Only 27% of enterprise software buyers now prefer outcome-based pricing, but fewer than one in five still want classic per-user licensing. The gap between what buyers say they want and what they actually sign is where the next three years of vendor competition will be decided.
For thirty years, enterprise software and enterprise services were priced on inputs. Seats, licenses, hours, headcount. The logic held because effort was a reasonable proxy for value: more people working longer produced more output, and buyers could audit the invoice by counting the inputs.
AI severs that link. When a system resolves a support ticket in four seconds, the hours consumed tell you nothing about the value of the work. Futurum Research's 1H 2026 Enterprise Software Decision Makers survey found that 43% of buyers now prefer consumption-based models and 27% prefer outcome-based structures, while fewer than one in five still favor classic per-user pricing. Futurum's conclusion was blunt: vendors restricted to seat-only pricing risk immediate disqualification.
Yet the contracts being signed have not caught up. Research published in July 2026 by HFS Research and EY, based on a survey of 304 Global 2000 executives and interviews with procurement professionals, found that 72% of enterprises are comfortable with outcome-based models but only 39% actually use them. Seventy-three percent still frequently write time-and-materials contracts for AI engagements. This article covers what outcome-based pricing actually means, why the shift is happening now, where it breaks, and how to structure an agreement that survives contact with reality.
What Is Outcome-Based Pricing for AI?
Outcome-based pricing charges the buyer when a defined result is achieved rather than for the time, seats, or compute consumed getting there. A resolved support ticket, a qualified lead, a processed claim, a closed reconciliation. Deloitte's June 2026 accounting guidance identifies four common structures: per-successful-outcome fees, performance-tier pricing, fixed fees with performance adjustments, and prepaid outcome credits.
The distinction that matters is not pricing versus free. It is where the risk sits. Under time and materials, the buyer carries delivery risk: if the work takes twice as long, the buyer pays twice as much. Under outcome pricing, the vendor bears the risk. That transfer is the entire product, and it is why buyers want it and why many vendors quietly resist it.
Why Is AI Breaking the Billable Hour Now?
Three forces are converging, and none of them will reverse.
Effort stopped predicting value. HFS and EY put numbers on the expectation gap: 40% of enterprise buyers expect AI-led services to cost 10% to 30% less than traditional delivery, and 34% expect reductions of 30% or more, with an average expected reduction of 19%. A buyer who believes AI makes the work cheaper will not accept an invoice that bills the same hours as before. The hour is no longer a credible unit of account.
The professional services model is already moving. Roughly one quarter of McKinsey's global fees now come from performance-based arrangements, according to remarks by the firm's UK managing partner reported in late 2025. When the firm that invented the billable-hour pyramid starts putting a quarter of its revenue on results, the argument that outcome pricing is impossible for complex knowledge work has lost its best defender.
Vendors have published the price. Intercom's Fin charges $0.99 per resolution and defines a resolution precisely: no further help is requested after Fin's last answer. Zendesk and Decagon price the same way. Once a competitor publishes a per-outcome number, every buyer in that category has a benchmark, and every rival who explains why their pricing must be per seat sounds like a firm protecting its own economics.
What Does Outcome Pricing Do to Your Margins?
This is the part vendors underweight, and it is why the shift is slower than the survey data suggests.
Kyle Poyar's 2026 State of B2B Monetization report, based on surveys of 230 or more B2B software and AI companies conducted in April and May 2026, found that the median target gross margin for AI products is 50%. Only 12% of companies target the 80% or higher margins that defined the SaaS era. Outcome pricing amplifies that compression, because the vendor now absorbs the cost of every failed attempt. If your agent burns tokens on four conversations to resolve one, you are paid for one, and you pay for four.
That math changes what you are allowed to promise. Outcome pricing is only viable where the outcome is (a) measurable without a dispute, (b) achievable at a success rate you can predict, and (c) attributable to your system rather than to twelve other things happening at the same time. Fail any one of those tests, and the model becomes a mechanism for transferring your margin to your customer.
It also explains why pure outcome pricing is not where the market is landing. HFS and EY found that 82% of enterprise buyers prefer hybrid structures that combine a fixed fee with outcome or usage components. The same report from Poyar found hybrid pricing is now the single most common model at 37%, up from 25% a year earlier. Hybrid is not a compromise. It is the structure that lets both sides share risk instead of one side eating it.
Where Does Outcome-Based Pricing Break?
