Outcome based AI pricing for services only protects your company if the baseline and metric are set before work starts. See the 6 clauses you should ask for.
Published
Last Modified
Topic
AI Vendor Selection
Author
Jill Davis, Content Writer

TLDR: Outcome-based AI pricing ties a services firm's fee to a business result instead of hours, days or scope. It only protects the buyer when the contract fixes the baseline, the metric, the attribution rule and who does the measuring before work starts. Most deals labelled outcome-based leave those to the firm selling the work. This post explains how the model works for AI services, where it breaks, and the six clauses to ask for, set out as Assembly's 6-Clause Outcome Term Sheet.
Best For: COOs, CFOs, operations and procurement leaders who have been offered an AI consulting, implementation or managed-service engagement "tied to outcomes" and want to understand the model before negotiating it. Private equity operating teams will find a short section of their own near the end.
Updated: November 2026
Outcome-based AI pricing is a commercial model that ties some or all of an AI services firm's fee to a measured business result, such as claims processed without rework or days removed from the monthly close, instead of to the hours or people it took. Consulting firms, implementation partners and outsourcers are all moving toward it: McKinsey said in November 2025 that about a quarter of its global fees now come from outcomes-based arrangements, as Business Insider reported. It sounds like the obvious answer to two years of AI projects with little to show for them. But the label tells you almost nothing about who carries the risk. A handful of contract terms decide that, and most of this post is about them.
What is outcome-based AI pricing, and why are services firms offering it now?
Outcome-based AI pricing is a contract structure in which the services firm's fee moves with a business result the buyer cares about, measured against an agreed starting point. Firms are offering it in 2026 because AI has broken the link between hours worked and value delivered, and because buyers who have paid for pilots that went nowhere are no longer willing to pay for effort alone.
Why services firms are moving
Professional services have always been priced on time. AI compresses the time, so a firm that keeps billing by the hour gets paid less for the same result, and a firm that keeps its fee has to justify it some other way. Tying the fee to the result is the way most of them have chosen. In January 2026 McKinsey and AWS went as far as launching a joint venture built around outcome-based pricing, according to the California Management Review. Outsourcers are on the same path. Everest Group's view, reported by CIO.com in 2026, is that outcome-based pricing in business process services is moving from pilots to scaled adoption because AI makes outcomes measurable enough to contract on.
Why buyers are listening
Buyers have been burned. MIT's 2025 GenAI Divide study, covered by Fortune, found about 95% of enterprise AI pilots delivering little to no measurable P&L impact. Only 14% of the 950 business leaders in Grant Thornton's 2026 AI Impact Survey said they had fully integrated AI into operations. After a few engagements that ended with a slide deck and a demo, a fee that depends on the result is an easy thing to say yes to.
Actual adoption is still thinner than the marketing. Gartner figures cited by CIO Dive in 2026 show 19% of services buyers using outcome-based arrangements today, and Gartner expects fewer than 25% of technology services contracts to use them through 2031. A Gartner research note from July 2026 puts the uneven adoption down to problems with measurability, value and accountability. So when a proposal like this lands on your desk, the odds are that neither you nor the firm has negotiated many of them.
Where the idea comes from, and what it is not
Paying for results is old. Energy services companies have long been paid out of the savings they deliver, outsourcers have used gainsharing clauses for decades, and recovery auditors take a share of what they find. AI is the newest version because it makes more kinds of work countable.
Outcome-based pricing is different from a performance guarantee. A guarantee promises a result and pays damages if it is missed. Outcome-based pricing only changes what gets invoiced. If the result is missed you pay less, but nobody compensates you for the lost quarter.
How does outcome-based AI pricing work in a services engagement?
Outcome-based AI pricing works in one of three ways in a services engagement: the buyer pays a share of a measured gain, pays a conventional fee with part of it held back against a target, or pays per unit of work the firm completes on the buyer's behalf. Most real contracts combine a fixed base with one of these.
Share of a measured gain
The firm takes a portion of a saving or a revenue gain, measured against a baseline. It suits process redesign work, where the result is a change in a number the company already tracks, such as rework rate or collections. Incentives line up well when the baseline is sound, and this form produces the worst disputes when it is not.
Fee with a portion at risk
The engagement keeps a conventional fee, but part of it is withheld until a target is met, sometimes with a bonus above the target. It is the easiest form to budget for and the easiest to water down, since a small holdback changes very little about how the firm behaves.
Pay per unit of completed work
Here the firm runs the AI-enabled process itself and charges for each finished item, such as a processed invoice or an adjudicated claim. EY has described this as a "service-as-a-software" approach, in the same Business Insider report. It is simple to count. Its weakness is that a completed unit is still an activity, and more activity does not always mean a better business.
