Four Numbers, One Cause
Four numbers have come out of four different corners of the market, and nobody put them in the same sentence.
Start with OpenAI. The Information reports — and this has not been independently confirmed — that the company has quietly begun letting a select group of its largest enterprise customers pay only when an AI task actually completes successfully. Pay for the result, not for the attempt. OpenAI has not announced it, has not said which customers, and has not disclosed the terms. What is known is that it exists, and that it exists for the biggest accounts in the book.
Now the shape of that practice across the wider market. Gartner has counted, and CIO Dive carried the figures. Nineteen percent of services buyers currently have outcome-based arrangements. On the seller side, thirteen percent of service agreements carry them. More of the market wants this than has signed it. Hold onto that. Then a number from a completely different body of research.
The National Bureau of Economic Research surveyed nearly six thousand senior executives across the United States, the United Kingdom, Germany and Australia and asked what AI has actually done inside their companies. Nine in ten reported no impact on their own firm’s employment over three years. On labor productivity, eighty-nine percent said the same. Six thousand executives, three years of spending, and the effect does not show up in the measurement. The layoffs have continued regardless.
And one more, this one from inside the channel.
Techaisle mapped how small and mid-market customers move through AI adoption and broke it into four stages. Demand is real at every one of them. What is not consistent is who can serve it. Partner capability is strongest at the first stage, the foundational estate work, and it thins as the customer moves forward, until at the most advanced stage it approaches zero. The further a client gets, the fewer providers exist who can take the next piece of work.
So. The largest buyers in the market are being offered payment on results. The appetite for those terms runs ahead of the contracts that carry them. The productivity that would justify any of it does not appear in the measurements. And the ability to deliver runs out exactly as the customer gets more ambitious.
Four facts. One of them is producing the other three. And it is not the one you would pick.
What You Ask For When You Can’t Tell
Outcome-based pricing is not what a market does when it matures. It is what an executive does when they cannot tell.
Consider the position that executive is actually in. Three years ago they stood up in front of a board, or a staff meeting, or an earnings call, and committed to artificial intelligence. Budget moved. Headcount moved. In a lot of cases people were let go and the reason given was this technology. That commitment is on the record, and it cannot be walked back.
And they cannot demonstrate the return. Thomson Reuters put a number on it — and note where this one comes from, because Thomson Reuters sells AI products to exactly the professionals it surveyed. In their research, ninety-one percent of eighteen hundred professionals said their own organization is falling short on delivering AI value. That is not a finding that the technology does not work. It is a finding that the people inside these companies cannot see whether it did.
So the buyer changes what they are purchasing. Not the work — the certainty. If I cannot prove this paid off, I will stop paying until somebody else proves it. That is the whole appeal of paying on outcomes, and wanting it is completely rational.
Here is what it does not do. It does not create the measurement.
Forrester’s Abhijit Sunil, writing in Channel Dive about outcome-based contracts in sustainability work, describes a constraint that travels well past his subject. Organizations lack reliable starting baselines. Data is fragmented across clouds and suppliers. And — this is the line — results often depend on customer decisions as much as on provider performance. He says the model works only in tightly defined initiatives where the provider controls the relevant levers and the customer has a credible baseline. The instrument requires the exact thing that is missing.
So outcome pricing does not solve the measurement problem. It relocates the obligation to solve it. Think of a homeowner who cannot tell whether the roof was actually fixed. They cannot get up there, they would not know good work if they saw it, so they stop trying to evaluate the work at all and say: I will pay you when it stops leaking. That is a completely reasonable thing to say. It also hands the roofer the entire question of what fixed means. The buyer who could not prove the value now buys from someone contractually required to.
Which leaves one question standing, and it decides everything after it: who in this chain is allowed to say no. OpenAI can hand risk back to a customer, absorb the occasional failure across thousands of accounts, and still be the only place that customer can go.
Not everybody has that. The further down the chain you go, the less of it there is.
Nobody Underneath You
Here is where that lands on you. You are the end of the chain. There is nobody underneath you to hand it to.
