Most product metrics are internal. Define "active user" generously and the dashboard looks better, the quarterly review goes smoothly, and nobody anywhere receives an invoice because of it. Outcome pricing removes that slack. When a vendor charges per resolved conversation, the definition of "resolved" is no longer an analytics choice. It is the invoice.
A practice answer on outcome pricing needs to connect the event definition to what the buyer can verify on the bill.
The reference case
As of 6 November 2026, Fin's published pricing lists $0.99 per outcome (resolutions, Procedure handoffs, disqualifications and self-serve routings) and $9.99 for a qualification, with custom terms for voice and negotiated contracts. Fin, formerly Intercom, has been part of Salesforce since September 2026. Its outcome rules require an actual answer before a resolution can be confirmed or assumed. A later request for help in the same conversation reverses the resolution charge, even across billing periods; an unanswered clarification question is not billable. A configured Procedure handoff can be billed, while a standard escalation is treated differently.
These are vendor-defined billing events, reviewed on that date. A buyer still needs to map them to its own definition of a solved customer problem.
Every edge case is a pricing decision
Use the following hypothetical contract cases to test your own design. They are not claims about how Fin handles each situation.
A customer receives an unhelpful answer and closes the window. If your contract bills an assumed resolution, decide how sampling, complaints and credits distinguish abandonment from success. Chat inactivity alone cannot tell you.
A customer gets an answer in chat, then emails support the next morning about the same problem, and because the chat ended without a follow-up question it sits in the resolved column while a human agent picks up the email. The problem was never solved. Unless the definition includes a re-contact window across channels, the buyer pays for a resolution and then pays a human to resolve it again.
Two issues in one conversation, one solved by the agent and one handed to a person: one resolution, half of one, or none? Each answer is defensible. Each produces a different bill.
A customer is told, correctly, that the refund they want is not allowed. They leave unhappy and do not come back. Resolved, from the vendor's side. From the buyer's, possibly a lost customer.
Estimate how often each case occurs in the buyer's traffic before agreeing a rate. A rare event in one queue can dominate another.
The incentive problem, stated plainly
Outcome pricing is attractive to buyers because it seems to align the vendor with the result they care about. It does, partly. It also gives the vendor a reason to make conversations end in a way that counts as resolved, which is not quite the same as solving them. Making the escalation path harder to find, or ending conversations with a confident summary that discourages follow-up, would both raise billable resolutions without raising solved problems.
That is not an accusation against any vendor. It is a property of the pricing model, and a PM on either side of the contract has to design around it from the first draft of the definition rather than discovering it in the first quarterly dispute.
If you are the vendor's PM
Pick a definition you could defend to a skeptical operations lead at your largest customer. That usually means a re-contact window (a few days is common), excluding conversations where the customer abandoned before the agent's answer, and a clear rule for multi-issue conversations. Give buyers audit access to a sample of billed conversations. A definition that survives an audit is worth more in renewals than a few points of reported resolution rate.
Track your own leading indicator of disputes: billed resolutions followed by contact on another channel. If that number rises, the bill and the customer's experience are drifting apart, and the renewal conversation will be hard.
If you are the buyer's PM
Before comparing vendors on price per resolution, normalise their definitions. A vendor at $0.99 with a strict definition can be cheaper per problem actually solved than one at $0.70 with a loose one. Ask for the re-contact rate on billed resolutions, and run a pilot where you measure it yourself.
Then compare against the alternative honestly: the fully loaded cost of a human-handled conversation in your operation, and the cost of an in-house agent once you include the model bill, the evaluation work and the people who handle its failures.
In the interview
Use this original practice prompt: "We are moving our AI agent from seat pricing to per-resolution pricing. Define the metric." Our rubric rewards explicit rules for abandonment, cross-channel re-contact and multi-issue conversations, followed by an audit and credit process. Only then discuss the rate.
Specify what would trigger a revision: for example, disputes on billed outcomes exceeding an agreed threshold for two consecutive billing periods. Assign an owner and a process for notifying buyers before changing the definition.
Questions about this topic
Charging per result, such as a resolved support conversation, instead of per seat or per request. The customer pays only when the AI succeeds, so the definition of success becomes part of the contract.