A new survey from Bpifrance, a state-backed French investment bank, reveals that generative artificial intelligence has become widespread among mid-sized French companies — but tangible productivity benefits remain hard to come by.
The annual barometer from Bpifrance, which tracks so-called Entreprises de Taille Intermédiaire, or ETIs, found that 77% of 534 company leaders surveyed reported their businesses are now using generative AI. However, just 17% of those firms said they had actually experienced time savings as a result of the technology.
The data paints a picture of adoption racing ahead of results, with many businesses unable to convert new AI tools into practical, measurable efficiency improvements.
Among the report’s key findings, companies using generative AI on a more frequent basis were more likely to report positive outcomes. Some 23% of regular users said they saw productivity gains, compared with only 12% of those who used the technology occasionally.
Despite the current gap between adoption and results, business leaders remain hopeful about AI’s future impact. Roughly 78% of surveyed firms said they believe generative AI will have a positive effect on productivity over time — a figure that rose 11 percentage points compared to the previous year’s survey.
The survey also found that generative AI adoption varied by industry. Service companies, along with industrial and construction firms, were more likely to be using the technology than businesses in commerce, transport, and tourism.
On the broader economic front, weak demand continued to be the biggest obstacle to growth, with 55% of companies pointing to current or anticipated softness in demand as a drag on their operations.
The outlook for cash positions in 2026 slipped two points to a balance of -12, with industrial and construction companies expressing the most pessimism. Meanwhile, the revenue outlook balance climbed eight points to +18, though that figure still falls well short of the long-term average of +29 recorded between 2011 and 2025.
