SMEs: Two-Thirds Use Generative AI, But Applications Remain Isolated

Small and medium-sized enterprises are testing AI in close proximity to their everyday tasks, often without a formal plan. According to the Banque de France, two-thirds of organizations with at least 20 employees are already using generative AI, most often in a limited or experimental capacity. The figures from CPME regarding administrative burdens and discrepancies in technology adoption highlight a need: to connect tools, maintain context, and frame assistants with clear rules to truly free up time for teams.

Controlled Assistants Based on Shared Memory

For an AI assistant to be useful in SMEs, its actions must be understandable and justifiable. Its access remains limited to what is necessary, and any sensitive operation requires human validation. The memory it relies on must be consultable and exportable, as it belongs to the company and its employees, not to a vendor. The idea is to enable AI to better understand the professional environment and to intervene between different applications. If an absence occurs, the assistant can suggest a reorganization of the schedule and draft a message for the client, while leaving control in the hands of the concerned individual, within a structure where humans and agents share a collective memory. This memory allows for the reconstruction of context, improves coordination, and avoids having to restart conversations from the beginning.

Widespread but Often Experimental Use, According to the Banque de France

In early 2026, the Banque de France surveyed around 7,000 companies. It reports that two-thirds of organizations with at least 20 employees use generative AI, most often in a limited or experimental manner. In the 20 to 49 employee range, 64% utilize generative AI, but only 31% use predictive AI, which requires structured data and more advanced integration. In contrast, 60% of large companies report using predictive AI.

Administrative Burden Weighs Heavy, Initial Steps Remain Very Concrete

In 2024, CPME surveyed 1,612 business leaders from small and medium-sized enterprises: 28% reported spending at least two days a week on administrative tasks. In this context, the initial uses of AI are very concrete and highlight tools that can be tried without requiring complex projects. Adoption occurs closely aligned with daily needs: automating repetitive tasks, locating a document searched for over twenty minutes, and supporting those in the company who embody the processes—from the manager who follows up on invoices to the accountant who continues to reconcile entries manually.

Isolated Tools, Context to Preserve to Save Time

AI functions integrated into software often remain confined to their scope, while difficulties arise between tools: a sick leave can impact payroll, scheduling, and a delivery. Those who manage the files perceive these connections, but software does not always link them. When AI answers a question while losing the context of the client and the file, it necessitates re-explanation, which can negate the time saved in drafting. SMEs have already engaged in these uses without much fanfare, through concrete trials such as drafting emails or summarizing contracts. The next step is to start from operational problems and gradually connect what does not yet communicate. The advanced indicator is not the number of tools installed, but the time returned to teams and the usage the company wishes to make of it.