How to buy AI training that changes something

AI training is easy to buy and hard to evaluate. The feedback forms are good, the slides are current, the room was engaged. A month later the same people are doing the same work the same way, and the budget holder has no way to tell whether that is the training’s fault or nobody’s.

The outcome is usually decided before the session starts, in what was bought. A few questions put to the provider at that stage predict the result better than any feedback form.

Whose material will you be working on?

The most important question, and the one generic courses fail. A session about AI in general produces people who know more about AI in general. A session built on your invoicing process, your service desk tickets or your quality reports produces people who know what to change on Monday.

Ask the provider what they need from you before the session. If the answer is a headcount and a room, the content is already written. If the answer is documents, process descriptions and a handful of awkward real cases, preparation is happening, and preparation is most of the work.

What exists at the end of the day?

Every session should end with an artefact: a ranked shortlist the room agrees on, a one-page usage policy, a working prototype, a redesigned process step. A session that ends with enthusiasm and no artefact has produced nothing anybody can use or check.

Ask the provider to name the artefact in the proposal. If they cannot, they are selling hours.

Who should be in the room, and how many?

Mixed levels for decisions, similar levels for skills. A process workshop needs the person who decides and the people who do the work, or the shortlist it produces will not survive the first meeting afterwards. A hands-on engineering workshop needs people at roughly the same level, or the fastest person builds and the rest watch.

Small enough that everybody builds. Hands-on workshops work best at six to ten people. Above that, people watch rather than build, and watching does not transfer.

Full attendance or none. A workshop where half the room keeps stepping out for calls is worse than no workshop, because it produces the appearance of alignment among people who were not there for the decision.

In a room, if people are building. Executive sessions work well remotely. Hands-on workshops are meaningfully better in person, mostly because the useful conversations happen while somebody is stuck, and remote formats hide being stuck. If the budget forces a choice, make the hands-on part the one that happens in a room.

Infographic

A generic course and a session on your material

Generic course

• –Examples from other companies
• –Lecture and tool demos
• –Ends with a satisfaction survey
• –A large room, mostly watching
• –Over at the last slide

Session on your material

• →Your processes and your awkward cases
• →Mostly doing, little lecture
• →Ends with a decision or an artefact
• →A small group, everybody building
• →A return visit some weeks later

What happens afterwards?

The useful learning happens weeks later, when people try it on real work and hit obstacles nobody anticipated. A follow-up session at that point — what was tried, what broke, what to change — is where training turns into capability. Ask whether it is in the scope. If it is an optional extra, it will be the first thing cut.

What it costs you, beyond the fee

The fee is the smaller cost. The larger one is the time of the people in the room, plus the preparation on your side: somebody has to choose the material, clear it for use, and be available to the provider before the session. And somebody with the authority to do it has to clear the participants’ calendars, or their phones will decide who is actually in the room.

That cost is worth paying only if the session produces something. Which is why the right way to make the case for training is not “our people need to understand AI” but “by the end of it we will have decided X” or “we will have a working Y on our own data”. If the case cannot be written that way, the training is not ready to be bought.

The warning signs

A certificate as the main deliverable. Demonstrations of other companies’ successes. A fixed agenda before anybody has seen your processes. And a request to use the session to sell an already-made decision internally — useful to somebody, perhaps, but it is not training.

That is how we build AI workshops and training: on your own documents and processes, ending with a decision or something working, and with a clinic some weeks later, once the real obstacles have appeared.