Outcome-based AI pricing vs. time and materials, fixed-fee and retainer models
Outcome-based AI pricing differs from the other services models in what the fee follows. Time and materials follows hours, a fixed fee follows scope, a retainer follows the calendar, and outcome-based pricing follows a result. Each one leaves a different risk with the buyer, and each one gives the buyer something different to police.
How the five models compare
Model | What the fee follows | Who carries delivery risk | What the buyer has to watch |
|---|---|---|---|
Time and materials | Hours or days worked | Buyer | Effort that expands, and no reason for the firm to finish faster |
Fixed fee | An agreed scope | Shared on effort, buyer on results | Deliverables accepted with no link to a result |
Retainer | Time elapsed | Buyer | Months that pass without a defined output |
Pure outcome-based | A measured result | Firm, on paper | Who defined the result, who measures it, and which workflows the firm agreed to take on |
Hybrid: fixed base plus outcome fee | Scope for the base, a result for the rest | Shared | The baseline, the attribution rule and the size of the at-risk portion |
Why the hybrid usually wins
Assembly's position is that buyers should not ask for 100% outcome-based pricing. We know that sounds backwards. A firm carrying all the risk will do one of two things: build a premium into the outcome fee, or accept only the workflows where results are easy to count and nearly certain. The hard, valuable work, the kind that touches several systems and teams, gets left out of scope. Gartner has warned the sellers themselves about this. A February 2026 research note on services argues that outcome-based pricing sounds appealing but that hidden complexity often makes it impractical.
Buyers have trouble with the pure version too. In an Andreessen Horowitz survey cited by BCG, 36% of buyers worried about how predictable their spending would be under outcome-based pricing. A fixed base that covers delivery effort, with a real portion of the fee riding on the result, keeps the firm working on the difficult problems and keeps the budget forecastable.
What should buyers actually ask for? The 6-Clause Outcome Term Sheet
Buyers should ask for six things in any outcome-based AI services contract: a signed baseline, a metric that sits on the P&L, a written attribution rule, measurement owned by the buyer, fees that can go down, and a term short enough to renegotiate. Assembly calls this the 6-Clause Outcome Term Sheet. A deal missing any one of them is outcome-based in name only.
1. A baseline both sides sign before work starts
The baseline is the contract. Every later argument about whether the AI worked is an argument about where things stood at the start, so the starting figures need to be measured, written down and signed by both parties before any build begins. That means establishing an AI ROI baseline for the specific workflow: volumes, cycle times, error rates and the hours people spend on it today. Everest Group names verified baselines and governance as what now holds these deals back. If the baseline takes four weeks to build, spend the four weeks.
2. A metric that sits on the P&L
Pick a metric the CFO already reports, or one that converts to a P&L line with a single agreed step. Services firms tend to propose activity measures such as hours saved or documents processed, because those can improve whatever happens to margin. Defining the metric is where most buyers get stuck: 47% in the survey BCG cites said they struggle to define clear, measurable outcomes. The wording matters as much as the choice. "Hours saved" and "hours removed from the budget" are different outcomes, and only one of them reaches the P&L.
3. An attribution rule written in advance
Results move for many reasons at once: a price increase, an acquisition, seasonality, a new manager. The contract has to say how the AI's share is separated from the rest, and it has to say so before anyone knows which way the numbers went. In the same survey, 25% of buyers found it hard to agree on how value is attributed, and 24% said outcomes often depend on things outside the provider's control. A workable rule names the comparison (a control group, a prior-period run rate or a volume-adjusted formula) and lists the events that trigger a reset. We have written separately about AI value attribution methods.
4. Measurement owned by the buyer's finance team
Whoever counts the outcome controls the invoice. TechTarget's 2026 reporting on outcomes-based deals found that sellers largely write the contract language defining an outcome, and that it is often unclear who controls measurement or how disputes get settled. Ask for the reverse: your finance team produces the number from your own systems, the firm gets read access to check it, and a named third party decides disagreements.
5. Fees that can actually go down
Gartner analyst Tom Coshow's test, as CIO Dive reported it, is whether the seller is taking on any risk at all. If it is not, the outcome label adds nothing. A lot of proposals fail that test. They keep the standard fee intact and add a bonus for good results, which gives the firm upside and the buyer nothing. Ask what the firm is paid if the outcome is missed entirely, and expect the answer to be less than a normal fixed fee for the same work. The base should cover the effort of delivery and no more. The firm earns its margin on the result. Expect pushback on this one.
