The hour that actually changes what people do

The session that does not work

You book an hour. You show the interface, demonstrate six impressive things, share a list of prompts, and answer questions. Everybody leaves interested. Two weeks later, three people are using it, one of them is pasting client data into a personal account, and nothing else has changed.

This is the standard outcome and it has a cause. A feature tour teaches what the tool can do. It teaches nothing about when to trust it, which is the only knowledge that changes behaviour.

Start with a failure, on their material

The first ten minutes should be a live, unrehearsed failure, on a task from the room's own work.

Ask somebody for a real question from their job — one where the answer depends on a specific they did not supply. Type it in. Read the confident answer aloud. Then check the specific together and find it wrong.

Nothing else produces the same effect. People arrive with a picture of either a magic box or a toy, and both pictures are wrong in ways that make them unsafe. Twelve minutes of watching it fabricate a case reference, a threshold or a supplier's terms replaces both pictures with the accurate one, and that calibration is the entire point of the session.

Be honest that it may not fail on the first attempt. Have two questions ready, and if both come back correct, say so — that is also data, and it is more credible than a rigged demonstration.

Then a real win, also on their material

Immediately afterwards, do something genuinely useful with a real task from the same person's week. Preferably one from the safe territory: rewriting something they wrote, structuring their dictated notes, drafting from a document they attached.

The sequence matters. Failure first, win second. Done the other way round, the win is remembered and the failure is filed as a caveat.

The four things worth an hour

1. Where did the facts come from? The question that sorts every task. If you did not supply them, they are unverified.
2. What does checking cost? If it costs more than doing, this is not a candidate.
3. What may not go in the box, in one page, specific to your organisation, with the answer to "what do I use instead".
4. Where a prompt that worked goes, so the next person does not start from nothing.

That is the whole curriculum. Everything else — the techniques, the tools, the clever prompting — can be learned later by people who want to. These four make the difference between a workforce that is safe and one that is not.

Work with their tool, their data, their desk

Generic training transfers badly. Somebody who watched a demonstration in a made-up scenario will not recognise the same situation in their own inbox on Tuesday.

If you can, sit with people individually for twenty minutes on their own work. It is slower per person and it is the only version that reliably changes what somebody does the following week.

The two people in every room

The enthusiast has been using it for months, has strong opinions, and is quite possibly the person pasting client material into a personal account. Do not embarrass them. Give them a job: ownership of the shared prompt library, and responsibility for bringing one failure a fortnight. Enthusiasm redirected into stewardship is the cheapest governance you will ever get.

The refuser is usually right about their own task. Somebody whose work is unrecoverable judgement about people, or who has spent twenty years developing exactly the skill being offered as a shortcut, has a sound objection. Say so, publicly. Agreeing with a correct objection buys you the credibility to make a recommendation elsewhere, and pretending the objection is fear costs you that person permanently.

A fortnight clinic beats an annual course

Twenty minutes, every two weeks, standing agenda: what worked, what failed, one prompt worth stealing. Attendance optional.

This is where the actual learning happens, because the questions are real and the failures are fresh. It also creates a place where somebody can say "I think I sent something I shouldn't have" early, which is worth more than every policy document your organisation will ever write.

Measure adoption honestly

Not licences issued. Not logins. Not enthusiasm in a survey.

Count tasks changed: how many named tasks are now being done differently, and what the before-and-after measurement showed, including checking time. The measuring lesson gives you the method. Three tasks genuinely improved is a real result. Forty licences and no measured change is a cost.

Never mandate

A requirement to use it produces compliance theatre: people run the tool, ignore the output, and do the work as before. It also hides the places where it is not working, because nobody wants to report that the mandated thing failed.

Make it available, teach it properly, remove the obstacles, and let the people whose tasks suit it demonstrate the benefit to the people whose tasks do not. That is slower and it is the only version that leaves you with an accurate picture of where this actually helps.