The seat stopped measuring the work.

The proxy that stopped tracking value

Per-seat pricing was never really about seats. It was about finding a number that grew when the customer got more value and stayed flat when they got less, and was easy to count. Headcount on a tool fit that job well enough for two decades: a company that put fifty people on a CRM was getting more out of it than one that put five, and the vendor billed accordingly. The model's whole legitimacy rested on that correlation between seats occupied and work done. Break the correlation and the price stops meaning anything — the customer is paying for a unit that no longer maps to what they receive.

The agent broke it from both ends. At the top end, a single person with an agent now produces the output that used to need a team, so the customer's value goes up while their seat count goes down — the vendor's revenue moves in the wrong direction relative to the value delivered. At the bottom end, the agent doing the work is not a person and does not need a login: it runs as a service, processes work in a loop, and never registers as the kind of user a per-seat contract knows how to charge for. The billable unit and the value-producing unit have decoupled. That is not a discounting problem a vendor can adjust for at renewal. It is the pricing model measuring the wrong thing.

The agent that never logged in

It is worth sitting with the specific mechanic, because it is what makes this different from every previous shift in software pricing. When a customer automates a workflow with an agent, the work still happens — invoices get coded, claims get adjudicated, documents get classified and routed — but it happens without a human in the seat the vendor is billing. The agent authenticates as a service account, does the volume of ten analysts, and shows up on the license report as, at most, one seat, or none. A vendor pricing that account by named users watches its revenue flatten while the amount of work its software is doing multiplies. The customer is not gaming the contract; the contract is simply blind to the way the work is now performed.

This is the quiet reason the seat model is not survivable rather than merely suboptimal. A vendor can hold the line on per-seat pricing for a while, but every customer that shifts work from people to agents is a customer whose bill stops reflecting what they use — and the more successful the vendor's own automation story, the faster it happens. The model punishes exactly the behaviour the product is supposed to enable. Meanwhile a competitor pricing the same account by the work performed bills more as the customer does more, and can point the buyer's finance team at an invoice that finally lines up with value. In a procurement conversation, "you pay for what the software actually did" beats "you pay for how many people we think are logged in," and the seat vendor loses the line item it can no longer justify.

Per-seat pricing charged for permission to do the work. The agent does the work without the permission, at a volume no seat count was built to hold — so the price has to move from who might act to what actually got done.

What replaced it: the hybrid, not the coin flip

The reaction to all this was not a stampede from pure per-seat to pure pay-per-outcome. That framing — seats versus outcomes, pick one — is the way the shift gets misread, and it leads vendors to overcorrect into a model that is impossible to forecast. What the market actually converged on through 2026 is a hybrid: a predictable base for the platform, plus a variable component priced by usage or outcome. The base gives the vendor a revenue floor and the customer a budget line they can plan around; the variable component is where the price finally tracks the work. It is now the most common structure among software companies repricing for the agent era, and the projections point the same way — a growing share of enterprise software spend moving to usage-, agent- or outcome-based terms over the rest of the decade, with the per-seat share shrinking rather than vanishing.

For a vendor whose product is document work, the hybrid maps cleanly onto what the software does. The base covers access to the platform, the templates, the integrations, the support — the things that have value whether or not a single document flows through this month. The variable component is the work itself, and the honest unit for it is not the page processed but the outcome produced: the decision approved, the case closed, the document extracted and accepted downstream. That is the same move we made the customer-facing case for in outcome pricing in document AI — charging for the result the buyer wanted, not the input they had to feed in. This piece is the other side of that coin: not how to present the price to the customer, but what the shift does to the vendor's own profit-and-loss, and how to run it without the variable component quietly eating the margin the base was supposed to protect.

The emerging frame, component by component

Vendors that have made the move without wrecking their margin converge on the same handful of pieces. None is exotic on its own; the discipline is in having all of them, because each closes a gap the others leave open — and the last two are the ones that separate a durable model from a billing experiment that loses money at scale.

• A billable value unit tied to a measurable, auditable outcome — the thing you charge for has to be something both sides can point at and agree happened: an approved decision, a closed case, a document processed and accepted downstream. If the unit is vague — "value delivered," "productivity gained" — it becomes an argument on every invoice. The unit is the contract; define it so precisely that whether it occurred is a fact, not an opinion.
• A per-outcome margin guardrail — because the cost of producing an outcome is now variable, driven by how much inference a hard case burns, the price of the outcome needs a floor of margin built into it and a monitored ceiling of cost. When a document class starts costing more to process than expected, the guardrail is what flags it before a quarter of negative-margin outcomes has already shipped. This is the piece most first attempts skip, and the one that turns outcome pricing from a growth story into a loss.
• A base-plus-variable structure — the predictable base that covers platform value regardless of volume, and the variable component that follows the work. The base protects the vendor from a quiet month and gives the customer a number they can budget; the variable part is where the price tracks value. Pure-variable pricing looks purest and forecasts worst — for both sides.
• Outcome telemetry as a billing prerequisite — you cannot charge for an outcome you cannot measure, and you cannot defend the charge to a customer who cannot see it. The instrumentation that records every billable outcome — what it was, when it happened, on which case — is not a reporting nicety added later; it is the thing the invoice is built on, and it has to exist before the first outcome is priced.
• A seat-to-outcome migration path — most vendors are not pricing a greenfield account; they are moving an existing customer off a per-seat contract without rupturing the relationship or spiking the bill. That means a mapped transition — often the base pegged near the old seat spend, the variable layered on top — so the customer sees continuity, not a renegotiation that invites them to shop the market. How you migrate is as much of the model as what you migrate to.

