Start with a problem, not a technology
The most common mistake is approaching AI from the technology side rather than the need: “we should be doing AI” rather than “this particular task costs us a huge amount of time every week, and could be automated”. The best first projects almost always share the same traits: a repetitive task, with low human added value, using data that’s already available somewhere in the company.
The data question comes before the tool question
Before choosing a tool, there’s a simpler question to answer: where does the relevant data live, and who should be able to access it? This is often the most neglected step — and the most decisive one for the project’s success. A poorly scoped AI agent, in terms of access, can expose information to people who shouldn’t see it, even with no malicious intent involved.
Data sovereignty isn’t a detail
For a company based in Europe, the question of where data is processed — and by whom — is not a side issue. The GDPR imposes specific safeguards whenever personal data passes through services located outside the European Union. Some projects can perfectly well rely on European cloud services; others, involving more sensitive data, are better designed from the outset with local hosting options, or even inference that never goes through a third-party service. That choice is made case by case, based on the actual sensitivity of the data involved — not as a matter of principle. It’s the approach we follow on our artificial intelligence projects.
Start small, measure, then scale up
An AI project that starts with a company-wide rollout is a project taking unnecessary risks. The method that works best remains progressive: one specific use case, a prototype limited to one team or process, an honest measurement of the results, then — only if the test succeeds — a wider rollout.
AI doesn’t replace your tools, it connects to them
The most realistic scenario for an SME is almost never “replace all our tools with AI”, but rather connecting an AI agent to the tools already in place — email, file sharing, CRM, ERP — to automate one specific task without disrupting an entire team’s working habits.
Have a repetitive process in mind but don’t know where to start? Let’s book 30 minutes to talk it through concretely — no obligation.
