Generative AI: over two thirds of French firms with 20+ employees use it, but gains remain limited
Banque de France, 3 September 2026: 64% of firms with 20 to 49 employees say they use generative AI, mostly in support functions, with economic effects still limited.
- Banque de France, 3 September 2026: over two thirds of firms with 20 or more employees use generative AI in early 2026.
- Use reaches 64% among firms with 20 to 49 employees and 83% at 1,000 or more, mostly in administration and sales.
- Observed effects remain limited: for an agency, aim at one precise task, with a person who approves.
In Banque de France Bulletin no. 266, published on 3 September 2026, Véronique Genre and Léo Parpais draw on a survey run between late February and early April 2026 among about 7,000 firms in the bank's monthly business survey. In early 2026, more than two thirds of firms with at least 20 employees say they use generative AI tools: 64% among the smallest (20 to 49 employees) and 83% among the largest (1,000 employees or more).
Use remains very uneven: most firms rely on general-purpose applications, in support or cross-cutting functions such as administration and sales. The economic effects seen so far are still limited and mainly concern productivity gains. Firms expect more within three years, with a moderately negative effect on employment, but no significant drop in their selling prices.
For AppH
- Support functions such as administration, mail and reminders are exactly where French firms already use AI. An agent that prepares these repetitive messages fits a use that is already widespread.
- The figures call for caution rather than excitement: every message the assistant prepares is read and approved by a person at the agency before it goes out.
Against / the honest limit
- The survey covers firms with at least 20 employees and measures declared use; it says nothing about very small businesses or domiciliation agencies.
- Using a tool is not the same as gaining from it: the Banque de France itself notes that observed effects remain limited, and its results about the future are firms' expectations, not measurements.
The Banque de France's finding shifts the question: the issue is no longer adopting AI but getting something lasting out of it. For a small agency, that means one well-chosen repetitive task, measured, and a person who stays in charge.
Reviewed by a human at AppH