How can I use AI in my SME?

Everyone tells you to get started with AI. In Belgium, 35 percent of companies used it in 2025, according to the Belgian FPS Economy. The EU average is 20 percent. Among small companies the figure is 29 percent, among large ones 76 percent. The same report names the biggest barrier: not enough expertise in-house.

I recognize that one. You run a company of fifteen people. There is no IT department, and your calendar is already full. You once asked ChatGPT to rewrite an email. Now what?

Every guide says "start small". None of them says where

Put this question to Google and you get dozens of articles. They all look alike. First comes a list of use cases, from marketing copy to stock management. Then comes the advice: pick one process and start small.

The advice is sound. It just doesn't get you anywhere, because the hard part is still yours. Which process? A list of 23 options only makes that choice harder.

There is a simple way to choose. It starts with something few people tell you about AI.

AI only knows what somebody wrote down

ChatGPT has read a large part of the internet. Your company isn't on it. ChatGPT doesn't know your pricing agreements, or the exception you made for your biggest customer. It has no idea why the export runs slowly for some users.

That knowledge lives in people's heads. Yours, and your most senior developer's. It also sits in sent emails, where nobody will ever find it again.

Think of a new hire on day one. They are smart and quick. Without any onboarding notes, all they can do is guess, or ask you everything. AI is no different. A model that knows nothing about your business gives you a generic answer. That is fine for a LinkedIn post. It is not enough for your customers.

So the place where AI works for you is the place where the answer already exists.

The test: which question did you answer more than once this week?

Take a software company with its own product and a team of twelve. Every week, a customer asks how to get their data into Excel. Every week, somebody on the team types out that answer again. Usually it's a developer, because the developer is the one who knows.

That is where you start. Three things come together there.

The answer already exists, because someone has given it ten times. The risk is low, because the right answer is the same every time. And you can count what you gain. Say one answer takes ten minutes and the question comes in five times a week. That is close to an hour a week, for one question. Most companies have about twenty like it.

In most SMEs, the repetition sits in the inbox. Customers, suppliers and your own colleagues ask things you have answered before. I wrote earlier about how much of your workday goes into that. It is more than you think.

Where AI doesn't belong yet

The test works the other way round too, and that matters just as much.

Does an answer call for a judgment with consequences? Then a person should give it. An angry customer or a contract with unusual terms stays with your team. That is also the work where your people make the difference.

And if you can't find a single question that keeps coming back? Then you don't need an AI project right now. I mean that. Buying a tool because everyone is talking about it gets you a subscription and a bad feeling. Check again in six months.

Write the answer down once

OK, you have found your question. Now the answer has to be somewhere AI can read it.

This is where many people give up, because it sounds like a documentation project, and nobody has time for one of those. Luckily, the material is already there. It's in your sent emails, in words your customers understand. All you have to do is collect it.

Start with the twenty questions that come up most. One question, one short article. Structure and layout can wait. If you have nothing today, here is how to build a knowledge base from nothing.

You do this work once. It pays off even without AI: a new colleague learns the job faster, and your senior developer gets interrupted less.

How this works in Sarrai

This is the part I build myself, so read it with that in mind.

Sarrai runs on that one rule. Your team answers a question once. The answer goes into your knowledge base. From then on, Sarrai answers that question on its own, in the chat on your website or by email. It only uses what you have written down.

If Sarrai isn't sure, or the answer isn't there, the question goes to a colleague. Your customer doesn't notice. That colleague's reply becomes the start of a new article, so next time Sarrai can handle the question.

Your people stay in charge. Sarrai takes the repetition, and your team has time left for the customer with the hard problem.

One question business owners almost always ask me: is this safe for my company data? Sarrai runs on European infrastructure and processes your data only within Europe. The AI models are never trained on your content. The details are on the security page.

Have your team keep a tally for a week

Repetition is hard to see. The person explaining the export for the fourth time doesn't stop to think about it. It's just work. And everyone only sees their own share, so nobody knows how big the total is.

Next Monday, ask your team to keep a tally for one week. Pen and paper will do. Every time someone explains, answers or does something they have done before, they add a mark, with two words about the topic.

Be clear about why you are asking. Nobody is being assessed. You are looking for the work nobody enjoys.

On Friday, lay the lists side by side. You will see which questions landed on three desks in the same week. Those are your first candidates, and you didn't have to buy a tool to find them.

The biggest effect comes afterwards. Someone who has kept a tally for a week keeps noticing. "I explained this last week as well." That habit is worth more than any step-by-step guide, because from then on your team points out where AI can help.