How to price and govern agent-run services
Written by Alex Vakhitov, founder of Comonad Limited. Published
On this page
How do fixed, outcome and hybrid prices differ?
A fixed fee is agreed before the work starts and does not move if the result is better or worse than you hoped. It fits work whose shape you can describe: an assessment, an operating design, a written control set.
An outcome price moves with a result you can check. The result has to be named before the work starts, in a measure you already report or can inspect. A general aim such as 'better' will not do; a gate pass rate, a cycle time or a defect count you already track will.
A hybrid combines the two: the fixed part pays for the work, and the outcome part pays a bonus or a share only if the agreed result is met. We use this structure for pilots and managed runs.
| Shape | What you pay for | Where it fits |
|---|---|---|
| Fixed fee | The work described in the scope | Assessment and operating design |
| Fixed fee plus outcome bonus | The pilot work, plus a bonus if the agreed outcome is met | A pilot beside the current team |
| Monthly base plus capped share | The run, plus a share of an agreed outcome up to a cap | A managed run, with an exit |
Where does the risk sit?
On a fixed fee, the supplier carries the risk of doing the described work for that fee. You carry the decision of whether to go on to a pilot. A fixed fee for an assessment should be payable even if the recommendation is to stop.
On a pilot, the fixed fee covers the work of running the pilot. The bonus is the part that depends on the outcome. If the outcome is missed, you do not pay the bonus, but you still pay for the pilot. Because the supplier is paid for the work either way, it has no reason to hide a poor result.
On a managed run, the monthly base covers operating the function. The share is the part that moves with the outcome. The cap stops that share from becoming an open-ended claim on your budget.
Why cap an outcome share?
A cap is a ceiling on the variable part of the fee, written as an amount or as a multiple of the base, agreed before the run starts. Above the cap, a better outcome does not increase the fee unless you sign a new agreement.
The cap stops a supplier being paid without limit for a result it only partly caused, and it means you know the maximum cost before the quarter closes. If the outcome measure turns out to be wrong, change the measure; raising the cap only hides the problem.
What should an exit clause cover?
An exit clause is the right to take the function back, on notice you agreed at the start, without buying the right to leave. It should name what transfers: the runbooks, the records, the evaluation sets, the prompts and policies you own, and the access needed to operate without the supplier.
It should also say what happens to work in flight, how long the supplier supports the handover, and that the agent can be switched off and the previous way of working restored. A managed run without that clause is a lock-in, whatever the monthly figure looks like.
What has to be auditable?
Three records, kept where your auditors already look. The scope: what was agreed, including the outcome measure, the cap and the exit. The run: what the agent did, what a person accepted, and what was rejected. The fee: how the fixed part and the variable part were calculated from those records.
If you cannot recalculate the outcome share from the records, you cannot check what you are paying for. The same records serve the operating model: decision rights, the rollback test, and the regulatory mapping. See AI operating model for the enterprise.
Who stays accountable?
A named person in your organisation. The supplier can operate the run. The supplier does not become the owner of the function because the invoice says 'outcome'.
An outcome fee also does not decide your role under the EU AI Act. Provider and deployer follow the activity, under Regulation (EU) 2024/1689 (opens in new tab) as amended by Regulation (EU) 2026/1744 (opens in new tab). That is for your counsel to decide; we map the controls but do not give legal opinions.
Why write the price and the controls before the pilot?
In a press release of 25 June 2025, Gartner said that over 40% of agentic AI projects will be cancelled by the end of 2027, because of escalating costs, unclear business value or inadequate risk controls. That is Gartner's prediction, not a result from your estate.
The more useful part is the three reasons Gartner gives. Cost, value and controls can all be written into a scope before a pilot begins: what the work costs, what outcome would count as success, and which controls have to pass.
How Comonad helps
This is how an AI operations engagement is priced: a fixed fee for assessment and design, a fixed fee plus an outcome bonus for the pilot, and a monthly base plus a capped share for a managed run. The usual first function is the AI-enabled SDLC. Accountability and the control set sit with AI governance.
Discuss AI operationsSources
Frequently asked questions
Should we pay only when the outcome is met?
Not for an assessment or a design. That work has a product even when the recommendation is to stop. For a pilot, pay a fixed fee for the work and a bonus for the outcome. For a managed run, pay a base for the run and a capped share for the outcome. A pure outcome fee with no base pushes the supplier to choose easy work and to argue about the measure.
What does the cap actually limit?
The variable part of the fee. The base is still paid. Above the cap, further improvement does not increase the fee unless you agree a new cap. The cap is not a limit on the agent's activity. Activity limits belong in the decision rights.
Can we take the function back?
Yes, if the contract says so. The clause should cover notice, the records and access that transfer, work in flight, and the right to switch the agent off. Agree it before the managed run starts, not when you want to leave.
Does Gartner's figure apply to us?
No. It is Gartner's prediction, published on 25 June 2025, that over 40% of agentic AI projects will be cancelled by the end of 2027. It is not a measurement of your programme. Use it as a reason to write cost, outcome and controls into the scope before the pilot.