Training that survives Friday

The situation

It is Thursday at two. Forty people are in a room or in a grid of tiles, ninety minutes booked. The facilitator is good. The demo builds a campaign brief in front of everyone, and it lands — except that three people in the room write campaign briefs and thirty-seven do not. The feedback form comes back glowing. On Friday, four people open the tool. By the following Wednesday it is one. Six weeks later someone in finance asks why the licences are idle and you have no answer that survives a follow-up question.

Nothing failed in the room. The session was clear, well-paced, well-received. It just did not leave anything behind. People left with notes about somebody else's work.

The principle

A session counts as training only if every participant leaves with a working asset built on a task they will have to do again — otherwise it was a demonstration they watched.

Adoption does not fail on motivation. It fails on the gap between the room and Monday. In the room the example was pre-tested, the data was clean, and someone was standing there to fix it when it stalled. On Monday the person has a messy input, no example, and twelve minutes before a meeting — so they do the task the old way, and the old way wins by default. The only thing that closes that gap is having already done the real task once, on the real input, with help available.

Three design rules follow. Most of the minutes are hands on keyboards. Every exercise runs on the participant's own live work, collected in advance. Nobody leaves without an asset saved where they will find it again.

Worked example

Shown here in ChatGPT.

Two weeks out, you ask every participant one question: name a task you do at least weekly that takes you more than thirty minutes. You get forty answers of wildly uneven quality. Some are buildable in a forty-five minute block. Some are "improve team communication". You cannot sort forty by hand the night before, and if you skip the sorting you spend the session unblocking people who brought impossible tasks.

Assign the sorting:

Job: Sort 40 submitted work tasks into what a first-time user can build
in a 45-minute hands-on block, and what they cannot.

Inputs: The pasted list, one task per line with the submitter's role.
Assume a general chat assistant with file upload, no connections to
company systems, and no code.

Output: Three tables. "Build in the session" — task, the output shape
the person gets back, and the input they must bring on the day.
"Too big" — task, plus the smaller piece of it that is buildable.
"Not an AI task" — task, and one line on why.

Standard: A task belongs in the first table only if every input fits on
the person's own laptop. Anything needing a system we have not connected
goes to "Too big". Do not invent tasks that were not submitted.

The third column of the first table is now your pre-session email — one line per person, naming the file they must have open. The "Too big" table is your second session, already scoped. The "Not an AI task" table is the one you read out loud at the start, because saying what this will not do buys you more credibility than any demo.

Then the shape of the ninety minutes:

0:00  What changed, and what this is not              10 min   talk
0:10  Assign one task together, out loud              15 min   hands on
0:25  Build your own                                  35 min   hands on
1:00  Two people show what broke, not what worked     10 min   talk
1:10  Save it where you will find it Monday           10 min   hands on
1:20  Name the next task you will assign              10 min   hands on

Seventy of ninety minutes are hands on keyboards. The block people cut first is the one at 1:10, and it is the block that decides whether any of this exists next week.

In your tool

What you need is a place to put the session's starting instruction so twenty people begin from the same line instead of typing it.

Product surfaces checked 2026-08-04.

Try it

Open the agenda of the last session you ran or sat through. Mark every block WATCH or DO. Total the DO minutes and divide by the session length.

You are done when two things are true: six of every ten minutes are DO, and you can name in one sentence the asset each participant walks out with. If that sentence contains the words "notes", "awareness" or "understanding", you have not got an asset — you have a memory, and memories do not survive Friday.

Common failures

• The facilitator's example. The demo runs on the trainer's own work or a vendor's sample library. It is flawless because it was rehearsed, and it teaches the room that the tool is good at somebody else's job. The person watching never sees it handle a messy input, which is the only condition they will ever use it in.
• The sandbox task. Fictional company, invented data, tidy numbers. Nobody in the room can tell whether the output is right or wrong, so nobody practises the one skill that matters — checking. The first time they verify anything is the first time it counts.
• The session with no exit asset. Everyone builds something good, the timer runs out, the tab closes. A week later they remember it worked but not what they typed, and rebuilding from memory costs more than doing the task by hand. Saving is not admin at the end; it is the deliverable.