11 July 2026

The 5 most common mistakes when a small company starts with AI

Updated September 2026. These five mistakes are the ones that come up most often when a small company or sole trader starts with artificial intelligence. None of them happens through lack of willingness: they are mistakes of approach, and all of them can be avoided simply by knowing they exist.

Mistake 1: putting client data into free tools

This is the most widespread one and the one that can cost most. Somebody pastes a client list, a contract or an invoice into a free chat tool to have it summarised, and that information leaves the company without anyone having decided it should.

The problem is not the tool: it is that in free versions the contract does not say what you need it to say. Many allow what you type to be used to train their models, and offer no data processing agreement, which is what the GDPR requires when somebody processes personal data on your behalf.

What to do instead: use business versions with a signed agreement, or tools that do not retain the data. And write one simple rule about what never goes in: health data, ID numbers, payslips, contracts.

Mistake 2: automating the wrong process

People start with what looks good in a demo — a chatbot on the website, a text generator for social media — and leave untouched the thing that actually eats the hours: moving data from one place to another, chasing invoices, answering the same question forty times.

What to do instead: for one week, write down every task you repeat and how long it takes. At the end you will have a boring and very useful list. Start with whatever repeats a lot, is easy to explain and has no serious consequences if it goes wrong one day. That combination is what produces quick results.

Mistake 3: expecting AI to work by itself from day one

AI gets things wrong. Not occasionally: often, and with great confidence in itself. It invents data, misreads a document, confuses two clients with the same surname. Setting something up and walking away is the fastest route to a problem with a client.

What to do instead: in the first weeks, have a person check everything before it goes out. Once some time has passed and you know where it fails, you can decide what can run on its own and what cannot. For anything involving money, deadlines or commitments to clients, human sign-off stays.

Mistake 4: telling the team nothing

A tool is bought, announced in a meeting, and that is the end of it. The result is that half the team does not use it, the other half uses it their own way, and nobody knows what is allowed and what is not.

On top of that, since February 2025 there is a legal obligation involved: Article 4 of the EU AI Act requires organisations to take measures so that whoever works with AI knows what they are handling. I explain it in full in the EU AI Act guide for small companies.

What to do instead: a short session and some written rules already do the job for a small company. What matters is that a record of it exists.

Mistake 5: buying a generic solution and forcing yourself to fit it

A large platform gets bought because everyone uses it, and the way of working ends up changing to fit it. You pay for twenty features and use two.

What to do instead: start from a specific problem and look for the smallest solution that solves it. An automation built around how you actually work usually costs less and delivers more than a platform that has to be tamed.

How to start without falling into any of them

1. Make the list of repetitive tasks over a week, with the time each one takes.
2. Pick just one, the one that repeats most and carries least risk.
3. Check the tool's contract before any real data goes into it.
4. Keep human review while you learn where it fails.
5. Tell the team and put the rules in writing.
6. Measure whether it really saves time. If after a month it does not, drop it without regret.

That list of repetitive tasks is, in fact, the starting point of the material on automation.