What a team actually ships during an AI training day

The usual AI training day ends with a room of people who enjoyed themselves and a slide deck nobody opens again. Two weeks later usage is back to where it started. The difference between a day that sticks and a day that evaporates is not the trainer's energy. It is whether anything was built.

The artefact test

We judge a session by what exists on the company's own shared drive when it ends. Not notes. Not screenshots. Things people can open on Monday and use on real work.

• ✳A prompt library for each role, written against that team's actual tasks
• ✳At least one finished piece of real work per person, produced during the session
• ✳A written rule for what company data may and may not go into which tool
• ✳One named internal owner per department who answers questions after we leave

Why generic examples fail

A prompt that summarises a made-up press release teaches nothing transferable. The same prompt aimed at the participant's own quarterly report teaches the shape of the problem, and the output is worth keeping. This is why we ask teams to bring real emails, real tickets, and real spreadsheets, and why we sign an NDA when that matters.

The week after is the hard part

Adoption dies quietly in week two, when the first awkward question has no one to answer it. Naming an internal owner before the trainer leaves costs nothing and is the single highest-return thing a manager can do on the day.