Four ways outcome pricing goes wrong

The failure modes are as predictable as the components, and each is a version of keeping the appeal of pricing-by-result while dropping the discipline that makes it survivable.

• Charging for the outcome without measuring its cost — the vendor sets an attractive per-outcome price, wins the deal on alignment, and never instruments what each outcome costs to produce. Then a hard document class, or a customer whose cases all need deep reasoning, starts generating outcomes that cost more in inference than the price collects, and the vendor is now paying to serve its best customers. Outcome pricing without cost-per-outcome telemetry is not a pricing model; it is a bet that the average case stays cheap, placed without a way to see when it stops. This is the same token tax that pulls agentic gross margins down — except here it lands straight on the price the customer agreed to.
• Holding per-seat while the competitor charges for results — the vendor sees the risk in variable margin and decides to wait, keeping the safe, forecastable seat model one more renewal cycle. Meanwhile the buyer's finance team is getting a proposal from a competitor whose invoice lines up with the work performed, and the seat vendor is now defending a price that measures logins in a world that measures outcomes. Waiting feels prudent and is the slower way to lose the account — the model does not fail loudly, it just stops winning the board.
• An outcome defined loosely enough to dispute — the contract says the vendor charges per "successful extraction" or per "resolved case," but never nails down what counts. Is a document processed and then corrected by a human a success? Is a case the agent closed and the customer reopened billable? Every ambiguity becomes an invoice dispute, and a pricing model that generates monthly arguments about what the customer owes destroys the trust the alignment was supposed to build. The outcome has to be defined to the point of boredom, or it becomes a standing fight.
• No telemetry the customer can audit — the vendor measures outcomes internally but gives the customer no way to verify the count, so the invoice arrives as a number the buyer has to take on faith. In a per-seat world the customer could check the user list; in an outcome world they are being asked to pay for events they cannot see. Without a customer-auditable record of what was billed and why, the model asks for a trust that pricing-by-result was supposed to earn, not assume — and the first surprising invoice is the last one before the renewal goes to committee.

The trade-off, stated honestly

The reason per-seat lasted so long is that it was easy to forecast on both sides of the table. The vendor knew its revenue from the user count and its cost from the infrastructure; the customer knew its bill from its headcount. Everyone could plan. Outcome pricing trades that predictability for alignment: the price finally tracks the value delivered, but revenue now moves with the customer's volume and margin now moves with the cost of each outcome — two variables the seat model held still. That is a real trade, not a free upgrade, and pretending otherwise is how vendors back into a model they cannot forecast and cannot defend to their own board.

So the honest framing is not "abandon predictability for alignment." It is that the hybrid exists precisely to buy back most of the predictability while capturing the alignment — the base is the forecastable floor, the variable component is where the price tracks the work, and the margin guardrail is what keeps the variable side from becoming a source of loss. Get the base right and a quiet month does not threaten the business. Get the guardrail right and a hard customer does not quietly become unprofitable. Skip either and you have taken on all of outcome pricing's variance for none of its protection. The vendor that does this well is not the one that swings hardest from seats to outcomes; it is the one that prices the base to survive the floor and the outcome to survive the cost — and knows, per document class, which one is carrying the account. This is the same margin discipline the whole sector spent 2026 relearning, applied to the pricing model itself.

Closing thought

Every era of software has priced against the best available proxy for value, and every proxy eventually breaks when the technology moves. The seat was a good proxy for as long as work required a person in front of the tool. The agent ended that, and it ended it fast enough that the market repriced the whole sector before most vendors had changed a single line item. The mistake now is to treat this as a marketing decision — rebrand the price sheet, call it "outcome-based," and hope the margin holds. It is not a marketing decision. It is a P&L decision, made at the level of what unit you charge for and whether you can measure what that unit costs you to produce. The vendor that gets it right is charging for the same result the buyer wanted all along, with a base that protects the business and a guardrail that keeps the price above the cost. The one that gets it wrong is either defending a seat count the work has already left behind, or charging for outcomes it never learned to cost — and both of those bills come due at the same renewal.

At Cogneris we priced for the outcome from the start, and the reason we can is that the platform is built to measure it. Every document that flows through is processed against a versioned template, and what happens to it is recorded as a discrete, auditable event: extracted and accepted, held for review, escalated, corrected. That record is what makes an outcome billable in a way both sides can see — the customer can audit exactly which decisions they paid for, and we can see exactly what each one cost us to produce, per document class, so the price stays above the cost even when a hard class runs expensive. It is cost per successful task read from the other side of the invoice: the same telemetry that tells a buyer their unit economics tells us ours. If you are moving your own product off per-seat and want the pricing model to rest on outcomes you can measure rather than ones you have to hope stay cheap, talk to our team.