AI Training: Why Most of Them Change Nothing
AI training builds knowledge, but rarely change. Why AI courses in mid-sized companies so often fizzle out, and what works instead.
Date: 20 May 2026 | Author: David Hefendehl
All 40 people sat through the AI training. Four weeks later, everyone is working exactly the way they did before.
That is not a one-off. Unfortunately, it is the norm.
And it is not because the training was bad. The slides were good, the trainer knew their subject, the feedback round afterwards was warm. Still nothing happened. Because training delivers knowledge, and knowledge is not what is missing in mid-sized companies.
The knowledge problem was solved a long time ago
Anyone in your company can find out what a large language model is in ten minutes. There is YouTube, there are free courses, and there are the tools themselves, which explain what they do while you use them.
Even so, according to the HKA study, 40% of mid-sized companies use AI in some form, while only 21% have an actual AI strategy.¹ The gap between those two numbers is not a knowledge gap. There are no 19% of companies out there that simply hadn't heard of ChatGPT.
The gap is an application gap. People know what AI can do. They just don't know what to do with their own process on Monday morning.
What happens in a typical AI training
The running order is nearly always the same. First the fundamentals: what is AI, what is generative AI, a short history, a few impressive demos. Then the tool section: this is how you use ChatGPT, this is how you use Copilot, here are ten ways of phrasing a request that get better results. Then an exercise built around an example that has nothing to do with your company.
At the end, everyone walks out thinking it was all rather interesting. Then they get back to their desk, where the same 200 emails, the same spreadsheet and the same back-and-forth with sales are waiting. And the jump from "interesting" to "I do my job differently now" never happens.
That is not a criticism of trainers. It is a problem with the format. Training is designed to move knowledge into people's heads. It is not designed to change how work gets done.
Why generic examples don't transfer
Most AI courses practise on examples that have to work for everyone in the room. Rewrite a marketing email. Summarise meeting minutes. Translate a text into English.
The problem is that a job isn't made of generic tasks. It is made of a quotation that needs three systems and two colleagues. Of a goods-in inspection where someone walks the floor with a caliper and a printed list. Of complaint handling where 80% of cases look alike and 20% are nothing like each other.
A generic example shows you that AI can rewrite text. It doesn't show you whether AI can pre-sort your complaints. And that is the question you actually care about.
If you want to see what one of those workflows looks like in practice, I put it into a free 30-minute video. You take a real workflow from your own week and work it through with a template. No fundamentals section, no tool tour.
Watch the free 30-minute video
The difference between knowing and being able
There is a reason nobody learns to swim from a book. Physical skill comes from doing, not from listening. AI at work is no different, except that almost nobody treats it that way.
What teams actually need to learn is not how to operate a tool. It is the judgement to answer:
- Which steps in my process are even candidates for AI?
- What exactly goes in at each point, and what comes out?
- Where does a human have to look at it before anything moves on?
- How would I notice that the output is wrong?
- When is AI the worse option compared to a simple rule or a macro?
That judgement can't be delivered from a stage. It shows up once a team has walked the whole path: their own process, their own data, their own result, and their own dead end somewhere in the middle.
What works instead
The formats that actually change something in mid-sized companies share three traits.
They work on the real process. Not a sample case, but the workflow that is costing the people in the room time right now. That sounds obvious, and it is the decisive difference. Take your own process apart and you pick up everything you need to know about AI on the way.
Something is running at the end. Not a slide deck, not an action plan, but a working prototype, however small. A live demo changes the conversation inside a company immediately, because it is no longer about what might be possible but about something people can see.
The team builds it themselves. If an agency builds the prototype, the team has seen something. If the team builds it, the team has understood something. That is the difference between a demo and a capability that stays in the building.
There is a useful side effect. Anyone who has built a small AI prototype themselves develops a very realistic sense of what works and what doesn't. That is exactly the sober view you need to make good decisions, and it comes from doing the work, not from a slide headed "limitations and risks".
When training is the right call anyway
There are cases where classic AI training is the right format. If the goal really is to bring a large group to a shared baseline, a talk is efficient. If a new tool is rolling out and everyone needs to know how to operate it, tool training makes sense.
You just need the right expectation attached to it. Training makes sure everyone knows the same things. It does not make a process change. Expect change, buy knowledge transfer, and you end up disappointed and convinced AI is overhyped.
The expensive mistake isn't the training itself. It's the conclusion that follows: "We gave AI a go, it didn't do much for us."
From knowing to a first running result
If it feels like your company has talked a lot about AI and changed very little, that is rarely down to the people. It is down to the format.
That is what the AI Hackathon is for: a 2-day AI Innovation Workshop where your team works on a real process from your own company and walks out with a running prototype. No coding skills required. No deck at the end, but something you can actually demo.
It is the difference between "we now know what AI can do" and "we built something that works here".
¹ HKA/KARL study 2025: AI in the German Mittelstand, Karlsruhe University of Applied Sciences, Prof. Dr. Steffen Kinkel (517 companies surveyed) - www.h-ka.de
