77% of French mid-sized companies already use generative AI, but structured adoption stays rare

77% of French mid-sized companies now use generative AI, up from 58% a year earlier, a 19-point jump in a single year. This figure, from Bpifrance Le Lab's 2026 mid-sized company barometer, traces a spectacular acceleration of AI usage among French mid-sized companies. Yet it deserves the same careful reading as other widely reported adoption figures, widespread usage does not mean structured adoption.

The nuance matters. This 77% figure measures generative AI usage broadly, an employee using ChatGPT, Claude, or another assistant on their own initiative to draft a document or summarize information counts in this statistic, just like a company that has rolled out a structured AI project across several departments with formal governance. Yet these two realities differ sharply in their actual organizational impact.

This figure contrasts sharply with a more restrictive one measured separately for French SMBs and mid-sized companies combined, where only about a third of companies have adopted AI in a structured way, according to a separate Bpifrance Le Lab study. The gap between these two measures illustrates a known but rarely quantified reality this precisely, mid-sized companies, larger and better resourced internally, move faster than SMBs on broad AI usage, without that lead necessarily translating, at the same pace, into comparable structuring.

Individual, informal generative AI usage implies no governance, no integration into existing business processes, and no measurement of impact on company performance. An employee who saves time drafting emails delivers real value, but a one-off, invisible one in a company's usual management indicators, unlike a structured project whose impact can be measured, communicated, and deliberately scaled.

This sits within a broader, and itself mixed, economic picture for French mid-sized companies in 2026. The opinion balance on revenue growth prospects rose 8 points to reach +18, a fairly positive signal. But 21% of mid-sized companies report difficulty financing their investment projects, up 2 points year over year, which mechanically limits the resources available to turn diffuse AI usage into a structured project with dedicated governance.

A mid-sized company, or an advanced SMB, that observes widespread but uncoordinated AI usage across its teams is not starting from a blank page, it is starting from largely cleared ground, with a specific risk to address, fragmentation. Dozens of spontaneous, uncoordinated uses, each individually legitimate, can collectively create inconsistent practices, unmanaged confidentiality risks, and difficulty identifying which uses deserve to be officially scaled rather than staying informal.

The shift from broad usage to structured adoption does not start by ignoring what already exists to launch a new project in parallel, a common mistake that creates confusion rather than clarity. It starts by precisely mapping these spontaneous uses, which almost always reveals use cases mature enough to be officially scaled, with the governance and impact measurement they previously lacked.

This is exactly the interplay between two steps of the IMPACT method, the Inventory step, which captures spontaneous uses already present in the company, and the Framing step that follows, which then gives them formal governance, turning diffuse, individual adoption into a genuinely steered company-wide AI strategy. 77% usage is not a destination, it is a starting point to structure. A five-business-day diagnostic maps your existing AI uses before structuring them.