Training the team on AI: what works
Train staff to use AI: what they need, what does not work and how to measure whether training delivered results.
Reading time: 9 min
AI training in the company almost always fails in the same way: you organize a session on what artificial intelligence is, people come out interested, and three weeks later nothing has changed.
The reason is that you taught the subject instead of the task. What produces real use is half an hour on how to do better the thing that person does every day.
In summary
- Training on own task, not about AI in general.
- Little and repeated It beats a lot and only once.
- Start with those who have a concrete problem, not all of them.
- It is measured by looking at the use, not satisfaction.
- Resistance has three different causes and requires different responses.
- The policy goes explained, not distributed.
What Works
- Start with a real task. «How to prepare the response to a complaint», not «introduction to AI».
- Make them do it., immediately, on a real case brought to them.
- Also show failures: when the tool is wrong and how to notice it.
- Leave something written: three examples that work, not a pantry.
- Come back after three weeks and ask what they used and what they didn't.
Point 3 is the one that is missing in almost all formations and that generates the most confidence: showing where the wrong tool teaches you to use it judiciously and prevents the opposite problem, that is, to trust too much.
Who to form first
Starting from here
- Who has a repetitive task he hates
- Who is already curious and tries on his own
- Who has informal influence on colleagues
- Who will follow the project internally
Don't start here
- The whole company together
- Who doesn't have their own use case
- Who is openly opposed, in the beginning
- Only those responsible, without those who work
The last point in the right column is a frequent mistake: training those responsible hoping they will broadcast. It doesn't happen, because they won't use the tool and they won't know how to answer practical questions.
The Three Resistances
It is worth distinguishing them, because they require different answers and treating them all in the same way does not work.
Fear of error. "What if I send the customer a wrong thing?" It is a legitimate concern and the answer is procedural: to show where human control is, when it is necessary to reread, what should never be sent without verification.
Fear of being replaced. Rarely said out loud. The honest answer is not to reassure in vain but to be clear about what changes: which activities disappear, which ones remain, what is expected of them afterwards.
The instrument is uncomfortable. The most common and the least considered. If using the system requires more steps than the current method, people will not use it, and no training will change it. It's a design problem, not an adoption problem.
How it is measured
- How many people use the tool after a month, not after a day
- On what tasks, and if they are those planned
- How often: every day or once in a while
- How many questions come, and if they change in nature over time
The last point is a good qualitative signal: if after a month the questions are still "how do you...", the training has not taken root. If they have become ‘you can also do that...’, people have begun to explore.
Practical examples
General session in the company. Thirty people, two hours, topic "artificial intelligence for work". High liking, no measurable change after a month.
Forty-minute sessions per ward. Each on a specific task of that department, with cases brought by participants. Real use in three out of four departments after one month; in the fourth it emerged that the tool was more inconvenient than the existing method, useful information.
Policy training. Twenty minutes of spoken explanation of why certain documents must not leave the business, using examples from real work. Much more effective than distributing the document — see corporate policy on the use of AI.
The Material That Remains
A training produces lasting effect only if it leaves something consultable when the session is over and the person is in front of his real case.
What works is not a dispensation: they are three or four concrete examples of your work, with the instruction that worked and the result obtained. One page, not twenty.
- Examples taken from the real work of that department, not generic
- The exact instruction that worked, copyable
- A note on what to check before using the result
- The name of the person to ask if it doesn't work
This material grows over time if someone treats it: whenever a person finds an effective way to do something, that becomes an example for others. It is the cheapest way to circulate expertise, and does not require any platform.
The link to policy
Technical training and rule training must be kept together, because separates work badly both.
Explaining how to use the tool without saying what should not come out of the company produces effective use and accidents. Distributing policy without showing how working well produces rules perceived as obstacles.
Half an hour that covers both — that's how you do it, that's what you don't put in, that's why — is more effective than two separate sessions and costs less. The content of the policy is described in corporate policy on the use of AI.
What works and what doesn't
| Approach | Effect after one month |
|---|---|
| General session on what AI is | Interest, no use |
| Half an hour on that person's task | Real use |
| Intensive day once | I remember vaguely |
| Repeated short sessions | Competence Growing |
| Distributed Dispense | Unread |
| Three examples of your work | Consulted and reused |
Mistakes to Avoid
- Teaching the subject instead of the task. It produces interest, not use.
