Told to use AI? Pick one task and run a proper trial.
Sixty minutes with your team. You leave with one supervised trial on real work — what the tool may see, who checks it, how you’ll judge it, and the rule for stopping.
14-day trial, no card. Runs in your browser.
60 minutes with the team
- Before
- 20 min, one person
- After
- 30 min, the reviewer
The problem
The instruction arrives from above: every team should be doing something with AI. Nobody says which task, what data is allowed, or how you would know it helped. So teams either do nothing, or try something nobody checks.
This session gets you to one small, safe trial that you can defend.
What you leave with
One supervised AI trial with a baseline, permitted input boundary, human reviewer, evaluation measure, and stop or retain criteria.
The hour
- 20 minutes
Before the session, one person
Bring one real task and a sample of how it is done now. Nothing personal, confidential or restricted unless the rules are already clear.
- 20 minutes
Step 1
Describe the work as it is today. Mark what information may go into a tool, what needs permission, and where a person stays accountable.
- 10 minutes
Step 2
Rule out anything without permitted data, an accountable reviewer or a way to judge it. Choose one candidate from what is left.
- 20 minutes
Step 3
Design the trial: approved tool, allowed data, how much work, who reviews, who approves, where to escalate.
- 10 minutes
Step 4
Agree how you’ll compare before and after, and write the stop rule now — before anyone has seen a result.
- 30 minutes
After the session, the reviewer
Compare trial work with the baseline. Record quality, errors, and the time spent checking. Keep, change or stop.
A worked example
- The task
- Replying to “where is my delivery” emails. About 40 a day, each needing a look-up in the order system.
- Ruled out
- Refund decisions (cost of an error too high) and anything using customer payment details (not permitted).
- The trial
- For two weeks the approved assistant drafts replies for delivery-status emails only. Priya reads and sends every one. Nothing is sent without her.
- How they’ll judge it
- Time per reply including Priya’s checking, how many drafts needed rewriting, and any complaint about a reply.
- The stop rule, written in the session
- Stop if more than one draft in five needs rewriting, or if checking takes longer than writing the reply herself.
- Review, two weeks later
- Change it. One draft in four needed rewriting, mostly for split deliveries. The trial continues for tracked single parcels only; Priya reviews again in two weeks.
Sometimes the right answer is not to use AI. If the work is unclear, the data isn’t permitted or nobody can review the result, the session says so and stops short of a trial. That is a useful result.
What it doesn't do
It doesn’t run any AI tool for you, approve one, or send anything. It doesn’t score your “AI readiness” or promise a saving. It helps your team make one careful decision and check it.
14-day trial, no card. One person pays; the team joins free.
Planning something bigger? The longer version is in the catalogue for consultants.