Most AI pilots do not fail. They just never end. Three months in, people use it sometimes, nobody can say if it worked, and the renewal conversation is awkward for everyone.
We run every pilot the same way to avoid that. It fits on one page, and it is agreed before any work starts.
One process, written down
Not "AI for the HR team". One process: screening CVs for open roles, the Monday client report, replies to stock and price questions. Written down with what goes in, what comes out and who checks it.
Often we build the first version in about 48 hours on the team's own example, before the pilot is even signed. Seeing their own work come back done settles more questions than any demo.
A champion, not a committee
One person on the client's side uses it every week, tells us what is wrong and decides whether a result is good enough. Without that person, feedback arrives late and in committee language, and the pilot slows to the pace of the slowest meeting.
One measure, agreed first
Pick one number and agree it before starting. Some that work:
- Hours a week the process takes, estimated before and after.
- The report is out by nine on Monday, every week.
- Drafted replies sent with only light edits, out of all drafted.
- Time from a role opening to a shortlist the hiring manager accepts.
One measure is uncomfortable, because the pilot will do other useful things that do not count. That is the point. Three measures give everyone a way to call it a success.
An end date, with the review in the calendar
The pilot has a fixed last day, and the review meeting is booked on the day it starts. The review has three possible outcomes: it stays, it changes, or it stops. All three are fine. Drifting is not.
The price, before it starts
The scope says what the pilot turns into if it works: the monthly price and what it covers. Nobody should find out the cost at the review. It changes how seriously both sides treat the measure.
We also ask, at the start, whether we can tell the story if it works. Asking later is harder for everyone.
The whole thing, on one page
- The process, with what goes in, what comes out and who checks it
- The champion, by name
- The one success measure and today's estimate
- The end date and the review meeting
- The price it turns into
- Whether we may tell the story
If you are choosing which process to start with, the four questions are a good place to begin.