AI Automation for Small Business: Where to Start

Start with one process that you repeat often, that has clear rules and where mistakes cost little. Think of processing files, sorting mail or retyping data. Do not automate your whole business at once. One process that works well delivers more than ten half solutions, and you learn from it right away.

What is AI automation for a small business?

Automating processes has been possible for years: a program performs a fixed series of steps without anyone touching it. That works well as long as the input is tidy. An invoice in exactly the same format, a form with fixed fields.

AI adds something where the input is messy. A language model (the kind of AI that understands and writes text) can read an email and decide what it is about, read a file with odd columns anyway, or summarise a long document. Where ordinary automation gets stuck on an unexpected format, a language model can still make a reasonable guess.

For a business owner that is the core: the goal of AI automation in your company is that the boring, repeated work no longer lands on you or your staff. Whether a language model is involved is a detail of the implementation.

Which process do you automate first?

I use four questions. A process that scores yes on all four is a good first candidate.

1. Does it repeat often? Every day or every order, not a few times a year.
2. Are there clear rules? If you can explain what has to happen, a system can do it too.
3. Is it boring and error-prone? Retyping and copying go wrong when you are tired.
4. Can a mistake be fixed? A wrongly sorted email is bad enough to watch for, a wrongly sent invoice is worse. Start with the first kind.

A few examples that often work well:

• Sorting incoming emails and forms and passing them to the right person.
• Reading files from a supplier or customer automatically and putting them in your system.
• Having product texts or descriptions drafted, after which a person checks them.
• Preparing a first answer to frequently asked customer questions.

What does this look like in practice?

Two things I built myself, without naming names.

For a logistics company I made a bot that processes files automatically every day. The process was boring, it came back daily and the rules were clear. That is the ideal profile: nobody has to open and work through those files by hand any more.

For a webshop I built an AI pipeline that rewrote more than 2,000 product pages in three languages. It has a research step, a writing step and a checking step. The checking step matters here. AI writes fast, but it also makes mistakes, and a second step that reviews the work catches a lot before it goes live. I described exactly how that works in how I rewrote 2,000 product pages with AI.

In both cases I did the same thing: one clearly bounded process, clear rules and a way to see whether it is right.

When does AI automation not pay off?

You rarely hear this from people who want to sell you something, so I say it here.

It rarely happens. A task you do a few times a year is not worth building. Do it by hand.

The rules do not exist. If every situation needs its own judgment and you cannot explain it yourself, a system cannot do it either.

A mistake is expensive. For decisions about money, legal matters or health you want a person to decide. AI may prepare, not settle.

Your process is still messy. Automating makes a bad process badly faster. Tidy up first, then automate.

A standard integration is enough. Many programs already connect to your bookkeeping or webshop out of the box. Then you do not need anything built.

Where does it run and what happens to your data?

An automation has to run somewhere, and that is a choice. You can use a provider's service, or run it on a server of your own. I host automations on my own servers, so you know where your data is. That is especially useful when customer or staff data is involved.

As soon as personal data is in a process, the GDPR applies. Among other things, that means you must know which data you process, why, and with which party you have agreements about it. If you use an external AI service, you send data to that party. So ask beforehand what they do with it and put it in writing. The Dutch Data Protection Authority (Autoriteit Persoonsgegevens) published guidance on generative AI and the GDPR in July 2026, which you can find on its site.

How do you go about it, step by step?

1. Pick one process with the four questions above.
2. Write down what happens now: who does what, with which data, and where it goes wrong.
3. Decide what good looks like. How do you recognise a correct result? You want to know that before you build anything.
4. Build small and let a person watch along at the start.
5. Measure and expand. If it works, pick the next process.

If you want someone to look at which process pays off first, you can do that with an AI audit. What I build in automation is on the page about AI automation.

What does it cost?

AI projects with me are on request, with a fixed price up front. Extra work afterwards is billed at € 95 per hour in blocks of 15 minutes. The pricing page explains how that works. If you doubt whether your process is worth it, get in touch and I will tell you honestly, even when the answer is no.

Frequently asked questions

Where do I start with AI automation in my business?

Pick one process that you repeat often, with clear rules and where a mistake can be fixed. Build that out small first and have a person check it. Only then choose a second process.

Is AI automation only for large companies?

No. Small businesses often have a lot of manual work that costs a person, and a well-bounded automation can make a big difference there. The process does have to come back often enough to be worth building.

What is the difference between automating and automating with AI?

Ordinary automation follows fixed rules and works on tidy input. AI can also handle messy input, such as an email in free text. You often combine both: fixed steps where possible, AI where needed.

Can I have AI automation built in the Netherlands by one point of contact?

Yes. I work from Borne in Twente as a one-person studio, so you talk to the person who builds it. I host automations on my own servers. If you use an external AI service, data does go to that party, and I discuss that with you beforehand.