Build workflows, don’t rehearse examples

Why most AI training evaporates within a month.

The training goes well. People are engaged, someone laughs at the right moment, three participants say afterwards that they had no idea it could do that. Feedback forms come back strong.

Four weeks later, nothing has changed. Not because the training was bad, but because it was a training.

The demo is always clean

Sessions run on prepared material. A tidy document, a sensible question, an answer that lands. The example is chosen because it works.

Monday morning is not that. Monday morning is a half-finished press review, a document in the wrong format, three unanswered questions from a country office, and eleven minutes before a call. The distance between the demo and that is where enthusiasm goes to die.

What blocks adoption is rarely skill

People understand the tools faster than anyone expects. What they lack is somewhere to put the understanding.

There is no agreed place for the output. No approved tool for the material that is actually sensitive. No decision about which part of the process this now replaces, so it gets added on top of the old process instead of into it. Everyone is doing their job plus an experiment, and the experiment loses.

The difference between rehearsing and building

Rehearsing means learning how to phrase a request well, on an example. Building means taking the thing a team does every week, the media summary, the donor report, the interview preparation, and reconstructing it so that the new step is in the workflow rather than beside it.

The output of a training is a set of notes. The output of building is a process that someone runs on Thursday because it is now how Thursday works.

Inside the environment, not next to it

A workflow that lives in a tool nobody has approved is not a workflow. It is a liability with a countdown on it, and the countdown ends at the first data protection question.

Working inside the environment an organisation has already cleared is slower at the start and it is the only version that survives. Any capability that cannot be described to the person responsible for compliance will be abandoned quietly, usually without anyone reporting that it happened.

The part that should not be handed over

Not everything belongs in the workflow. Deciding what a development means, whether a sentence can be said publicly, which of three framings is the honest one: that stays with a person, and the people building the workflow should be able to say why.

Capability includes knowing where the line is. A team that has only been shown what the tools can do has learned half the job.

The test

Three weeks after the last session, is anyone still using it without being asked?

That is the only measure that matters, and it is not a measure of how good the training was. It is a measure of whether anything was actually built.