Artificial intelligence adoption: 18% of French businesses, and who is missing

INSEE measures a tripling of artificial intelligence adoption by French businesses in two years. The overall figure hides a sharp size gap, a barrier other than price and an entire category the survey does not even examine.

Artificial intelligence adoption by French businesses reaches 18% in 2025, triple the 6% of 2023 (INSEE, Insee Première no. 2120, July 21, 2026). Behind the average lies the size gap: 15% among businesses with 10–49 employees, against 58% above 250 employees.

The figures by size and profession

One business in two above 250 employees uses AI, compared with one small business in seven; the middle band of 50–249 employees stands at 31%. The gap widened by 8 points in a year across the overall average. Sector figures show the same contours: 59% in information and communication, 33% in scientific and technical activities, 17% in manufacturing, 13% in trade and 10% in construction. A construction tradesperson therefore works in the country's least equipped sector, at one in ten. The European average of 20% places France slightly below. These gaps follow each profession's raw material: an information business handles text all day, a plumber handles jobsites, and consumer tools of the last two years do the first much better than the second.

What the survey does not measure

The 2025 ICT survey questioned around 11,000 businesses with 10 or more employees. Businesses below 10, the bakery, physiotherapy practice, three-person estate agency and association with two permanent staff, are outside its scope. The 18% says nothing about them in either direction. When a press article headlines French business AI adoption, it describes a country starting at ten employees.

The number-one barrier: not seeing a use

Among businesses not using AI, 71% say they see no use, 54% cite lack of expertise and 38% incompatibility with existing systems. Cost does not appear among the three cited barriers, moving the debate far from tool prices. Seven non-users in ten see no use, one in two lacks internal expertise, and among existing users, 53% say that same expertise shortage limits implementation. Our reading: the blockage concerns translation; nobody internally knows how to connect a business need to what a tool can do. We tackled this directly in four routes to an application without coding and what to say when you lack the words.

The decision's budget context

These choices happen in a tense climate. Bpifrance Le Lab's 83rd half-yearly barometer, published July 9, 2026 from 4,637 responses by microbusiness and SME leaders, reports 38% investment intentions for 2026 and 31% finding cash flow difficult. This release contains no AI figure: it supplies the context, not the answer. In that context, software is judged on entry cost and what can be stopped partway, more than a promised three-year return. A project stoppable after requirements without paying for construction fits tight cash flow; a six-month agency package does not.

Usefulness is demonstrated on a case, not a category

Technology's usefulness is not judged generally but against a precise need expressed in professional words: the quote copied three times, forgotten customer follow-up, spreadsheet nobody dares open. An agent team writing a brief then French requirements lets you decide on a document readable in a quarter of an hour before committing anything. Directing this work without coding remains a skill to acquire, closer to project leadership than computing.

Where to start as a small business

Take the task costing the most hours each month and write ten lines describing what happens: who does what, in which order, with which in-house rule. This page is enough to see whether a market tool covers it or your working method exceeds it. It starts a requirements document and belongs to you whatever follows.