Training your teams on AI: where do you start?

The answer does not start with a tool, or with a list of prompts. It starts with a simpler question: who in your office works with documents today? Those people are probably already using AI, or will be soon. What determines whether that adoption holds over time is less the quality of the tool than the quality of what they know how to do with it, and on which tasks.

Why a single training session is not enough

Only 36% of employees consider their AI training sufficient, according to the BCG AI at Work 2025 study. Among those who had completed more than five hours of hands-on training, 79% used AI regularly, compared to 67% for those who had completed less. The gap widens further when training covers real company cases and includes follow-up, rather than generic demonstrations.

Those figures do not call for more hours so much as a different shape. Training works when it repeats and takes root in real work, not as a one-off event. A thirty-minute webinar on "how to write a prompt" does not produce the same effect as a follow-up that returns to the actual tasks of the office several times over the year.

The sticking point is often not where people expect it to be. A McKinsey study published in January 2025 (Superagency in the workplace) asked business leaders to estimate what proportion of their employees used AI for at least 30% of their work. The median answer was 4%. The figure measured directly from employees was 13%. Teams are adopting AI faster than leadership thinks. What is missing is rarely the motivation of the staff; it is the structure, the support, and the training on the tasks that actually matter.

What training needs to cover, without turning it into a full programme

Two dimensions come up in every office that takes AI adoption seriously. They do not require two separate modules or a full day, but both need to be addressed, even briefly.

The first is judgment: knowing how to recognise when a tool's output is correct and when it is not, and on which tasks a manual review remains necessary. Knowing which tasks genuinely lend themselves to AI is a topic in its own right, one that depends on the structure of each task and not just on the choice of a tool. Training does not need to cover everything, but it should at least signal that this sorting process exists and that the tool does not do it on the employee's behalf.

The second is the question of data: understanding what type of document can go into what type of tool, and why that distinction has real consequences. What using a consumer-grade tool without a contract covering data processing means for client data or confidential documents is itself a topic that deserves its own attention. What training should produce here is a reflex: before copying a document into a tool, ask what it contains.

These two dimensions form the foundation of what is called AI literacy. Training that covers only part of them leaves blind spots that do not show up on the day, but a few weeks later, when the habits formed meet a situation the training did not anticipate.

3 signs that training has actually trained, not just equipped

These three points allow you to assess a training you have received or are considering. They are not a programme: the precise content depends on the actual tasks of the office and the tools already in place.

• Staff know, on their own office tasks, which outputs need careful review and why, not only how to phrase a request.
• The question of what can or cannot be entered into a public tool was raised explicitly, with examples drawn from the company's usual documents.
• There is a clear answer to this question: what happens when the employee who was most skilled with the AI tool leaves the company?

That last point is often overlooked. In offices where adoption holds over time, the skill is not concentrated in a single person. It is shared widely enough to survive a departure, and documented well enough that a new employee can pick up the same practices without starting from scratch.

In practice, the few practices that genuinely changed how people work are written down somewhere, whether that is a shared page, a short memo, or a few lines added to an existing process. Not a full manual, but enough that a colleague or a new arrival can understand which tasks AI is used for, how, and what must always be checked. Without that, the skill leaves with the person who attended the training.

The construction site question

A doubt comes up regularly in construction companies: do teams need to leave the site to be trained? The answer is no, because the question itself rests on a mistaken idea of what AI literacy actually involves in this context.

In a construction company's office, the employees directly affected by AI day to day are those who work with documents: quotes, site meeting reports, technical specifications, supplier correspondence, subcontractor requests. These are rarely the site crews. A handful of people, sometimes two or three in a mid-sized office, handle the bulk of the document work. Training them on their own cases, without moving anyone off a site, is both more effective and easier to organise than the word "training" usually implies.

What the EU AI Act requires, without interpreting it for your situation

Article 4 of the EU AI Act has been in force since 2 February 2025. It requires any deployer to ensure that the employees who use AI systems have a sufficient level of competence to use them appropriately. The law does not prescribe a specific format, number of hours, or formal programme: it targets an outcome, not a checkbox. This framework may still evolve: an amending regulation (Digital Omnibus) was approved by the European Parliament in June 2026, but it has not yet been published in the Official Journal of the EU and therefore does not have the force of law at this date. The current wording of Article 4 is the one that applies.

In practice, keeping a record of training carried out and its content remains useful, whatever form the final text takes. A checkbox on a form, with no record of content, provides little. What actually matters is what the training produces: the ability to use a tool appropriately, which includes knowing when and how to check its outputs.

What Article 4 means precisely for your specific situation is a matter for legal advice, not a blog post.

Frequently asked questions

Is a single training session enough?

Rarely. According to the BCG AI at Work 2025 study, only 36% of employees consider their AI training sufficient. Among those who had completed more than five hours of hands-on training, 79% used AI regularly, compared to 67% for those who had completed less. The gap widens further when training covers real company cases rather than generic examples. A single session, however long, does not produce the same lasting effect as a follow-up that returns to the actual tasks of the office over time.

Do teams need to leave the construction site to be trained?

Not necessarily. AI literacy in a construction company's office mainly concerns the few people who work daily with documents: quotes, meeting reports, technical specifications, supplier correspondence. These are not the site crews, but the office staff. Training those people on their own tasks, without pulling them away from their work, is the most direct way to start.

Who does the AI Act obligation apply to?

Article 4 of the EU AI Act applies to deployers, meaning companies and organisations that use AI systems in their operations. This includes construction SMEs that use AI tools in their daily processes, not only technology providers. The regulatory framework may still evolve; for your specific situation, consult a lawyer specialised in data protection or regulatory compliance.

How long does it take to train a team?

There is no universal answer. The issue is not the total number of hours but whether training covers the concrete tasks of the office rather than generic examples, and whether it is followed by time to let the habits settle. A few hours targeted at the company's real cases produce more lasting results than a full day of training on a catalogue of features.

---

Where do you start in practice? With the few people who handle documents daily, and with their actual tasks, not with a catalogue of features. A first step, before committing to anything: identify the two or three document tasks that come up most often in the office, a type of quote, a standard meeting report, a recurring letter, and note who handles them. That is where training is most likely to stick, because repetition is what turns a practice into a habit.

The right depth and the right order then depend on what you find in the company: which tools are already in use, who actually uses them, and where the skills gap really is. Identifying that is usually faster with an outside perspective than starting from a generic training programme.

This content is provided for general information only. It does not constitute legal advice and does not replace the advice of a lawyer specialised in data protection or regulatory compliance. For your specific situation, consult a legal advisor.