6. A term short enough to renegotiate
An outcome formula that looks fair today can look generous to one side a year from now, once both parties know how hard the work really was. Analyst Brent Leary's advice to CIOs in the TechTarget piece was to keep commitments to 6 to 12 months while pricing is still changing, and that is a sensible ceiling for the at-risk measurement window too. Put a cap on the total outcome fee. The company should also own the workflows, configurations and data when the engagement ends, so it can avoid AI vendor lock-in.
Why do outcome-based AI deals go wrong?
Outcome-based AI deals usually go wrong in one of three ways: the firm chooses an outcome it can hit without changing the business, the measured result never reaches the P&L, or the buyer fails to deliver the access and people the firm needs. The AI itself is rarely the cause, and you can usually spot the problem in the contract or the operating plan before signing.
The firm picks the outcome
A firm offered free choice of metric will choose one it already knows it can hit. The contract then pays out on schedule while nobody on the buyer's side can point to a margin change. Clauses 2 and 4 exist to prevent this.
The result is real but never reaches the P&L
Saved hours only become savings once the hours are redeployed or removed. An outcome metric defined as "hours saved" can be met in full while the P&L stays flat, which is how a company ends up inside MIT's 95% while holding a contract that says the project succeeded.
The buyer cannot hold up its side
Outcome deals bind the buyer as well. The firm needs data access, process owners with time, and decisions made on schedule, and most contracts suspend the at-risk portion if those arrive late. Gartner predicted in 2025 that over 40% of AI agent projects will be canceled by the end of 2027, with unclear business value and weak risk controls among the reasons. A company that cannot staff its half of the deal will pay the base fee and collect nothing.
The hardest questions buyers ask
Buyers weighing outcome-based AI pricing tend to push back on the same points: a confident firm should take all the risk, a company without clean data cannot use the model, and firms will simply charge more. These are fair objections, and the answers below are the ones we would give a peer.
"If the firm believes in its work, why not put the whole fee at risk?"
Because you would get a narrower engagement. A firm with everything at risk scopes down to what it can control, and most of what moves the result sits outside any outside party's control: your data, your managers, your willingness to change a process. A shared structure keeps both sides working on the hard parts.
"We have no clean baseline. Does that rule us out?"
No, but it changes the order of work. Run a short paid diagnostic first, with the baseline as its deliverable, and sign the outcome terms after. A firm that offers to skip this and estimate the baseline is offering to set its own target.
"Won't the firm just charge more to cover the risk?"
Often, yes, and that can be acceptable. A higher fee for a delivered result can be a better trade than a lower fee for an attempt. What you should refuse is a premium with no downside attached, which is why clause 5 comes before any conversation about the upside.
Where to start before the next term sheet
The first step before negotiating outcome-based AI pricing is to pick one workflow and build its baseline, because every other clause depends on it. Buyers who arrive at the negotiation with their own measured starting point, their own metric and their own finance team ready to count can accept or reject a firm's offer on its merits.
It also helps to fold the six clauses into the standard questions to ask an AI consulting firm before signing, and into whatever scorecard you use to evaluate AI consulting firms. By the time anyone is looking at a dashboard, most of what decides whether the AI work shows up in the numbers was settled in the contract.
A note for private equity owners
PE-backed companies face the same model with a shorter clock. In the 2026 StepStone and Bain GP survey, 39% of general partners expected AI to have no material financial impact on their portfolio companies this year, which explains the appetite for fees tied to results. Two extra terms matter here. The measurement window has to close well inside the hold period, and the contract should say what happens to an ongoing outcome fee at a change of control, since the next owner will price it in. The metric should also be one the sponsor can use to track AI impact on EBITDA. In Accordion's Q2 2026 benchmark, only 38% of operating partners said they track AI's effect on revenue or margin.
Frequently Asked Questions
What is outcome-based AI pricing?
Outcome-based AI pricing is a commercial model that ties some or all of an AI services firm's fee to a measured business result instead of hours, days or scope. The result might be claims processed without rework or days removed from the monthly close. Gartner figures cited by CIO Dive in 2026 show 19% of services buyers using outcome-based arrangements today.
How is outcome-based AI pricing different from time and materials or fixed-fee pricing?
Outcome-based AI pricing charges for a result, while time and materials charges for hours and a fixed fee charges for scope. Under the older models the buyer pays whether or not the business improves, so the buyer carries the results risk. Under outcome-based pricing part of that risk moves to the services firm, provided the contract fixes the baseline and the buyer controls measurement.
What should buyers ask for in an outcome-based AI contract?
Buyers of outcome-based AI pricing should ask for the six terms in Assembly's 6-Clause Outcome Term Sheet: a baseline both sides sign before work starts, a metric that sits on the P&L, a written attribution rule, measurement owned by the buyer's finance team, fees that can go down as well as up, and a term short enough to renegotiate. A contract missing any one is outcome-based in name only.
