Your vendor's exit is your problem now.

The twelve months that consolidated a category

Consolidation in document AI did not arrive as one headline deal. It arrived as a pattern, and the pattern is what matters more than any single transaction. Over the past year the acquisitions clustered into four recognisable moves. RPA majors — companies whose automation story stalls the moment a workflow hits an unstructured document — bought document-compliance and extraction vendors outright, including targets that had raised hundreds of millions and were, on paper, supposed to be consolidators themselves. European enterprise-content groups, sitting on large installed bases and thin AI roadmaps, bought tax- and invoice-focused OCR houses to keep their regulated customers from looking elsewhere. Data-cloud and analytics platforms acquired agentic document capabilities because their customers were already piping documents at them and the build-it-internally clock was too slow. And the ingestion vendors that feed LLM pipelines bought adjacent startups — parsing, notebook tooling, evaluation — to widen a wedge before someone wider bought them.

Each acquirer bought for a different reason, but every deal removed the same thing from the market: an independent vendor whose roadmap a customer was depending on. That is the part the trade press underplays. An acquisition is reported as a milestone for the vendor and a strategy move for the acquirer; for the customer mid-contract, it is an unplanned dependency change in a system that touches their invoices, claims, loan files or compliance evidence. The wave is not a story about founders getting outcomes. It is a story about a market deciding, fairly abruptly, that the standalone document AI feature is worth more inside a bigger platform than outside one — and about every buyer who now has to work out whether their vendor is the platform or the feature.

An acquisition is a milestone for the vendor and a strategy slide for the acquirer. For the customer mid-contract, it is an unplanned change of dependency in the system that processes their most regulated documents — announced to them last.

The thesis: under $50M, horizontal, unanchored — a target

Strip the deals down and the underlying thesis is uncomfortable in its simplicity. Extraction accuracy — the thing horizontal IDP vendors spent a decade competing on — stopped being a moat when general-purpose models made very good extraction available to anyone with an API key. What is left as defensible is everything around the extraction: the vertical workflow it sits inside, the decision logic and audit trail a regulated customer needs, the distribution that a platform partnership or installed base provides, and the data advantage of running deep in one domain. A vendor that has none of those — horizontal by design, sub-scale by funding-market standards, anchored to no vertical and partnered with no platform — is not priced as a business anymore. It is priced as a feature with a customer list attached, and features with customer lists are exactly what acquirers in this market are shopping for.

This is the same economic logic we wrote about from the buyer's side in the renewal-cycle consolidation piece — the budget collapsing from five point tools into one platform that closes the loop — now playing out at the level of the companies themselves. When the buyer consolidates spend, the revenue of the narrow vendor erodes; when the revenue erodes, the next funding round reprices or disappears; and the acquirer's offer starts to look less like an option and more like the plan. The $50M ARR line is shorthand, not a law — but it marks roughly where a vendor stops being able to fund the platform breadth the market now demands, and starts needing someone else's platform to live inside. By Q4 2026, the vendors below that line without an anchor will mostly have chosen — or had the choice made for them.

Consolidator, target, survivor

The useful way to read the wave is not deal by deal but profile by profile. Three roles keep recurring, and which one your vendor occupies tells you more about your next two years than anything on their pricing page.

The founder-side implication falls straight out of the table: verticalise or be absorbed. Both are legitimate outcomes — being acquired by the right consolidator is a fine end for a feature company and often the right one for its customers too. What is no longer available is the middle position: staying independent while horizontal, sub-scale and unanchored, on the strength of being slightly more accurate. The investors have already moved; mid-stage money is flowing to vertical specialists because vertical depth is the only profile on the table that compounds instead of commoditising. A founder still pitching "best horizontal extraction" in 2026 is pitching the target row and calling it the survivor row.

What the wave changes for the buyer

If you buy document AI rather than sell it, the consolidation wave rewrites three parts of your job that used to be routine. None of them is about picking better features; all of them are about pricing in the probability that the company across the table is a different company before your contract ends.

• Vendor-viability due diligence became strategy — the question "will this vendor exist, independently, in 24 months" used to be a procurement formality answered with a funding announcement. It is now the most consequential question in the evaluation, and it has to be asked structurally: where does this vendor sit on the consolidator–target–survivor map, what is their vertical anchor, who needs to buy them and why. We made the same argument about model providers; it now applies one layer up, to the vendor assembling those models into your workflow.
• The continuity clause became baseline — change-of-control terms used to be lawyer comfort. In a consolidating market they are the difference between a managed transition and a hostage negotiation: price protection through the current term, a guaranteed support window after any acquisition, source-and-data escrow or an export path if the product is sunset, and data portability in a documented format you have actually tested. A vendor that resists these clauses is telling you which row of the table they expect to occupy.
• The vertical platform earned its premium — when every vendor's independence was assumed, the horizontal toolkit's flexibility looked like the safer buy. With consolidation priced in, the calculus flips: the vertical platform with a real decision engine is the profile the wave does not wash away, and paying its premium is partly paying for roadmap continuity. The cheaper horizontal tool is only cheaper until its acquirer reprices it inside a bundle you did not choose.
• The acquisition question entered the RFP — alongside SLA, security and data residency, the RFP now needs one more standing question: what happens to this product, this contract and this data if you are acquired? A serious vendor has a real answer — contractual commitments, an architecture that exports cleanly, an honest account of who their plausible acquirers even are. A vendor who treats the question as insulting has not read their own market.

