Where to Start with AI When You Run an SME

Nearly one SME in two has now launched at least one artificial intelligence project, up from 29% two years ago (Bpifrance, Baromètre IA 2026). The trend is real, but it comes with a figure worth facing head-on: the average budget for an SME's first AI project reaches €79,000 (Bpifrance, Baromètre IA 2026). That's not a sum you commit on a hunch.

Before choosing a tool, a provider or a training course, three questions help you avoid most of the costly mistakes.

1. Is my need similar to everyone else's, or specific to my business?

81% of SMEs that invest in AI ultimately choose a custom-built solution or specific development, rather than a standard tool (Bpifrance, Baromètre IA 2026). This figure says something important: most businesses don't see themselves in generic off-the-shelf tools.

If your need is common (posting regularly on your social media, for example), an existing tool may be enough. If your need touches on something specific to your business (a particular product catalog, a unique internal process), a standard tool will almost always disappoint you.

2. Do I have the skills to run it, or do I need to rely on someone else?

60% of SMEs that want to build AI skills in-house don't succeed (Bpifrance, Baromètre IA 2026). Hiring a data/AI profile is slow, expensive, and a single person can never cover building the tool, running it day to day, and ensuring its regulatory compliance all at once.

If you don't have a dedicated IT department, that's not a barrier to AI: it's simply a selection criterion. Look for a provider that builds and runs the tool for you, rather than a solution that assumes an in-house technical team.

3. Do I know what I'm allowed to do?

52% of SMEs don't precisely know what they're allowed to do with AI (Bpifrance, Baromètre IA 2026). Since August 2, 2026, a new set of obligations under the European AI Act has come into force, with penalties that can reach €35 million or 7% of worldwide turnover. Many companies are already using AI tools internally, without realizing it, that haven't been properly logged.

This question needs to be settled early, not as an afterthought. Unresolved regulatory uncertainty is one of the leading causes of projects that never move forward.

An AI tool is only worthwhile if it helps the company grow. Not if it impresses people in a meeting.

The order to follow

• If you're not sure about your regulatory situation, start with a compliance diagnostic: it's quick, quantifiable, and it clarifies everything else.
• If your need is close to that of other companies, first try an existing tool, at low cost and with no commitment.
• If your need is specific to your business, ask for an audit before committing to development: it quantifies what a tool would deliver, and may well conclude that nothing needs to be built.