Free template
An AI pilot brief that fits on one page
Six parts, written before the pilot starts, and five gates it passes before production. If you cannot fill all six, you are not ready to start, and now you know what to do first.
A pilot judged by how it felt will always pass
People's sense of what AI did for them is not a measurement of what it did. In one controlled study, experienced developers using AI took longer on real tasks, and believed they had been faster.
So a pilot is judged against a number the business already tracks, not by asking the users whether they liked it. The brief makes you write that number down before you start. See why most AI pilots never reach production.
Six parts, one page
- The use case. Its tier on the Opportunity Pyramid, its data position and its route: research, product, a feature in a system you already have, or data first.
- The number. What you measure today that the pilot should move, by how much, by when. "Median ticket resolution from four hours to under thirty minutes within eight weeks" is a number. "Better customer experience" is a wish.
- The scope. The narrowest slice on which the number can be measured: one product line, one site, one ticket category.
- The owner. A named person in the business, not in IT and not at the vendor, who will run the result.
- The escalation. What the system does when it does not know, and who it hands to. Designed before the first line is written.
- The stop condition. The result, or lack of one, that ends the pilot. Pilots without one become permanent.
Then one more line: the whole process. Work removed, work moved elsewhere, work created in checking and correcting, any change in quality or risk, and the money, counted at the end of the process.
From pilot to production, in order
The director's checklist. Do not move to production until all five are ticked.
1. The data is audited
Somebody has looked. The data position is measured, not just reported in a workshop.
2. The owner has agreed
Not nominated. Agreed, having read the brief and accepted the number.
3. The escalation is designed
The system has a defined way of saying "I don't know", and a person who receives it.
4. The test set exists
Twenty to fifty real cases with the right answers, run again every time anything changes.
5. The governance fields are filled
Risk category, data protection, human oversight and transparency.
Then, a review date
Every system in production gets a monthly cost line, a six- and twelve-month benefit check, and a condition for switching it off.
Using the brief
Which use case should the first pilot be?
Something that already has users and a number, from the "ready now" group on your scorecard. Not the thing that would look best on a slide.
What if the pilot fails its stop condition?
Then it worked. It answered a question for a bounded cost. The use case goes back on the roadmap, usually as research or data-first, and you move to the next ready-now item.
Who pays for the system once it is running?
The owner. AI has a metered running cost that grows with use, so forecast it before production, meter it from day one, and review it monthly against the number.
Run a pilot that becomes a system
CONVERGE chapter 10 covers the brief, the gates and what ownership means in full. Or we can scope and run the first pilot with you.