Outcome pricing

How to write an outcome definition for AI work.

A template for outcome definitions that make AI services billable and auditable: what counts, the evidence, how long it holds and what does not.

An outcome definition says exactly what counts as one completed piece of work, what evidence proves it, and how long it has to hold. Write one for every unit you pay for. If a result is not covered by a definition, it should not be on the invoice.

Definitions are where outcome pricing succeeds or fails. Some AI vendors count a support conversation as resolved when the customer confirms it, or when the customer leaves without asking for a human. The second condition can include customers who gave up. Others have moved to billing only verified resolutions. The difference is written in the definition.

Step 1: Name the outcome in business terms

Use the words your team uses: "certificate of insurance issued", "invoice matched and posted", "renewal file complete". Avoid system words like "task processed" or "workflow run". If a manager would not recognize it as finished work, it is not an outcome.

Step 2: State the conditions that make it complete

List what must be true. For "invoice matched and posted":

  • Matched to a purchase order and goods receipt within agreed tolerances
  • Posted to the ledger in the correct period and cost center
  • Supplier and bank details verified against the vendor master

Each condition should be something a person could check.

Step 3: List the evidence that proves it

Say what the record must contain: the source documents, the checks run and their results, the system reference where the result landed, and any human approval with a name and time. If the evidence is not in the record, the outcome is not billable.

Step 4: Set how long it must hold

Some outcomes can be reversed. An invoice can be reversed, a certificate can be reissued because it was wrong. Add a holding period: "not reversed within 30 days". Billing waits until the period has passed, or is credited back if the outcome fails.

Step 5: Write down what does not count

Be explicit: duplicates, items the client had to redo, items that failed a check and were sent back, retries by the agent. Listing exclusions up front prevents most disputes later.

Step 6: Agree who can dispute it and how

Name who on each side can challenge an outcome, the time limit for doing so, and how it is settled. With a full record behind every item, most disputes take minutes.

Template

  • Outcome: name in business terms
  • Complete when: conditions, as a list
  • Evidence: what the record must contain
  • Holds for: holding period, if any
  • Does not count: exclusions
  • Disputes: who, how, time limit

At Agentic MSP, every billable outcome has a definition like this, agreed before go-live. See how outcome pricing works and what the Evidence File records.

Common questions.

What is an outcome definition?

A written agreement on exactly what counts as one completed unit of work, what evidence proves it, and how long the result must hold before it can be billed.

Why do outcome definitions matter for AI pricing?

Outcome pricing only works if the outcome is real. Without a precise definition, a vendor can count things that are not really outcomes, such as a support chat where the customer simply gave up.

How many outcome definitions does a service need?

One per billable unit. A servicing pack with five task types needs five definitions, each short enough to fit on one page.

Sources

Want this run for you.

Agentic MSP runs back-office work with governed AI agents and bills only for verified outcomes.