When a client asks to pay on results, you cannot pass that along to Microsoft, or to your RMM vendor, or to the model provider — every one of whom sold you their product with the outcome carefully excluded. Whatever you agree to, you hold. That is not a warning. It is just the shape of the seat you occupy. Techaisle put a name on this when they mapped those four stages. At the far end, they wrote, being accountable for what an autonomous system does means underwriting an outcome — and the overwhelming majority of the channel has no mechanism to price risk. Not no appetite for it. No mechanism. That is a gap somebody is going to sell into, and it will not be sold to you on your terms.
Which makes one thing decisive, and it is not whether you accept outcome terms. It is whether you were in the room early enough to define what the outcome was.
Look at who is getting that seat right now. The Next Web reports demand for fractional engineering talent up nine percent in ninety days, and the arrangement named in the piece is the outcome-based retainer. Daniel Kirichanski, who runs one of these practices, puts it plainly: senior technology expertise creates value through decisions rather than hours logged, and an outcome-based structure gives the client a more direct way to measure what that leadership contributed. Notice what that person is really selling. Not a deliverable. A judgment, priced against a result. And they can price against a result because they sit next to the decision. They are in the room when the goal gets set, so they help set it.
Then look at Okta’s most recent quarter. The company says partners were involved in all twenty of its largest deals, driven by demand for AI agent security. Twenty out of twenty. Those partners are not in those rooms because they resold a license. They are there because they brought a capability into a conversation that had not been decided yet.
Those are the same position, described two different ways. Present before the definition is written.
So here is the choice, and you will make it whether or not you make it on purpose.
Decide now which outcomes you are willing to be paid against, and pick only the ones where you control every input that produces them. Write them down. Put your own number next to each one. Take them to clients before a client brings you theirs.
Or wait — and let the first client who asks write the definition for you, and find out you have guaranteed a result that depends on their staff, their process, and a model that neither of you owns.
There is a much smaller way to start than the one you are picturing.
Because the first outcome you price on should be the smallest one you are certain of, not the most impressive one you can imagine. Pick a single line — a response time, a recovery objective, an onboarding window — attach money to it, and run it for a year with a client who already likes you. You are not repricing the business. You are finding out what it costs you to be wrong while the number is still yours to pick.
What to Consider
Price the first one off your worst month, not your average one. Pull your own last twelve months of data on whichever line you pick and find the month you performed badly, because that is the month that will decide whether this was profitable. Most providers set the number against a typical month and discover their variance for the first time in the month it costs them. The point of a small pilot is to buy that discovery cheaply, so price it to survive the bad one.
Write the client’s inputs into the agreement as conditions, at signing. Every outcome you could sell depends on something the client does — approving changes on time, replacing the hardware you flagged, keeping somebody in the seat on their end. Name those in the contract as conditions of the guarantee while the conversation is friendly and hypothetical. If you leave them out, you will be raising them for the first time in the month you missed the number, where they will sound like excuses rather than terms.
Keep it as one line, not as the shape of the whole contract. Sell the outcome line inside an otherwise ordinary agreement so that if it goes badly it costs you a line item rather than the relationship. It also leaves you a control group — the rest of the contract prices the normal way, so at the end of the year you can see exactly what the outcome line did to your margin instead of guessing. A provider who can say what one guaranteed outcome cost them is in a materially different negotiating position than one who has only ever guessed at it.
And picture the provider who ran the small version of this a year ago. They have one line, in one contract, priced against a result for four straight quarters — and they know exactly what it cost them. The two months it went badly. What those months did to the margin. What the client said afterward. So when a larger prospect turns up next spring with an outcome number already written into a draft, that provider is the only one in the running who can say, from their own books, whether the number is survivable. Everybody else is guessing, and it shows.
If this trend continues: by the middle of 2027 the outcome number will be arriving in the client’s first draft of the agreement rather than in yours, and the providers who never picked one will be negotiating against a figure somebody else chose for them.