An AI chatbot for a small business: when it makes sense and when it doesn't
An honest look at AI chatbots for small businesses: what they can and can't do, what a build looks like, what GDPR means for it and how to tell if a chatbot will pay off.
An AI chatbot can answer customers at 11 pm, on a Sunday, and while you're with another client. It can also annoy people, make up a price you don't charge, and put off someone who just wanted to call. Which version ends up in your business depends on how it's built.
I build AI assistants for small businesses, and my own, Rubin, runs on this site. In this post I say plainly when a chatbot makes sense and when it's a waste of money.
How an AI chatbot differs from an old-style chat
Old chatbots ran on rigid scripts: click "Opening hours", get the opening hours. Any question outside the list ended with "Sorry, I don't understand".
An AI chatbot understands plain language. Customers write the way they talk: "could you replace a bathroom tap on Saturday, and roughly how much would it be?". The assistant looks for the answer in the information about your business and replies in full sentences.
The key is in the words "the information about your business". A good assistant answers only from what you give it: your price list, hours, policies, services. When it doesn't know something, it says so and passes the question on to you.
When an AI chatbot makes sense
In my experience a chatbot earns its keep when at least two of these are true:
- The same questions come back every day. Hours, directions, rough prices, availability, what to bring. If you answer them several times a day, an assistant takes that off your plate.
- Customers write outside working hours. An evening enquiry you answer the next morning has often already gone to a competitor.
- You want enquiries collected in a tidy way. The assistant can ask for the details (address, date, scope) and send you a ready enquiry by email or into a spreadsheet.
- You take bookings or appointments. The assistant can check free slots and book the customer in, instead of a long back-and-forth.
When a chatbot is a waste of money
- You get a few enquiries a week. Then it's quicker and cheaper to reply yourself.
- Every job needs an individual quote or a site visit. An assistant can collect the details, but it can't replace the conversation.
- Your customers prefer to call. In some trades nobody uses chat. A call-back automation or a voice agent will do better there.
- The basic information isn't written down anywhere. An assistant is only as good as the knowledge it's given. That has to be gathered first.
What a build looks like, step by step
- A conversation and a list of questions. We write down what customers ask most often and what the assistant should do when it doesn't know the answer.
- A knowledge base. I gather your offers, price lists, policies and FAQs in one place the assistant draws on.
- Rules and limits. We agree what the assistant never does: no prices outside the price list, no promised dates, no legal advice.
- Handover to a person. When a question goes beyond the knowledge base, the assistant takes the customer's contact details and notifies you.
- Testing on real questions. I test the assistant on questions your business has actually received, before the first customer ever sees it.
- Fixes after launch. For the first few weeks I read the conversations and fill in whatever the assistant is missing.
AI chatbots and GDPR
The assistant talks to customers, so it processes their personal data. That needs thinking about from day one:
- say in your privacy policy that an AI assistant runs on the site and what data it collects,
- collect only the data needed to handle the enquiry,
- check where conversations are processed and whether the AI model provider offers a proper data processing agreement,
- decide how long you keep conversations.
During the build I help sort this out on the technical side. The final legal assessment is worth checking with whoever handles data protection in your business.
What an AI chatbot costs
The cost has two parts: building the assistant once, and running it (AI model fees, a server, fixes). A simple FAQ-based assistant is a completely different scale from an agent that checks your calendar, books customers in and connects to your systems.
I go into more detail in the post on what automation costs a small business. The best comparison is against the time you spend today answering the same questions.
Frequently asked questions
Can an AI chatbot give a customer wrong information?
It can, if it's built badly. That's why the assistant answers only from your knowledge base, has clear limits and hands the conversation to a person when it doesn't know the answer. Before launch I test it on real questions from your customers.
Can an AI chatbot talk to customers in Polish and English?
Yes. Today's AI models handle Polish well, including casual questions and typos, and they answer in English or other languages too if you have customers from abroad.
Can the chatbot also work on WhatsApp or Messenger?
Often, yes. The same assistant can answer on your site and in the messaging apps your customers use. It depends on the channel and its rules, so we check this when we talk.
Where do I start if I don't know whether a chatbot would help?
For a week, write down the questions customers send you and the time you spend answering them. With that list it's easy to judge whether an assistant will pay off. Or book a free call straight away and we'll go through it together.
Want to check whether a chatbot would pay off for you?
Book a free call. We'll go through the questions your customers send, and I'll tell you honestly whether an AI assistant makes sense or whether it's better to start with something simpler.




