Running an AI Skills Training Session That Actually Sticks

Most corporate AI training fails in the same predictable way. A trainer demos six impressive tools, everybody nods, nobody changes how they work on Monday, and six weeks later the organisation concludes that "AI did not work for us". The problem is rarely the tools. It is that the session was designed to impress rather than to transfer a habit.

Audit before you teach

I do not accept a training booking without a short discovery step first — usually a survey plus three or four conversations. I want to know what each group actually does all week: what they write, what they check, what they chase, where they wait for someone else. Out of that comes a list of real, boring, repeated tasks. Those tasks become the curriculum. A finance team learns AI through variance commentary and supplier emails, not through generating fantasy artwork.

Teach one workflow, not six tools

Tool tours are entertainment. A workflow is a sequence a person can repeat without me: input, prompt pattern, quality check, where the output goes. In a half-day I will teach two or three workflows at most, each end to end. Depth beats coverage, because confidence comes from having finished something, not from having seen something.

Build live, with their data

The turning point in every session is when participants stop watching and start typing with material from their own inbox. Prepared demos always work; that is precisely why they teach nothing. Real documents fail in interesting ways, and fixing those failures in the room is the actual lesson. I plan for at least half the time to be hands-on and I resist the urge to rescue people too quickly.

Teach the failure modes honestly

Everyone should leave able to name what the tools get wrong: confident invention of facts, outdated information, fabricated citations, mangled numbers, and a house style that reads like a press release. I teach a verification step as part of every workflow — never as an afterthought. Trust in a system depends on people knowing exactly where it is unreliable.

Address the fear in the room

In almost every session, someone is quietly worried the training is a prelude to redundancies. Ignoring that guarantees polite non-adoption. I name it early, and I frame the goal in terms of what the tools remove — the fifth draft of the same email, the manual reformatting, the blank page — and what remains human: judgement, relationships, accountability, taste. People adopt tools that make their day less irritating, not tools that feel like a threat.

Governance in plain language

Ten minutes on rules saves months of trouble: what may never be pasted into a public tool, what must be labelled as AI-assisted, who signs off on anything customer-facing, and where prompts and outputs are stored. Written simply, on one page, this is the difference between an organisation that scales AI use and one that quietly bans it after an incident.

Leave a kit, not a slide deck

Everyone leaves with a small prompt library tied to their actual tasks, written in their language and stored where they work — not a PDF of screenshots. Alongside it: a one-page checklist for verification, and a named internal champion per team who owns questions after I leave.

Measure the boring things

Feedback forms at the end of a good day are meaningless — energy is high, everyone scores it well. What matters is what happens two to four weeks later: how many people used a workflow last week, how much time they estimate it saved, what broke, and which workflow nobody touched. That follow-up conversation shapes the next session far more than any satisfaction score.

The shape of a session that works

• Twenty minutes framing: what the tools are, what they are not, why this team specifically.
• One workflow taught live, then rebuilt by participants on their own material.
• A deliberate failure: a task the tool gets wrong, diagnosed together.
• A second workflow, built almost entirely by them.
• Governance and the prompt kit.
• Each person commits out loud to one task they will run with AI this week.

That last commitment is small and slightly awkward, and it is the highest-leverage five minutes of the day. Training does not change organisations. Repeated use does, and repeated use starts with one task somebody actually tried on Monday morning.