Should a buyer ask for 100% outcome-based AI pricing?
A buyer should not ask for 100% outcome-based AI pricing in most cases. A services firm carrying all the risk either adds a premium or accepts only easy, certain workflows, which leaves the valuable work out of scope. A Gartner research note from February 2026 argues that hidden complexity often makes outcome-based pricing impractical in services. A fixed base plus a real at-risk portion usually serves both sides better.
Which outcome metrics work best in an outcome-based AI contract?
The outcome metrics that work best in an outcome-based AI contract are ones the CFO already reports, or that convert to a P&L line in one agreed step, such as rework rate, days to close or revenue recovered. Activity measures like hours saved are weak because they can improve regardless of margin. In the Andreessen Horowitz survey cited by BCG, 47% of buyers said they struggle to define clear, measurable outcomes.
Who should measure the outcome in an outcome-based AI deal?
The buyer's finance team should measure the outcome in an outcome-based AI deal, using its own systems. Whoever counts the outcome controls the invoice. The services firm should get read access to verify the figures, and a named third party should settle disputes. TechTarget reported in 2026 that sellers largely write the definitions in these contracts, which is the arrangement to avoid.
How do you set a baseline for an outcome-based AI contract?
A baseline for an outcome-based AI contract is set by measuring the target workflow before any build starts: volumes, cycle times, error rates and hours spent today. Both parties sign the figures. If the company cannot produce them, a short paid diagnostic should come first, with the baseline as its deliverable. Assembly's guide to establishing an AI ROI baseline covers the method.
How is attribution handled in outcome-based AI pricing?
Attribution in outcome-based AI pricing is handled by a rule written into the contract before results are known. The rule names the comparison method, such as a control group or a volume-adjusted formula, and lists events that trigger a reset, such as an acquisition or a price change. In an Andreessen Horowitz survey cited by BCG, 25% of buyers found it hard to agree attribution with providers.
Why do outcome-based AI deals fail?
Outcome-based AI deals fail for three main reasons: the services firm picks an outcome it can already hit, the measured result never reaches the P&L, or the buyer does not supply the data access and people it promised. All three can be spotted before signing. MIT's 2025 GenAI Divide study, covered by Fortune, found about 95% of enterprise AI pilots delivering little to no measurable P&L impact.
How long should an outcome-based AI contract run?
An outcome-based AI contract should usually commit the buyer for no more than 6 to 12 months at a time while pricing is still changing, which is the range analyst Brent Leary gave in TechTarget's 2026 reporting. The measurement window should be long enough to show a result in reported financials and short enough to renegotiate before the terms go stale.
What forms does outcome-based AI pricing take?
Outcome-based AI pricing takes three main forms in services: a share of a measured saving or gain, a conventional fee with a portion held back against a target, and a fee per unit of work the firm completes on the buyer's behalf, such as a processed claim. Most real contracts pair a fixed base with one of these. EY has called the per-unit form "service-as-a-software", Business Insider reported.
How common is outcome-based AI pricing in 2026?
Outcome-based AI pricing is growing in 2026 but is still a minority model. Gartner figures cited by CIO Dive show 19% of services buyers using outcome-based arrangements today, and Gartner expects fewer than 25% of technology services contracts to use them through 2031. McKinsey said in November 2025 that about a quarter of its global fees come from outcomes-based arrangements.
Is outcome-based AI pricing the same as a performance guarantee?
Outcome-based AI pricing is not the same as a performance guarantee. A guarantee promises a result and pays damages if it is missed; outcome-based pricing only changes what the firm can invoice. If the result is missed, the buyer pays less but still absorbs the lost time. Buyers who need a committed result should negotiate a guarantee or service levels as separate terms.
What is the first step before negotiating outcome-based AI pricing?
The first step before negotiating outcome-based AI pricing is to choose one workflow and measure its baseline. Every other term depends on that starting point. Buyers who bring their own baseline, metric and measurement process can judge a firm's offer on its merits. Those who do not end up accepting the firm's version of all three.
Who should own an outcome-based AI contract inside the company?
An outcome-based AI contract should be owned jointly by the executive who runs the workflow and the CFO. The workflow owner is accountable for supplying data and people, and finance owns the baseline and the measurement. Procurement alone is the wrong owner, because the terms that matter are operational. TechTarget reported in 2026 that business sponsors tend to favor these deals while procurement often resists.
What should an AI transformation partner commit to under outcome-based pricing?
An AI transformation partner working under outcome-based pricing should commit to a signed baseline, a metric on the P&L, measurement by the client's finance team, and a fee that falls if the result is missed. A partner that offers only a bonus on top of its normal fee is not sharing risk. The partner should also hand over workflows and data at the end of the term.
Legal
