Where to start with AI when you are an SME

If you run an SME, you have probably already tried ChatGPT, read three contradictory articles and closed the tab thinking "we'll see later". That is normal: the topic is vast, the promises are huge, and the fear of getting it wrong is real. Yet the gap keeps widening between companies that use AI daily and those that are waiting for the right moment.

The good news: getting started with AI needs neither a huge budget, nor a technical team, nor an IT overhaul. It needs a method. Here is one, proven in the field.

1. Forget the tools, start from your pain points

The first mistake is to pick a tool before defining a problem. You subscribe to a trendy app, try it for two weeks, then drop it because it fits no real need. The result: money spent and the feeling that "AI is not for us".

Take the problem the other way round. List what wastes your teams' time, every week, on repetitive tasks with no added value:

• Answering the same customer questions several times a day.
• Re-entering information from one piece of software into another.
• Searching for a document or a procedure in a badly organised shared folder.
• Writing meeting notes, quotes or follow-ups.
• Sorting and routing incoming emails.

Each of these pain points is a candidate for AI. And above all, each has a measurable cost in time that you can compare to the expected gain.

2. Quantify the value before quantifying the solution

For each pain point, ask three simple questions: how many hours per week does it represent? How many people are involved? What could they do better with that time? A use case that saves two hours a day for three people has nothing to do with a gadget that impresses in a demo but serves no one.

This step gives you a natural prioritisation. Usually two or three use cases clearly stand out: those are the ones to tackle first.

3. Start small, but start for real

Once the use cases are identified, the temptation is to do everything at once. Bad idea. You build one use case, just one, and put it in your teams' hands in real conditions. Not a demo: a real test, on your real data, for two to three weeks.

This prototype has two virtues. It proves the value concretely, with figures, before any heavy commitment. And it brings your teams on board: nothing convinces a sceptical employee better than saving an hour on a task they hate.

4. Keep control of your data from the start

Many SMEs give up on AI out of a legitimate concern for their data. The question deserves to be raised from the first project: where do your documents, your exchanges, your client files go? With consumer tools, the answer is often "onto US servers, out of your control".

Alternatives exist. Depending on how sensitive your data is, AI can run in a French sovereign cloud, or directly on your own servers. Handling this point from the start saves you from rebuilding everything the day a client or an insurer asks you for guarantees.

5. Train, measure, then expand

A tool nobody uses is worthless. Once the first use case is in production, train the teams involved, not in a lecture, but on their own tasks. Measure the real gain after a few weeks. And only then move on to the next use case.

This step-by-step approach avoids the tunnel effect of big projects that never ship. You move forward use case by use case, each profitable, each validated before moving on to the next.

And if you still don't know where to start?

That is exactly what a diagnosis is for. In a few exchanges, we map your company, identify your two or three most profitable use cases, each costed, and you leave with a clear plan. Whether you implement it alone, with us or with another provider, you will finally know where to put your first euros.