AI Training at Work: Making Adoption Succeed in SMEs

AI training in the workplace is no longer a comfortable option. Since 2 February 2025, Article 4 of the European AI Act has required every employer that deploys AI tools to ensure a sufficient level of AI literacy among its staff [1][3]. Successful adoption in an SME takes clear governance, skills development by job function and GDPR safeguards.

Why has training your teams in AI become unavoidable?

Two forces push in the same direction: the law and the field. On the legal side, the European AI Act has been entering into application in stages, and since 2 February 2025 its Article 4 has imposed an "AI literacy" obligation on every provider and every deployer of AI systems [1][3].

The text requires them to "take measures to ensure, to their best extent, a sufficient level of AI literacy of their staff", taking into account their knowledge, experience and the context of use [2]. An SME that lets its teams use AI tools with no framework or awareness-raising is not meeting this obligation of means.

On the field side, adoption is growing but remains a minority: around 13.5% of EU enterprises with at least 10 employees used AI technologies in 2024, according to Eurostat [5]. In other words, the window to get ahead is still open, provided you treat AI as an organisational project and not as an individual gadget.

What does successful AI adoption look like in an SME?

Successful adoption is not measured by the number of licences bought, but by the number of tasks that have genuinely been transformed. It is first and foremost a change management project: you are changing working habits, not just a software stack.

Three markers set apart an SME that has succeeded in its adoption from one that has simply "tried AI":

• Defined, shared uses rather than a patchwork of individual initiatives invisible to management.
• A deliberate, gradual approach: one pilot function, a learning phase, then a wider rollout, instead of a big technology bang.
• Ownership by the teams themselves, who understand the tool's limits and know when not to rely on it.

The classic mistake is to deploy an AI assistant to the whole company at once, with no training and no rules. The outcome is predictable: a few enthusiasts take it up, the majority ignore it, and nobody knows which data goes where. To choose the first projects, it helps to identify concrete AI use cases in SMEs before equipping anyone.

How do you govern usage with an AI charter?

Governance is the foundation that makes training useful. Without written rules, each employee invents their own, and the best intentions produce data leaks. A short, readable AI charter answers three questions: which tools are allowed, for which uses, and with which data.

The core of the charter is the list of authorised tools, split by the confidentiality level of the data handled. This distinction is the most concrete line of defence against incidents.

Beyond the table, an effective charter names an AI lead, provides a channel for proposing new tools, and reminds everyone that AI output is always reviewed before it is shared. Governance is not meant to hold people back: it protects the company while giving teams a clear framework in which to dare.

Which risks should you avoid: data leaks, GDPR, dependency?

The first risk is the leak of confidential data to public AI tools. Pasting a client contract, a prospect database or an HR file into a consumer service amounts to entrusting that information to a third party whose hosting and reuse you do not control. Personal data is protected by the GDPR: any processing, hosting or transfer of data must meet the applicable obligations, notably the legal basis, minimisation, EU data residency and the supervision of subcontracting.

The Belgian Data Protection Authority also points out that it is essential for operators of AI systems to respect the principles of personal data protection, and provides dedicated resources on AI and the GDPR [4]. A credible training course treats this point as a priority, with examples of what must never be entered into a public AI.

The second risk is dependency. Building an entire critical process on a service whose price, availability and terms of use you do not control creates fragility. Good practice is to keep an exit route: document the uses, avoid proprietary lock-in where possible, and distinguish tasks where AI assists from those where it decides. On this last point, no performance commitment should be presented as a contractual guarantee that cannot be honoured: AI remains a fallible tool, not a promise of results.

How do you build skills function by function?

Gradual skills development is what separates lasting adoption from a flash in the pan. Rather than training everyone in everything, you proceed in waves, starting where the gain is clear and the risk is low.

1. Raise everyone's awareness of the common foundation. A short session on what generative AI is, its limits (errors, "hallucinations"), and the charter's rules. This is also the basis of the AI literacy required by the AI Act [1].
2. Choose a pilot function. Support, administration or marketing often offer repetitive, low-sensitivity tasks, ideal for a first measurable learning experience.
3. Train on real cases from the job. Assisted writing, document summaries, first-level replies: you work on the team's actual tasks, not abstract exercises.
4. Document and share the good practices found by the pilot, then extend to a second function.
5. Appoint internal champions, those comfortable colleagues who support their peers day to day and feed needs back up.

This stepped approach has a decisive advantage: each wave funds the next through the gains achieved, and the company learns to govern AI before rolling it out widely. For highly repetitive tasks, training naturally leads to agentic workflows and automation once the uses are under control.

How do you measure the gains from AI training?

Without measurement, adoption stays a hunch. The aim is not to build a complex dashboard but to track a few simple indicators, chosen before the pilot starts.

• Time saved on a specific target task (for example, writing a first draft or handling a support request).
• Actual usage rate: how many trained people really use the tool each week, not just at launch.
• Quality and compliance: number of reviews needed, GDPR incidents avoided, absence of sensitive data in public tools.
• Team satisfaction, often the best early signal of adoption that sticks.

These benchmarks must stay honest. A real, modest, well-documented time saving is worth more than a flattering, unverifiable percentage. This rigour is what lets you decide, with figures in hand, which function to train next.

From strategy to implementation

At ITOPS.be, we approach AI training and adoption as a two-phase project. A Blueprint first: mapping of uses, AI charter, choice of authorised tools and GDPR framing tailored to your SME. Then a Build: training of teams by job function, putting safeguards in place and supporting the first pilots, with the same contacts from start to finish.

This continuity between advice and delivery avoids the classic gap between a fine AI strategy on paper and adoption that never happens on the ground. The point is not to buy AI but to get it adopted, with confidence and within the rules.

Frequently asked questions

Is AI training mandatory for a Belgian SME?

Yes, in part. Since 2 February 2025, Article 4 of the European AI Act has required every deployer of AI tools to ensure a sufficient level of AI literacy among its staff, proportionate to their uses [1][3]. It is not a certification but an obligation of means that must be documented.

Can you use ChatGPT at work without GDPR risk?

Not without a framework. Entering personal or confidential data into a public AI can amount to an uncontrolled transfer or processing, contrary to the GDPR [4]. An AI charter must list the authorised tools and forbid sending sensitive data to services not covered by a contract.

Which job function should AI training start with?

The one where the gain is most measurable and the risk lowest, often support, administration or marketing. A restricted pilot lets you learn before rolling out more widely, rather than a mass, ungoverned deployment.

How many companies already use AI in Europe?

Around 13.5% of EU enterprises with at least 10 employees used AI technologies in 2024, according to Eurostat [5]. Adoption therefore remains a minority, which leaves a competitive advantage to SMEs that structure their skills development now.

Sources and References

1. European Commission: EU regulatory framework on artificial intelligence (AI Act)
2. EU AI Act: Article 4, AI literacy
3. European Commission: Questions and answers on AI literacy
4. Data Protection Authority (Belgium): Artificial intelligence and data protection
5. Eurostat: Use of artificial intelligence in enterprises