Implementing AI in a small business: where do you start?

Implementing AI in a small business starts with one concrete process that comes back every week, not with a tool. You describe how the work runs today, write down what the AI may see and do, and test one prototype on real data. At Goudbeek in Almere, that is how we work: process first, technology second.

That sounds less exciting than a demo of the latest model. But the hard part is rarely the technology. It is being clear about which problem you are solving and who will work with the result. This piece walks through the steps, from the first choice to the moment you expand.

Why start with the job, not the tool?

Starting from 'which AI tool do we need?' means looking for a problem to fit a solution. Start with the work instead: which task eats time, causes mistakes, or stalls when someone is off sick?

Picture an online shop that answers the same questions about delivery times and returns every day. The job is not 'a chatbot'. The job is giving customers a correct answer quickly, based on the returns policy and the order status.

Only once the job is clear do you choose which of the AI solutions for businesses fits. Sometimes that is a chatbot. Sometimes it is a better FAQ page or a reply template, and that is a good outcome too.

Which process should you pick first?

Pick one process that comes back every week, has a clear start and finish, and currently takes manual work. The more often it happens, the sooner you can tell whether an improvement actually helps.

Think of an installation company where someone retypes the engineers' job sheets into the admin system. Or an accounting firm that sorts and enters clients' receipts and invoices every month, work that document processing with AI is built for. Dull work, and that is exactly what makes it a good candidate.

Avoid starting with the process that has the most exceptions, or where a single mistake has serious consequences. You get to those later, once you know how a system behaves in your business.

Then write down how the work really runs today: who does what, with which information, and where the work waits for someone. That description often turns up improvements before anything is built. More on that approach under improving business processes.

What may the AI see, and what may it do?

Before anything is built, write down which sources the system may read and which actions it may take. A system that answers returns questions needs the returns policy and the order status. It does not need the whole shared drive, payroll included.

Separate what can be undone from what cannot. Summarising, sorting and drafting a reply are things the system can do on its own. Sending, paying, deleting or changing a customer record waits for a person by default.

If you process personal data, the GDPR applies. The EU AI Act adds obligations of its own, depending on what you use AI for and how. Check the official texts for your own situation. How we set up access, approvals and logs is described on our AI security page.

Why one working prototype on real data?

A demo with tidy sample text tells you little about your business. Real job sheets are messy, real customer questions are incomplete, and real invoices each have a different layout. Only on that data do you see what works and what does not.

So build one small, working prototype for that one process, and run it on real data, properly shielded where needed. Let the people who do the work today judge the results. They spot problems fastest.

Keep it deliberately small: one kind of input, one output, one person reviewing. You can expand once the principle holds.

How do you measure whether it works, and when do you expand?

Agree up front what you will look at. How much manual work is left? How often does someone have to correct an output? Does the work reach the next person sooner? Record how things run now, so you have something to compare against.

A system that goes live is not finished. You adjust instructions, sources and boundaries based on what you see in real use. An action that waits for a person now is only released once you have watched it go right for long enough. Whatever cannot be undone stays with a person by default.

When the first process runs steadily, pick the next one. Sometimes that extends the same system; sometimes it is a different process altogether. That way AI grows in your business one step at a time, rather than as one big project.

What does your team need to learn?

An AI system changes people's work, so those people belong in the project from the start. They need to know what the system does, where it can go wrong, and when to step in.

It also helps when your team is comfortable with the tools it already has, such as ChatGPT, Claude or Copilot in Microsoft Teams. Asking clear questions, checking outputs and knowing which data never goes into a tool are skills you can practise. That is what our AI workshops for teams are for, where your team works on its own tasks.

The AI Act also asks organisations that use AI systems to ensure, as far as they can, a sufficient level of AI literacy among staff who work with them. The official text sets out what that means for you.

When is AI not the answer?

Not every problem needs AI. If a process stalls because nobody knows who is responsible, a model will not fix that. If the work follows fixed, predictable rules, ordinary process automation or a form with field checks is often simpler and more predictable.

Where a mistake has serious consequences and nobody can check the output, caution is wise too. So a good analysis sometimes ends with the advice not to use AI. That is not a failure; it is what the analysis is for.

How does Goudbeek approach AI implementation?

We start with the work, not the technology. Together we map how the process runs today and choose the simplest thing that works. Sometimes that is an AI agent or a chatbot; sometimes it is plain automation without AI.

Then we build one working prototype on your real data, with written rules on what the system may see, what it may do on its own and what waits for a person. After a short intake we agree a fixed price up front.

Once it is live, it runs on managed hosting from €39 per month and we adjust it based on real use. Your domain, code and content stay yours. Read more about how we implement AI for businesses, or get in touch to talk through your process. You will hear from us within one working day.