Outcome-Based Pricing Works Only if the Buyer Does the Measuring

← Back · ACT II · Chapter 14: Who Controls Value Engagement?

The Value Standard holds that a purchase is governed only when the buyer can verify, on its own terms, that the value it paid for arrived. In The Trade You Priced I set out the one line that belongs in a contract before signature: what evidence will show the value arrived, who produces that evidence, by when, and what follows if it does not. This post is about the second of those clauses, who produces the evidence, because a pricing model now spreading through enterprise services answers that question on the buyer’s behalf, and answers it in the seller’s favor.

In July, Serge Tkach, writing as a contributor to CIO, described the AI-native services pitch he nearly signed. The agents demoed cleanly, and the pricing was the kind procurement loves: per resolved case rather than per seat. What he caught in time was who got to define the word resolved. On an earlier outsourced-support arrangement, the vendor’s reported resolution rate had looked excellent until his own team pulled the reopen numbers. In his words: “Auto-closed tickets had been counted as wins. Users had quietly stopped logging issues at all.” The dashboard was green and the service was failing underneath it.

That gap is now being priced across the enterprise. Everest Group wrote in December that the foundations are in place for outcome-based pricing to move from pilot programs to scaled adoption, that the real challenge now lies in mindset and governance, and that benefits attribution has become one of the most debated elements of outcome-linked contracts. Gartner predicts that more than forty percent of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value or inadequate risk controls. His own reading is that the providers are moving first and the contracts are changing faster than most buyers can govern them.

The word outcome makes the deal sound safe. Tkach puts the risk plainly: in the near term, outcome pricing “hands the advantage to whoever can measure the outcome,” and in most buying organizations that party is the vendor. When the vendor decides what counts as success and then reports and audits it too, the buyer has bought a number it does not own. Priced per unit, the arrangement leaves the risk of the outcome with the buyer and the evidence of it with the seller.

Where the incentive points

Read the outcome contract against the standard and the flaw shows at once. The vendor is paid on the reported outcome. The vendor also defines the outcome and measures it. Its incentive attaches to the number on the dashboard rather than to what the user receives, and the two diverge exactly where it costs the buyer: auto-closed tickets, suppressed reopen rates, demand quietly turned away. The contract can be met in full while the service degrades.

This has an ancestor on the site. Define Success Before the Vendor’s Product Defines It for You describes how the definition of value arrives pre-written in the seller’s template, and Software Now Reports Its Own ROI, and the Vendor Keeps the Record describes the platform version, where the vendor holds the transaction records and reports observed return. AI Token Pricing: The Seller Sets Both the Price and the Quantity found the same pattern in model pricing, where the seller decides how many tokens a task consumes. An outcome-priced service contract is the same asymmetry with a price attached to every reported unit.

Custody has to be built into the arrangement

The CIO piece gives the right advice and then stops one level short. It tells buyers to make resolved mean what their users experience, measured against a baseline they captured themselves and tied to metrics they own, and to treat auditability as a gate. All of that is correct. Framed as negotiating tactics, though, those gains erode as soon as the bargaining position shifts, which usually happens at renewal, after an outage, or when procurement wants a deal closed quickly. Custody of the number has to be a structural default. The baseline, the metric definition and the audit trail live on instruments the buyer owns, independent of the vendor, before anything is signed. The place to set that down is the request that goes out before any supplier bids: what the purchase must produce, how it will be measured and on whose instruments, the conditions that must hold, the evidence the supplier must supply, and how bids will be compared on it. Written by the buyer, that document is a Request for Value. If keeping the number means having the same argument every cycle, the buyer will eventually lose one, and the vendor’s definition of done becomes the buyer’s.

Four properties decide who owns the number

Custody is four conditions, and an outcome deal that fails any one of them is pricing a risk transfer as a service.

Visibility. The buyer can see the outcome it is paying for, at the level it is paying for it, on an instrument the buyer reads, and not only on the vendor’s dashboard.

Audit by default. The trace is immutable and the buyer’s from the start, rather than reconstructed after an incident from the vendor’s logs and slide decks. IBM’s 2025 breach report, the first edition to study security and governance for AI, found that 63 percent of the 600 breached organizations it studied either had no AI governance policy or were still developing one, and that 97 percent of those whose AI models or applications were breached reported lacking proper AI access controls. Its 2026 edition found that more than 20 percent of organizations had reported a breach targeting AI models or applications. A trace rebuilt after an incident, from the vendor’s records, cannot stand in for one the buyer held from the start.

Dismantle by default. When the vendor stops producing value, the arrangement dissolves on terms the buyer set, with data portability and knowledge-base ownership already the buyer’s, so that an embedded agent platform does not become, as Tkach warns it can, stickier than the staffed incumbent it replaced.

Permeable borders. The seam between the buyer’s operation and the vendor’s agents is where attribution goes to die, and Everest’s own finding is that attribution is one of the most debated elements of these contracts. Someone on the buyer’s side has to own that seam. If no one does, the green dashboard is the only version of events that survives.

The gap is older than AI

The distance between the contracted number and the lived service has been around a long time. Outsourced support ran on it for years. What AI changed, as Tkach puts it, is that outcomes became measurable enough to contract on, and the measuring instrument now sits inside the vendor. That is why the reckoning arrived this summer: the misalignment between reported and realized value became sellable, at scale, priced per outcome. New C-Suite Roles Coordinate Strategy but Own None of the Results describes the organizational version of the same failure, a role built to coordinate value that disclaims ownership of the result. The buyers who hold up under either version are the ones who can hold any vendor to a result on their own numbers.

Tkach closes on a practical step, and it is the right one: take one high-volume workflow you can measure, capture its baseline yourself, and instrument it on metrics you own. What I would add is where that gets written down. It belongs in the buyer’s own request, before the bids come in, with an audit trail the vendor cannot edit and that sits with the buyer. Once that exists, the choice between paying per seat and paying per outcome matters much less, because the buyer holds the number either way. A buyer who skips it tends to find out at renewal what Tkach found on his earlier support contract: when the vendor owns the measurement, the vendor’s count is the only one there is.

References: Serge Tkach, Your service vendors are being rebuilt around AI, CIO, July 14, 2026. Kunal Verma and Agnivesh Jha, Outcome-based metrics: the new value currency in BPO, Everest Group, December 9, 2025. Gartner, Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027, June 25, 2025. IBM, Cost of a Data Breach Report 2025, and its release of July 30, 2025. IBM, IBM Study: One in Four Malicious Breaches are AI-Enabled, July 29, 2026, reporting the 2026 Cost of a Data Breach Report.

Keep reading: ACT II · Chapter 14: Who Controls Value Engagement?