The four ways the wave gets misread

Consolidation is visible enough that almost nobody denies it. The failures come from misreading what it implies, and four misreadings account for most of the damage on both sides of the table:

• Treating the acquisition announcement as good news by default — "our vendor was acquired by a bigger company, so the product is safer now" is true only when the acquirer bought the product. Often they bought the engine, the team or the customer list, and the product you run is the part with a sunset date. The day the deal is announced is the day to ask which of those was actually purchased — not eighteen months later when the migration notice arrives.
• Diligencing the demo and not the balance sheet — evaluation processes still spend ninety percent of their effort on accuracy bake-offs and integration fit, and a courtesy slide on the company. In this market the company is the risk. A technically superior tool from a textbook target profile is a worse buy than a slightly less polished one from a survivor — because you are not buying this quarter's accuracy, you are buying two years of someone else's roadmap decisions.
• Founders mistaking funding for an anchor — a large round extends the runway; it does not change the profile. The past year's deal list includes targets that had raised more than most survivors ever will. Capital buys time to verticalise or to partner; spent instead on scaling a horizontal feature, it mostly raises the eventual purchase price — for someone else.
• Buyers waiting for the wave to finish — "we'll choose a vendor once the market settles" sounds prudent and costs more than it saves. Consolidation waves take years to resolve, the work the documents represent does not pause, and the strongest survivors get capacity-constrained and choosier about whom they onboard. The hedge is not waiting; it is choosing for survivability and contracting for the scenario where you chose wrong.

The trade-off, stated honestly

It would be convenient for us to tell you consolidation is simply bad and independence simply good. It is not that clean. Consolidation delivers real benefits to real customers: a document AI capability inside a suite you already run is one less vendor to manage, one less integration to maintain, one less security review to repeat — and an acquirer with deep pockets can fund engineering that a struggling independent never could. For a buyer whose document volumes are modest and whose workflows are generic, the absorbed feature inside a consolidator's bundle may genuinely be the right answer, and a vendor who pretends otherwise is selling you their fear, not your interest.

The cost lands on a narrower group, and it lands hard: the customers for whom the document workflow is core, regulated or competitive. They are the ones who bought a roadmap, not a feature — who need the audit trail to deepen, the decision engine to keep learning their domain, the vendor to say no to the bundle in favour of the workflow. For them, the acquirer's rational behaviour — integrate the engine, standardise the stack, sunset the edges — is precisely the erosion of everything they paid a premium for. The honest statement of the trade-off is this: consolidation trades the upside of a focused roadmap for the safety of a balance sheet, and which side of that trade you should want depends entirely on whether document intelligence is plumbing or product for your business. Decide that first; it decides everything else.

Why "what happens if you're acquired" belongs next to the SLA

The SLA question earned its permanent place in the RFP because downtime is a risk every buyer eventually experiences. Acquisition risk has now reached the same status in document AI, for the same reason: at current deal velocity, a meaningful fraction of the vendors answering your RFP this year will not be independent when the contract you are negotiating expires. That is not an insult to any of them; it is arithmetic. And as with the SLA, the value of asking is only partly in the answer — it is in watching how the vendor handles the question. The one who has thought about it gives you specifics: here is the change-of-control language we sign, here is how your data exports and in what format, here is the escrow arrangement, here is — frankly — the kind of acquirer that would make sense for us and the commitments that would survive that deal.

The same logic that made margin discipline a board metric applies here from the buyer's chair: a vendor's unit economics are now part of your risk surface. A vendor losing money on every page processed is one funding winter away from the target column no matter what their roadmap deck says, and you are entitled to ask the uncomfortable question. Survivors will answer it gladly — sustainable economics are the cheapest differentiation in a market where most competitors cannot claim them. Evasion, in 2026, is data.

Closing thought

Every maturing software category goes through this: the funding wave plants a hundred vendors, the capability commoditises, and the market sorts the platforms from the features through M&A. Document AI is simply taking its turn, compressed into a faster cycle because the model layer commoditised the core capability faster than anyone's five-year plan assumed. The wave itself is neither good nor bad; it is weather. What is in your control is the position you hold when it passes — as a founder, whether you own a vertical loop someone must buy at a premium or a horizontal feature someone can buy at a discount; as a buyer, whether your contract assumes your vendor's independence or prices its end. The consolidation will be finished in a few years either way. The contracts signed this year will still be running when it is.

At Cogneris we sit inside the market this piece describes, so here is our own answer to the question we just told you to ask. We built for the survivor column on purpose: document AI anchored in the workflows of finance, insurance and accounting teams — extraction, validation, decision and write-back with the audit trail carried through every run — priced pay-per-page on economics that do not depend on the next funding round. And we sign the clauses we recommend: change-of-control protection, tested data export, continuity commitments in the contract rather than in the sales call. If you are evaluating document AI vendors and trying to work out who on your shortlist is a platform, who is a feature, and what to put in the contract for the day you find out — talk to our team: bring your shortlist and your draft terms, and we will walk you through the diligence questions we would ask in your chair, including the ones about us.