- Forming all together. It disperses and leaves no trace.
- One session only. Useful questions arise after use.
- Not showing failures. It generates misplaced trust.
- Measuring the liking. It does not predict adoption.
- Treating resistance as an attitude problem. Often it is design.
How to apply it in the company
Choose three people with a repetitive task and ask them to bring a real case. Forty minutes, you work on their case, you also show where the tool is wrong, you leave three written examples.
After three weeks come back and ask what they used. The answers will tell you whether to continue with that format or change, and especially which colleagues to involve next.
The link with the adoption of projects is in Why AI Projects Fail, the general method in AI project: how to get started.
The role of the driver
A factor that affects more than any teaching method: whether those who lead the company or the department use the tool or not.
You don't need to become an expert. He needs to be visibly one who tries, mistakes, asks. The implicit message questathis thing we all do, even I.is worth more than any formal communication.
The opposite is equally true: an instrument presented as "useful to you" by someone who does not touch it is read as something imposed from above, and adoption is affected regardless of the quality of the training.
It is a point that never appears in the training plans and that explains much of the difference between similar companies with different outcomes. The link with the adoption of projects is in Why AI Projects Fail.
To train those who will follow the system
There is one category of training that no one plans and that serves more than the others: that of the internal contact person who will have to follow the project after release.
It's not technical training. It's knowing where you look at the numbers, how you recognize that something is changing, what you can fix yourself and when you call the provider, where the documentation is.
- Where are the logs and what they tell
- How to read the exception queue
- What parameters can be changed without risk
- Who to call, for what, with what urgency
It is a few hours of passing deliveries that should be explicitly asked to the supplier and put into the contract. Without it, the referent remains with a system that does not know where to open, and the project enters the degradation described in Why AI Projects Fail.
The Right Pace
A practical question that always arises: how often to return to the subject.
In the first few weeks after the introduction of a tool people have many questions and little experience: a short close session works. After a couple of months the questions become more specific and less frequent: it pays a moment each quarter, when those who have found effective ways tell them to others.
The latter format —the people who teach each other —is the most effective and the least expensive, but it is only born if someone organizes it. Left to chance, knowledge remains in the individual heads and those who arrive after start over, exactly as described in AI for Internal Training.
Conclusion
AI training works when it's short, repeated, tied to a real task and done with who that task is. It doesn't work when it teaches technology to people who don't have a problem to solve.
And when people don't adopt the tool, before talking about resistance to change it's worth checking out the simplest thing: is it really more comfortable than the method they used before? If it isn't, they're right.
If you want to understand what processes in your company can be automated with AI, Giallo Studio can help you turn the problem into a real workflow.
I close with the simplest verification: after a month, ask three people to show you how they use the tool on a real case. In three minutes you understand if the training has taken much better hold than with any questionnaire, and above all see where they lock — which is the useful information for the next session.
Giallo Studio guides: methodology and sources
How we apply it in the company
Related resources
FAQ
What kind of training is really needed?
Training focused on people’s own work, rather than AI in general. A course explaining what a language model is generates interest without changing anything; half an hour showing someone how to use the tool for a task they perform every day leads to actual use.
Does everyone need to be trained?
No; trying to do so spreads your effort too thin. Start with people who have a concrete problem that AI can ease and those who are already curious: they will bring others along. General training without a use case of their own is forgotten within weeks.
How long should the training last?
Keep it short and repeated, rather than long and one-off. Brief sessions tied to a specific task, repeated at intervals of several weeks, work better than an intensive day that nobody remembers.
How do you measure whether the training helped?
Look at usage rather than satisfaction: how many people actually use the tool after a month, for which tasks and how often. The satisfaction questionnaire at the end of a session measures enthusiasm at that moment, which predicts nothing.
What should you do about people who resist?
Understand why. Resistance is rarely ideological: it is more often fear of making mistakes, of being replaced, or simply that the tool is less convenient than the current method. These three causes require different responses, and none is resolved by insistence.
Should training also cover the internal policy?
Yes, and this is the most neglected part. Rules on what must not leave the company need to be explained, rather than merely distributed: an understood rule is followed, while a rule that is merely read is circumvented with the best intentions.