AI in SMEs: where to start? A 5-step method
Updated on 10 November 2026. This article gives the method; to choose the tool itself (ChatGPT, Copilot, Gemini, Claude, Mistral…), see our guide Which AI for which task?.
AI is no longer just for large companies
In 2023, talking about artificial intelligence in a ten-person SME raised a smile. Today, most of your staff have already tried ChatGPT, Copilot or Gemini — often on their personal account, sometimes with company documents. So the question is no longer "should we go for it?" but "how do we go about it without spreading ourselves thin or taking unnecessary risks?".
The good news: getting started requires neither a big budget nor a team of data scientists. What matters is the method.
Step 1 — List the tasks that waste time
Before choosing a tool, look at your week. For a few days, note down the repetitive, text-based or time-consuming tasks:
- writing similar emails (reminders, replies to price requests, confirmations);
- summarising long documents, contracts or minutes;
- formatting offers, product sheets, procedures;
- looking for a piece of information buried in dozens of files;
- translating exchanges with Dutch- or English-speaking customers.
These are precisely the areas where AI is most useful.
Conversely, a task that requires a decision committing the company (approving a quote, a diagnosis, a hire) should probably stay in human hands.
Step 2 — Choose a single pilot use case
The classic mistake is wanting to "put AI everywhere". Choose one use case, ideally:
- frequent (several times a week);
- easy to evaluate (you can see straight away whether the result is good);
- low-risk (no sensitive data, no automated decision).
A typical example: drafting first versions of replies to customer enquiries, then reviewed by a human.
To find out which tool suits this pilot, our guide Which AI for which task? offers a task-by-task table, from free to team plans, with current prices.
Step 3 — Set the ground rules (data and confidentiality)
Even before the first test, define what can — and cannot — be sent to an online AI tool. Named customer data, health data, detailed financial information or trade secrets must not end up in an unmanaged consumer tool. We go into this in our article on AI and GDPR.
Depending on the level of sensitivity, there are three options:
| Option | For what | Confidentiality |
|---|---|---|
| Free consumer tool | Tests, texts with no internal data | Low |
| Business plan (Copilot, ChatGPT Business…) | Managed day-to-day use | Good (contract, no training on your data) |
| Locally installed AI | Sensitive documents | Best control |
Careful: a paid personal subscription (ChatGPT Plus, Claude Pro, Mistral Pro…) is not a business plan. Your conversations may be used to train the models until you switch the option off. The details for each vendor are in our guide.
A local installation does not on its own guarantee that nothing leaves the building: updates, telemetry, connected tools or the infrastructure used can transmit data.
With a business plan, also check the conditions for processing, retention, access and sharing, as well as the service settings: the absence of training on your data is not, on its own, enough to guarantee good confidentiality.
Step 4 — Test for a month and measure
Give yourself four weeks. Measure simply: how long did the task take before, and how long after (including review)? Is the result of equal or better quality? What caused problems?
Working with an AI means supervising rather than doing. It is not necessarily easier, nor always faster: that is exactly what the measurement should show.
If the gain is real, you have your argument for going further. If not, you have learned it for the price of a monthly subscription and perhaps a little expert support — not a five-figure project.
Step 5 — Train, document, expand
An AI used badly gives mediocre results, and the team concludes too quickly that "it doesn't work". A short practical training session — how to phrase a request, how to check an answer, what must never be sent — radically changes the outcome. It is also an explicit expectation of the European AI Act, which asks companies to support their staff's AI literacy (see what changes for SMEs).
Only then do you move on to the next use case.
Article 4 of the AI Act (Regulation (EU) 2024/1689) on AI literacy has applied since 2 February 2025. It was reworded by the AI "Digital Omnibus", which entered into force on 27 July 2026.
And in practice, for you?
Every organisation is different: an accounting firm, a garage, a construction company or an online shop will not have the same priorities. We offer an initial meeting to identify your two or three most profitable use cases and choose a solution suited to how sensitive your data is.