflow sits in — and that buyer does not think in pages. They think in the cost of the work the system is supposed to remove: the salaried hours of keying, matching and chasing that a clean decision replaces.
To that buyer, a price denominated in pages is a price they cannot tie to their own business case, and a vendor who insists on it is asking them to do the translation from input to value in their own head, every renewal. The vendor who prices in approved decisions has done that translation for them — the invoice speaks the same language as the budget it comes out of. This is the same legibility advantage that decides the ROI gap: the tool whose value the buyer can read off their own ledger survives diligence, and the one that asks the buyer to take the value on faith gets discounted or cut. Pricing is just where that legibility either exists or doesn't, written into the unit on the invoice.
Closing thought
The pricing conversation that ages well in document AI is not "how many pages did we process". It is "how many decisions did we get right, end to end, with no one having to touch them — and can the customer verify that count against their own books". Per-page pricing was the right unit for a category that sold throughput to a technical buyer. It is the wrong unit for a category that sells decisions to an economic one, because it meters the input the buyer stopped budgeting for and hides the outcome the buyer started paying for. The vendor who defines an auditable outcome unit, aligns the incentive so automation lifts revenue, guards the margin with a floor and a collar, and instruments the result so the bill is verifiable keeps the economic buyer through the renewal. The vendor who stays on per-page because the meter is easy keeps the meter and loses the deal.
At Cogneris we build for the outcome the buyer is actually paying for: document AI that runs extraction, validation and write-back to a decision a finance, insurance or accounting team can act on without reworking it. Our pay-per-page pricing is deliberately transparent on the input so the cost is never a surprise — and the same per-tenant telemetry that meters it is what lets us frame, and stand behind, the outcome underneath: which documents completed without a human, which were touched, and what the cost and margin per decision actually are. That instrumentation is the precondition for the whole conversation — you cannot price on a result you cannot measure, and it sits inside the operating model from day one rather than bolted on at the contract. If you are pricing a document workflow and want to pressure-test where the value sits — input, outcome, or somewhere between — talk to our team: bring the workflow and the mix, and we will walk the unit economics with you before anyone writes a unit into a contract.