Choosing AI training for your SME team: which format fits, and how to make it stick

In trainings I see the same thing every time: on the day itself the energy is high, everyone discovers something they think "I'm going to use this tomorrow", and two weeks later the team has fallen back into the old routine. The difference rarely sits in the slides or the trainer. It sits in what happens afterwards, or rather, in what doesn't happen.

So the question is no longer whether you train your team, because since the AI Act that is mandatory. The question is which format fits your business, and how you avoid spending money on a pleasant day with no lasting effect.

First the obligation: why you have to do something here anyway

Since 2 February 2025, article 4 of the European AI Act obliges every employer to ensure a sufficient level of AI literacy among everyone who works with AI on your behalf. That applies to your permanent team, but also to freelancers and partners who operate your AI tools.

The law works on the basis of risk. For most SMEs using low-risk tools like ChatGPT, Copilot or Claude, a solid foundation is enough: what is generative AI, what are the limits, what can and cannot be done with customer data, how do you recognise a hallucination. Only when you deploy AI for decisions with direct impact on people (selection, assessment, credit) do heavier requirements kick in.

For a grounding in what that appropriate level means, and how you arrange it pragmatically, I refer you to the piece linked above. This blog is about the step after that: which training format do you pick, and how do you make sure it is worth the investment.

The formats, with their pros and cons

There are roughly five axes on which trainings differ. They are not mutually exclusive, you usually combine them.

Open enrolment or in-company. An open training is cheaper per participant and you learn from people at other companies. The downside: the examples are generic and it rarely connects seamlessly to your own processes. In-company is more expensive, but the trainer can work with your documents, your inbox, your actual working day. For teams of four or five people and up, in-company often still works out favourably.

One-off workshop or ongoing programme. A one-day workshop is low-threshold and good for getting everyone to the same starting level. But a single day is precisely the format in which knowledge fades fastest. An ongoing programme (several shorter sessions across weeks or months) demands more commitment from the team, but works fundamentally better.

E-learning or classroom. E-learning is flexible and repeatable, ideal for pure foundational knowledge. It does require self-discipline, and anyone who doesn't block time for it won't do it. Classroom delivers interaction, direct feedback and group dynamics, but costs travel time and calendar space. Blended (foundation online, depth and practice in the room) combines the two and is usually the most practical format for SME teams.

General or role-specific. A general AI training gives the whole team a shared foundation and a common language. A role-specific training (marketing learns prompts for content, sales learns to prepare customer conversations, finance learns to query documents and figures) is immediately applicable and has the highest chance of sticking. The combination works best: first the foundation together, then the application apart.

What to look at when choosing

Five questions you answer up front, so you aren't disappointed afterwards.

What should the team be able to do after the training? Make it concrete. Is it about awareness (everyone understands the risks), about doing daily tasks more efficiently (concrete time savings per role), or about strategic choices (where do we deploy AI in the coming year). These three call for completely different trainings, and providers who promise everything for every goal are often the weakest pick.

How relevant is the practice? A good training starts with your work processes, not with the tools. Ask a provider up front: which examples do you use, will we practise with our own documents and cases, can I give input beforehand about what the team is struggling with. If you get vague answers, it will turn into a demo of various tools.

What is the level of the participants? Management and operational staff have different questions. Management wants to know where the value sits, what it costs and what the risks are. The team wants to know how they will work faster tomorrow. It often pays to make separate sessions for these, otherwise one half gets bored while the other checks out.

Who is standing at the front of the room? A trainer with demonstrable SME experience weighs more heavily than a trainer with only technical depth. AI tools change every quarter; what doesn't change is what an entrepreneur in a company of fifteen people can and cannot organise. Ask for concrete examples from businesses your size.

Is there attention for responsible use? GDPR, bias, hallucinations, confidentiality of customer data. This is not a separate module at the end, it should be woven through the whole training. You need it for the AI Act anyway, and it prevents enthusiasm without brakes from leading to a data breach or a bad decision.

The real test: does it stick?

This is where most trainings go wrong. Success gets measured by attendance and an evaluation form right after the session. Both say nothing. Anyone who has just learned something new is enthusiastic, certainly if the lunch was good.

The only meaningful measure is behavioural change: does the team use the tools structurally, also when it's busy, also two months later. Forming a new habit takes roughly two to three months of conscious repetition before it becomes automatic. Without follow-up, a team falls back into the old routine after two weeks, not because the training was bad, but because the old way feels faster under pressure.

That is not a failure of the staff, that is how habits work.

How you arrange follow-up and embedding

Five things that make the difference between a pleasant day and a lasting change.

• Measure usage after two to four weeks, not right on day one. Look at the usage statistics of your AI tools, or ask people concretely: when did you use it this week, what for.
• Observe what people do when they are in a hurry. That's where you see whether the new way has really landed, or whether it only works when there is time for it.
• Ask for concrete output, not opinions. A set of prompts for the five tasks that recur weekly. A report that is now drafted automatically. A procedure that has become shorter.
• Schedule short follow-up sessions, one hour per month for the first three months. What works, where does it get stuck, what should we approach differently. This is cheaper than a new training and more effective.
• Build the application into the existing workflow, with explanation at the moment someone needs it. Nobody opens a manual after the fact. Everyone does click a button that sits in the tool they already use.

This is not a separate phase after the training, this should be part of the proposal. If a provider doesn't bring this up themselves, you are buying a day of fun, not a lasting change.

To close

A training that only inspires is wasted money. The value doesn't sit in the room, but in what happens after and how the application gets embedded in daily work. That is where most programmes fail, and that is what you have to select on.

If you are still orienting on where AI delivers the most in your business, before you start training, also read how to get started with AI in the SME. A training has the most effect when you know which two or three processes you want to strengthen with it first.