The 30-day AI pilot

The one idea

When you are told to bring AI to your team, the instinct is to buy everyone a license and launch it on the same day. That fails for three reasons: nobody has a specific reason to change, you learn nothing because fifty people are poking at it fifty different ways, and you burn your one good shot at the team’s attention.

The fix is a pilot with four numbers, easy to remember:

• 5 people. Five or six, not three and not ten. Enough to keep momentum, small enough to meet for fifteen minutes and actually talk.
• 1 task. Real, recurring, and already disliked. Not a demo.
• 30 days. Enough repetitions to see the AI actually get better at the task as the group learns how to ask.
• 1 number. Agreed on before day one, so thirty days later nobody is arguing about whether it worked.

I call it the 5-1-30 pilot.

Step 1: pick the 5 or 6 people

Step 2: pick the one task

The task needs to be three things:

• Real. Actual work the team is responsible for, not a demo. “Summarize this article” is a demo. “Turn the weekly account notes into the update our director expects” is real.
• Recurring. Weekly or daily, so thirty days gives the group ten or twenty attempts, not two.
• Disliked. The status report, the first draft of the client email, the meeting notes, the monthly deck. If people already want it gone, they will push to make the tool work.

Pick one. If you have two good candidates, the second one is your next pilot, not a bigger first pilot.

Step 3: send the kickoff message

Tell the five people why they were picked and what you are asking, in plain terms. Copy this and fill in the brackets.

Hi team,

I'm running a 30-day test to see if AI can take the annoying part out of [THE TASK, e.g. "our weekly status report"].

You're one of five or six people I picked for this. Not because anything's wrong, just [WHY THEM, e.g. "you own this task every week" / "you already asked me about AI" / "you'll tell me straight if this doesn't work"].

Here's the deal:
- For the next 30 days, use [TOOL, e.g. Claude or ChatGPT] on [THE TASK].
- In week 1, we'll spend 15 minutes comparing notes on what's working. This is the most useful meeting of the whole thing, so come with what you tried.
- At the end of 30 days, we compare [THE ONE NUMBER, e.g. "how long the report takes"] to where we started.

If this works, you helped the whole team skip the slow way. If it doesn't, you saved everyone the trouble of a bigger rollout. Either way, this stays low-key. No dashboards, no steering committee.

Sound good?

Step 4: run the 30 days

Before you start, write down where the one number sits today. Nobody remembers this in thirty days if you do not.

Three checkpoints, fifteen minutes each:

• Week 1. Everyone shares how they are asking the AI to do the task. This is the most important meeting — people copy each other’s good prompts and the whole group jumps forward at once.
• Week 2. Look at what is still not working and adjust.
• Week 4. Compare the number to where you started. Get one concrete before-and-after example, not just a feeling.

Use the checklist below to keep the whole thing on one page.

THE 30-DAY AI PILOT CHECKLIST

Task (one sentence): [what the pilot group is testing AI on]
Pilot group (5 or 6 names): [ ]
Start date: [ ]        End date (+30 days): [ ]

BEFORE (fill in on day 1, before anyone touches the AI)
The one number: [e.g. "90 minutes to write the weekly report" / "3 rounds of edits on the client email" / "slips 2 of the last 4 weeks"]

CHECK-IN 1, end of week 1 (15 minutes): what's working
-

CHECK-IN 2, end of week 2 (15 minutes): what to adjust
-

CHECK-IN 3, end of week 4: the number, before vs. after
Before: [ ]     After: [ ]
One concrete example, side by side: [ ]

NEXT: run the balanced scorecard below before you make the call.

Step 5: run the balanced scorecard

The one number tells you if the task got faster. It does not tell you whether the team actually trusts the output, would keep using it on their own, or could teach it to someone else. Those matter just as much when you decide whether to put this in front of twenty more people. So before you make the call, run one more thing.

At the end of the 30 days, send your 5 or 6 pilot people an anonymous survey, no name field, Google Forms or Typeform or anything similar. Ask them to score the pilot 1 to 10 on each of the criteria below. Anonymous matters: someone who is not convinced will tell a form the truth faster than they will tell you in a meeting.

The weights add up to 100. For each criterion, average your respondents’ scores, divide by 10, then multiply by the weight. Add the seven results together for one number out of 100. Worked example: five people score “time or effort saved” at 8, 9, 7, 8, and 9. The average is 8.2. Divide by 10 to get 0.82, multiply by that criterion’s weight of 20, and it contributes 16.4 points out of 100. Do this for all seven and add them up.

80 to 100: a strong, consistent signal. Scale it. 60 to 79: promising but mixed. Run it again for 30 more days with the adjustment your team named in check-in 2, then re-score. Below 60: the crowd is not behind it. Stop, and use what you learned to pick a better task or a better group next time.

Look at the spread, not just the average. If four people score trust a 9 and one scores it a 2, that one answer is worth a private conversation. Averaging alone can hide a real problem.

ANONYMOUS PILOT SCORECARD (paste into Google Forms, Typeform, or similar; collect no names)

For each question, answer on a scale of 1 (not at all) to 10 (completely).

1. Time or effort saved: how much did this actually save you time or effort on the task, compared to before?
2. Output quality: how does the AI-assisted result compare to how you did this task before?
3. Voluntary use: if nobody required it, would you keep doing the task this way?
4. Trust: how much can you trust the output without double-checking or redoing it? (10 = rarely check, 1 = always redo it)
5. Teachability: how easily could you teach a new teammate to do this in 15 minutes?
6. Stress and energy: compared to before, does this task feel lighter or more draining? (10 = much lighter, 1 = much more draining)
7. Fit with how we work: how well did this slot into your existing workflow without extra hoops?

Optional: what's the one thing that would make this better?

Step 6: make the call, then spread it

Say the decision out loud to the group, backed by the scorecard number, not a feeling: scale it, adjust and run it again, or stop. A manager who runs a clean pilot and reports “this one did not pay off” builds more trust than one who declares victory on everything.

If it worked, your five people are now the ones who know how to do the task with AI. They are the teachers. Each of them sits with three or four teammates and shows the exact prompts, on the exact task. That is fifteen or twenty more people taught by a peer, on real work, no training vendor required. Then pick the next task and the next five.