AI for SMEs: where to start, concretely?

For many SME leaders, the question is no longer whether AI will enter their company, but how to introduce it without wasting time, without exposing their data, and without paying for a tool that ends up unused. AI is no longer the preserve of large groups: the first tools are accessible, often for a few dozen francs per month per user, and SMEs of all sizes are seizing them. But equipping yourself with a tool is not enough to derive value from it. This article takes stock of how SMEs really use AI, the gaps widening between them, and the methods to follow to start effectively.

Where do SMEs really stand on AI?

SMEs are adopting AI faster and faster, but remain clearly behind large companies. In Switzerland, a third of them now integrate it into their processes, and across the OECD, large companies use it more than three times as often as small ones. The movement is real, but uneven.

On the dynamics side, the trend is clear. In Switzerland, the share of SMEs integrating AI into their processes rose from 22% to 34% in one year, while the share of companies that never use it fell from 45% to 29% (AXA / Sotomo 2025 study). In France, the number of very small businesses and SMEs using AI doubled in a year, reaching 26% (France Num 2025 Barometer). Across the European Union, enterprise AI use rose from 13.5% to 20% between 2024 and 2025 (Eurostat).

These rates are not directly comparable, because they come from separate surveys, with definitions of the SME and question wordings that vary from one country to another. They nonetheless show a common trend. In Switzerland as in France and across the European Union, adoption is accelerating and the gap by company size persists.

But behind this progress lies a persistent gap by size. Across the OECD, 40% of large companies use AI, versus only 11.9% of small ones — a ratio of more than three to one, consistent across all G7 countries (OECD, 2025). European data confirm this gap: in 2025, 55% of large EU companies use AI, versus 30% of medium-sized and 17% of small ones.

The most striking gap is not so much in usage as in intentions. According to a Bpifrance Le Lab study of more than 1,200 executives, 58% consider AI important, even vital, to their company's survival within three to five years. But a nearly identical share (57%) has yet to define any strategy on the matter (Bpifrance Le Lab, 2025). It is precisely this gap between awareness of the stakes and taking action that the rest of this article seeks to bridge.

The message for a leader is twofold. On one hand, not having yet taken the plunge is nothing unusual, since the majority of small structures are in this situation. On the other, the gap with companies that are getting started is real and tends to widen, which makes “how to start” a genuine question of competitiveness.

Why do SMEs struggle to take the step?

SMEs do not stumble on the technology itself, which has become accessible, but on four recurring obstacles: the lack of in-house skills, data preparation and governance, uncertainty about the expected return, and the question of data sovereignty. These are organizational barriers more than resource barriers.

Institutional studies converge on the same sticking points:

These findings are shared by independent sources. The Swiss Academy of Engineering Sciences cites “the quantity and quality of data, the cost of implementation, new skills, and digital sovereignty” as the main obstacles (SATW). For its part, a Deloitte survey of Swiss executives confirms that deep integration of generative AI remains limited, the main obstacles cited being the lack of talent and technical skills and the difficulty of identifying viable use cases (Deloitte Switzerland, 2025).

One point deserves emphasis, because it changes how to approach the subject. Cost is not the main obstacle. A large OECD survey of more than 5,000 SMEs concludes that, since many tools are cheap or even free, the real barrier is not the ability to pay but the confidence to use them. The barrier most cited by non-using SMEs is in fact the perception that AI would be “unsuited” to their activity (57%), a perception that recedes as soon as a first concrete use is tried (OECD, “Generative AI and the SME Workforce,” 2025). In other words, the obstacle is often in the mind before it is in the budget.

At what adoption stage is your company?

The OECD distinguishes four SME profiles when it comes to AI, from simple trial to strategic integration. Placing your company on this scale helps choose the right next step, because you don't steer the same way when you've never tried and when you're seeking to scale up.

The majority of SMEs today fall into the first two profiles, where they use AI for simple tasks, without an overall strategy. That is not a problem in itself. It is even the right way to begin. The challenge is not to stay stuck there for lack of method.

Where to start, concretely?

The best starting point is not a big project, but a simple, frequent use case with a visible benefit — often writing, translation, or information retrieval. You frame data usage upfront, measure the real gain, then broaden. This small-steps progression is what separates a project that holds up from one that gets abandoned.

A five-step approach, applicable whatever the company's size:

This logic is exactly what public authorities recommend: start with a simple use case, such as writing assistance, to familiarize teams and measure a first return, before tackling more ambitious projects. The decisive factor is not the budget, but the method.

Which first AI uses for an SME?

The most accessible uses concern tasks present in almost every company, such as writing, summarizing, information retrieval, or document preparation. Depending on the function, this ranges from preparing bids in sales to document processing in administration. The challenge is not to do a lot, but to choose the right first one.

Here is a range of the most common and most realistic uses for an SME, function by function:

But this range is a trap if you try to do everything at once. An AI diagnostic surfaces on average fourteen use cases in an SME, the vast majority of which have strong productivity potential (Bpifrance Conseil, 2025). No company can launch them all at once. The real challenge is not to identify uses, but to prioritize the one that will bring the most value for the least effort, in a given context.

This is where a project's success or failure is decided. The finding from McKinsey points the same way. Globally, a significant impact of AI on the company's overall results remains rare. Only 39% of organizations report an effect, even a modest one, on their overall profitability, with most of the gains often confined to a few functions. Most often, it is not the technology that is lacking, but the prioritization of the right uses and their real adoption by teams. In other words, the question is not only “which tool,” but “which use, in what order, and for what measurable benefit.” That is precisely the framing work to carry out before getting started.

Which mistakes to avoid at the start?

AI project failures in SMEs almost always have the same causes: starting without a framework for data, aiming too big from the outset, choosing a tool before defining the need, and neglecting team training. These are mistakes of method, easy to avoid once you know them.

These mistakes explain why some SMEs invest without seeing results, while others, more methodical, derive real benefit from otherwise identical tools. The difference lies less in the technology than in the way it is introduced.