Attribution. A support agent deflects 40% of tickets. How much of that is the agent, and how much is the help-center rewrite that shipped the same month? Buyers who cannot answer this will dispute the invoice. Vendors who cannot answer it will lose the dispute.
Definition drift. "Resolution" sounds objective until you ask whether a customer who abandons the chat resolved anything. Every outcome definition needs an explicit exclusion list written before the contract is signed, not after the first billing cycle.
Measurement infrastructure that does not exist yet. Outcome pricing assumes both parties can observe the outcome. Most enterprises cannot. HFS and EY found that only 13% use a formal, enterprise-wide framework to evaluate AI-led proposals, while 41% assess them case by case without structured governance. You cannot pay for results you are not instrumented to see, which is the same measurement problem we mapped in our analysis of where AI returns actually show up.
Revenue recognition. Deloitte's June 2026 technology spotlight walks through the accounting problem: an outcome-priced agreement may be a stand-ready obligation recognized over time, or a promise to deliver a specified quantity of results recognized as they occur. Getting that classification wrong is a restatement risk, not a pricing quibble. Finance needs to be in the room when the commercial model is designed.
How Should You Structure an Outcome-Based AI Agreement?
Define the outcome in engineering terms, not marketing terms. The definition should be something a system can emit as an event, with a timestamp and a payload. If your outcome cannot be logged automatically, it cannot be billed automatically, and you have written a dispute rather than a contract.
Price a floor, then price the outcome. A fixed platform or retainer fee that covers your delivery cost, plus a per-outcome component that captures upside, matches what 82% of buyers say they want, and protects the vendor from a bad quarter of model performance. This is the same logic that governs how engagement models are structured, as we covered in our guide to AI staffing models.
Instrument before you commit. Run the engagement on a shadow meter for one quarter. Log what the outcome definition would have produced, compare it against what both parties believe happened, and fix the definition before money moves. Building that measurement layer is engineering work, and it belongs in scope from day one of any AI development engagement.
Start where attribution is clean. The best first candidates are workflows with a hard, countable result and a single owner. A credit decision engine that scores an application either scores it or does not, which is why the outcome logic was legible from the start in our AI credit scoring engine work. Content production has the same property: pieces produced and shipped are countable, as we saw when building generative AI for marketing operations. Save the ambiguous, multi-touch outcomes for after you have proven the meter.
Put an agent behind the outcome, not a headcount. If the commercial promise is a result, the delivery mechanism has to be a system that produces results at a predictable rate. That is a different build from a staffing plan, and it is why tailored AI agents and AI orchestration are becoming the substrate for outcome-priced offers.
Cap your downside explicitly. Volume ceilings, success-rate floors below which the fixed fee applies, and a renegotiation trigger if the underlying model economics move. Token prices have fallen fast, but they are not guaranteed to keep falling, and a contract written on the assumption that they will is a bet, not a business model.
Common Questions About Outcome-Based Pricing
What is outcome-based pricing?
Outcome-based pricing charges a customer only when a defined result is achieved, such as a resolved support ticket or a processed claim, rather than for seats, hours, or compute consumed. It shifts delivery risk from the buyer to the vendor, which is why buyer preference for it is rising while vendor adoption lags.
How many companies actually use outcome-based pricing?
About 39% of enterprises use outcome-based pricing for AI services today, according to July 2026 research from HFS Research and EY covering 304 Global 2000 executives, even though 72% say they are comfortable with the model. Hybrid structures combining a fixed fee with outcome components are the most common landing point, preferred by 82% of buyers.
Is outcome-based pricing better for the buyer or the vendor?
It favors the buyer on risk and the vendor on upside. Buyers stop paying for effort that produces nothing. Vendors who can hit a high success rate at low marginal cost earn more than an hourly rate would allow. Vendors who cannot predict their success rate lose money, which is why median target AI gross margins now sit around 50% rather than the 70% to 80% typical of subscription software.
What should be measured in an outcome-based AI contract?
A single, machine-loggable event with an agreed exclusion list, a baseline measured before the system goes live, an attribution method both parties accept in writing, and a review cadence. If the outcome cannot be emitted as a timestamped event by a system that both parties can audit, it is not ready to be priced.
If your commercial model still bills for effort while your delivery model runs on AI, the two are going to collide at your next renewal. Talk to the Golabs team about defining outcomes you can measure, instrumenting them properly, and building the systems that can actually be priced on results.