The AI diagnostic for SMEs: a step-by-step method before you invest

By Jérôme Denis, AI & data governance consultant · 6 min read · Updated 11 June 2026

The AI diagnostic for SMEs has become the mandatory step before any investment: according to Bpifrance Le Lab, 55% of French SMEs were using generative AI by the end of 2025, up from 31% a year earlier — and European markets follow the same curve. Behind that spectacular figure hides a more mundane reality: individual subscriptions, unframed usage, and very few structured gains at company level.

Why diagnose before investing?

Because failed AI projects in SMEs almost always share the same cause: a tool was bought before the problem was identified. The diagnostic reverses the approach — it starts from the processes that cost time or money, and only then asks whether AI (and which AI) can improve them.

The France Num 2025 barometer confirms it: SME AI usage is still dominated by individual content generation, far from the structural processes (invoicing, production, customer relations) where the most measurable gains actually sit.

What are the 5 steps of an AI diagnostic?

The complete method fits in five steps, applicable to any SME:

1. Map processes and their pain points

   Interviews with each function (admin, sales, production, HR): which tasks are repetitive, time-consuming, error-prone? Quantify: hours per week, error rate, delays.

2. Assess the available data

   AI feeds on data. Is it digitised, accessible, of sufficient quality? A process without exploitable data cannot be automated, whatever the budget.

3. Prioritize by ROI and feasibility

   Each use case is positioned on two axes: estimated gain and implementation difficulty. Launch the quick wins first — they fund what follows.

4. Check GDPR and EU AI Act compliance

   Risk classification of each use case, personal data involved, internal usage policy. Building compliance in from the start costs ten times less than retrofitting it.

5. Quantify before committing

   Implementation cost, monthly running costs (API, licences, maintenance) and expected gain, in black and white. That document makes the decision — not the vendor's demo.

Which processes should you automate first?

The best first candidates share three traits: high volume, clear rules, low decision stakes. In practice, across the SMEs I advise, the quick wins are almost always on the same list:

• Sorting and answering recurring inbound emails (quote requests, tier-1 support);
• Extracting and entering document data (invoices, delivery notes, applications);
• Assisted drafting: minutes, commercial offers, product sheets;
• Answering recurring internal questions from procedures (a RAG assistant);
• Lead qualification and follow-up.

Conversely, highly variable or high-stakes processes (HR decisions, contractual commitments, safety) come later, with a human in the loop. The technical building blocks — agents, RAG, MCP, low-code — are explained in AI agents, RAG, MCP: understanding AI solution architecture.

What do you get at the end?

A well-run diagnostic produces a decision document, not a slide deck of generalities: quantified and prioritized use cases, a realistic implementation budget, the monthly running costs, the compliance requirements per use case — and, for French entities, the public funding schemes that may apply (Bpifrance's Diag Data IA, IA Booster France 2030, France Num support, OPCO funding for training). International clients get the same document, with the funding section replaced by a financing-options note for their market.

Conclusion

An AI diagnostic is not a formality: it is the step that decides whether your investment produces a measurable gain or just one more subscription. Five steps, two to five days, and a costed decision document — that is what separates SMEs that succeed with AI from those that pile it up.