Instrument
The Measurement Clause Review
Ten questions, in four parts, for any AI agent contract priced on an outcome. Use it before the clause makes the vendor both player and scorekeeper.
Who owns the instrument
- Who defines the metric? A metric with only one honest way to improve it has stopped being a measure.
- Who operates the measurement? If the vendor's platform does the work and reports the result, the scorekeeper and the player are the same party.
- Who can audit it — at your discretion, not theirs?
Is the baseline honest
- What is the baseline, and who set it?
- Does the baseline stay frozen for the whole contract term?
- Is attribution isolated from what your own team changes anyway?
What can be gamed
- Can the metric improve while the outcome degrades?
- What does this do to your own team's incentives?
What's the evidence grade, and what's the exit
- Has this survived production, anywhere, for nine months or more — without the vendor on the reference call?
- What's the exit? Who owns the rebuilt process, prompts, and data if you leave?
If the vendor proves the savings, the vendor owns the deal. Verification belongs to the buyer — or it isn’t verification.
Bring an actual proposal to The Attempt — we’ll run this review against it together.