At a glance: AI in a small business in 2026
- Where do small businesses stand? In France, 40% of small and mid-sized businesses use at least one AI solution in 2026, and 53% of mid-sized ones, up from 34% a year earlier (France Num barometer, September 2026).
- What for? First, producing content, answering questions, sorting and analyzing documents, and automating repetitive tasks.
- Where to start? One frequent, measurable task, an off-the-shelf tool before any custom build, and a pilot of a few weeks with a before/after metric.
- What rules? Written usage rules (which data can go into which tool), data protection law, and the user-training obligation in the EU AI Act.
- What about ROI? It is measured, not assumed: time saved, errors avoided, shorter turnaround, compared with the real cost of tools and training.
Where French small businesses stand
The France Num barometer, published in September 2026 by the French government’s Directorate General for Enterprise from 9,655 company responses, shows adoption accelerating sharply.
40%
of small and mid-sized businesses use at least one AI solution (13% in 2024).
53%
of mid-sized businesses, up from 34% a year earlier.
68%
of users rate the impact as positive; 84% among those paying for their tools.
Uses remain simple. Across all small and mid-sized businesses: content generation (34%), chat assistants (24%), document analysis and classification (13%), task automation (11%), data analysis and forecasting (9%).
Spending is modest too: 19% of small and mid-sized businesses pay for an AI solution (32% of mid-sized ones). Money goes first to subscriptions (17%), well ahead of training (5%), custom development (3%) and consulting (2%). In other words, most companies start with off-the-shelf tools, and that is often the right call.
Where to start: 5 steps
1. Pick a task, not a technology
List repetitive, frequent, easy-to-measure tasks: customer inquiries, meeting notes, invoice entry, finding information in internal documents. Keep one or two where a mistake is harmless.
2. Start with an off-the-shelf tool
ChatGPT, Microsoft Copilot or Claude cover most first uses with no development. Our ChatGPT, Copilot and Claude comparison helps you choose, especially on data privacy.
3. Write the ground rules
One page is enough: which tools are allowed, which data must never be entered into consumer plans (sensitive customer data, HR information, trade secrets), and who approves content before it goes out.
4. Pilot for a few weeks and measure
With a small volunteer group, measure before and after: time spent, number of cases handled, errors, response time. A pilot with no metric cannot support a decision.
5. Train, then expand
Training is the difference between a tool you bought and a tool people use. See our AI training program for teams. Only expand once the first use is established.
The use cases that pay off fastest
Writing and replies
Emails, quotes, product sheets, answers to common questions: a first draft in seconds, reviewed by a person.
Document summaries
Contracts, tenders, meeting notes: summarize, compare, pull out the key points.
Data entry and filing
Invoices, purchase orders, incoming emails: read, classify and pre-fill the fields.
Searching internal documentation
An assistant that answers from your own documents: that is the idea behind RAG.
End-to-end automation
AI built into a workflow: an email arrives, it is read, classified and entered into the ERP. See our n8n use cases.
AI agents
Assistants that chain several actions. Powerful, but best kept for processes you already master: see our concrete AI agent examples.
Measuring return on investment
The math fits on one line: annual gain = time saved per task × number of tasks per year × loaded hourly cost, minus the cost of licenses, setup and training. Add errors avoided and shorter turnaround when you can measure them (a quote sent the same day, an invoice entered without delay).
Three precautions: measure real time spent before the pilot, not an estimate; count review time, which does not disappear; and compare with the simplest alternative (a document template, automation without AI).
For more ambitious projects, our article on the cost and ROI of AI agents breaks down the spending items.
The rules and the pitfalls
- Data. A consumer plan does not offer the same guarantees as a business plan. Check where data goes and whether it is used to train models: see Claude for business: data and security.
- Data protection. Processing personal data with AI is still processing: legal basis, informing people, a properly contracted processor.
- The EU AI Act. Since February 2, 2025, companies using AI systems in the EU must ensure their users have a sufficient level of AI literacy. The details are in our article on theAI Act for small businesses.
- Tools nobody uses. Licenses bought for everyone, used by a few: start small and expand after measuring.
- Blind trust. Generative AI can make up a figure or a reference: anything sent to a customer gets reviewed.
Frequently asked questions
How many small businesses use AI in France?
According to the France Num barometer published in September 2026, 40% of French small and mid-sized businesses use at least one AI solution, up from 13% in 2024. The share reaches 53% among mid-sized businesses.
What are the first AI use cases for a small business?
Writing content and replies, summarizing documents, entering and filing documents such as invoices, searching internal documentation, then automating complete processes.
Do you need to hire a data scientist to integrate AI into a small business?
Not to get started: first uses rely on off-the-shelf tools. Occasional support helps choose use cases, write usage rules and measure results; technical skills become useful for integrations and agents.
How much does it cost to integrate AI into a small business?
To start, mostly subscriptions billed per user per month, plus training time. Integrations into your business tools and agents are quoted case by case, depending on the process.
How do you measure the ROI of AI?
Measure the real time spent on the task before and after, multiply the time saved by annual volume and hourly cost, then subtract the cost of licenses, setup and training.
Bring AI into your business
Describe your business and your most time-consuming tasks: I reply within 24 hours with the two use cases to start with.
Sources
- France Num Barometer 2026, French Directorate General for Enterprise, published September 17, 2026 (9,655 companies surveyed).
- Regulation (EU) 2024/1689 on artificial intelligence, Article 4 (AI literacy).