Generative AI in small and midsize businesses: 3 mistakes to avoid before you start

The AI rush and its illusions

According to the Baromètre France Num 2025 (a French government survey of small business digital adoption, in French), 26% of French small and midsize businesses now use artificial intelligence, a rate that doubled in one year. But how many get real value from it? From what I see in the field, less than a third of the companies that adopt AI get tangible results. The rest? Unused subscriptions, disillusioned teams and budgets spent with no measurable return.

The technology works. The gap between intention and reality comes from the way SMBs approach the subject. They move too fast, won over by spectacular demos and promises of instant productivity. Yet generative AI changes how people work, and that takes far longer than rolling out a standard SaaS tool in a morning.

After helping about 20 SMBs through this transition, I see these 3 mistakes come up every time, and all of them can be avoided.

Mistake 1: buying a tool with no defined use case

This is the most common mistake. An executive reads an article about a new AI tool, sits through a convincing demo or gets a recommendation at a business dinner. The next day, the team has a new subscription and no clear idea of what to do with it.

I saw a 45-employee SMB in Nantes sign up for 3 different content generation tools at the same time. Total monthly cost: €850. Actual use after 2 months: a single employee was using one of the 3 tools, now and then, to rephrase emails. The other 2 licenses sat idle.

The instinct is to start from the tool. The right move is to start from the problem. Before subscribing to anything, ask 3 simple questions:

• Which task eats up time without creating any distinctive value?
• What measurable result do we expect from automating that task?
• Who will use the tool every day, and is that person involved in choosing it?

If you can't answer these 3 questions, you're not ready to buy.

Mistake 2: underestimating change management and training

Many executives think putting an AI tool in an employee's hands is enough, and that adoption will happen naturally because the technology is “intuitive.” That's rarely the case.

It isn't technology that transforms a company, it's its teams' ability to make it their own. And that ownership can't be decreed, it has to be built.

Cédric Villani, France IA 2025 seminar

In the SMBs I work with, the pattern is almost always the same. In week 1, enthusiasm dominates. In week 2, the first doubts appear. After a month, without structured support, most employees have dropped the tool or only use it for trivial tasks. That's what I see in the field, and studies confirm it: companies that invest in training reach an adoption rate of 80%, compared with less than 30% without support.

Training can't be reduced to a 2-hour workshop. What you need is a progressive program: an introduction to the concepts, hands-on workshops on the team's specific use cases, then regular follow-up with checkpoints. Companies that succeed also appoint internal “AI champions,” people who become the go-to contact for questions and best practices.

Skipping training costs far more than training. A €200-a-month tool used at 15% of its potential means €170 a month wasted.

Mistake 3: ignoring data governance and compliance

This is a blind spot for many SMBs. Generative AI runs on data. When an employee pastes a client brief into ChatGPT, when a marketing team feeds its contact database into an AI tool, when a salesperson uses an assistant to personalize proposals, data is moving. And that data is subject to GDPR.

In 2025, the CNIL (France's data protection authority) stepped up its checks on how companies use AI. It published clear recommendations on algorithmic transparency, the consent of the people concerned and minimizing the data sent to models. Ignoring these recommendations exposes your company to penalties, and also to a loss of trust from your customers.

In practice, every SMB that brings in AI should ask itself these questions:

• What data is sent to the AI tools we use?
• Do these tools store our data? Do they use it to train their models?
• Have we informed our employees and customers about these uses?
• Have we updated our record of processing activities?

This is a concrete matter of trust and accountability. For a hands-on approach, I wrote a complete guide to anonymizing data with Presidio before handing it to an AI.

A structured approach: the 4-phase integration framework

SMBs that get concrete results apply a method. This is the framework I use with my clients, tested on more than 20 engagements.

The framework flexes with company size, the team's digital maturity and the available budget. The logic stays the same: understand before choosing, support before rolling out, measure before scaling.

What successful SMBs do

After 2 years of working with SMBs on these topics, I see a few constants among the ones that make it work.

They start small. One use case, one team, one measurable target. They skip the 18-month strategic plan and aim for proof of value in 6 weeks.

They involve end users. Tool choice stays close to the people doing the work: the employees who will use AI every day take part in selection, testing and validation.

They measure from day one. Before rollout, they document the starting point: time spent on the task, volume produced, error rate. After rollout, they compare. Without that baseline, there's no way to prove ROI.

They accept iteration. Prompts get refined. Workflows get adjusted. Processes evolve. AI behaves like a living system: it improves with use and with feedback from the field.

Generative AI is a real opportunity for SMBs, and one that takes preparation. The 3 mistakes described in this article are common, predictable and costly. Avoiding them already puts you ahead of most companies that jump in without a method.