You answer the same questions every week. Opening hours, prices, whether you can take on one more job. An enquiry lands at nine in the evening, and by the time you see it the next morning the person has already called someone else.
So you start looking at those AI chats everyone’s talking about. Within ten minutes you’ve run into three completely different solutions and a string of companies all telling you theirs is the right one.
Which one fits your business? That’s the question this guide sorts out. We’ll walk through how to think about an AI chatbot for your business, without anyone trying to sell you something along the way.
Why it’s trickier than it sounds
The problem starts with the word itself. Three fairly different things all get called a chatbot, and whoever explains the difference usually makes their money from one of them.
There are ready-made services you pay a monthly fee for. There are simple bots you can build yourself in an afternoon. And there are custom solutions wired into your own systems.
They solve different problems, cost different amounts, and suit different businesses. Understanding the difference is what decides whether you spend your money on the right thing.
Three ways to get a chatbot
Buy it ready-made (a SaaS service)
The fastest route is to rent a ready-made service. You train it on your own content, meaning your website and your common questions, and have it running in no time. Services like ImBox and Tidio work this way.
The upside is that you skip building anything and get going fast. The downside shows up as you grow. Many services charge per conversation, and Tidio’s pricing (October 2026), for example, is counted in conversations per month, with a separate quota for AI conversations. That makes the cost hard to predict the more customers you get. And if the bot needs to plug deep into your own systems, that’s rarely what they do best.
This suits you if you have steady, predictable question volume and want to be up and running quickly.
Build a simple one yourself (inside an AI tool)
You can also build a simple assistant yourself, inside an AI tool you already pay for. You feed it your own text and instructions, and it costs essentially just the subscription.
The classic example has been a custom GPT in ChatGPT, but that particular route is closing. According to OpenAI’s help center (September 2026), personal accounts, Plus and Pro included, can no longer create new GPTs, and OpenAI plans to retire them in favour of what it calls plugins. The tools change, but the principle is the same.
But there’s a catch. An assistant like that lives inside the AI tool. It doesn’t sit as a chat box on your own website meeting your customers, and it doesn’t talk automatically to your booking system or your customer records.
As an internal tool for you and your staff, it’s excellent. As a public entry point on the site, it’s weak.
Custom-build it on the API (a custom build)
The third route is to build your own solution using, for example, OpenAI’s or Anthropic’s API. That gives you more control over data flows, integrations, logging, and how the whole thing is built.
It costs more and takes longer. But it’s the only way if the bot really needs to connect to how your business works, such as your booking system or your customer records.
What it lands at in money varies enormously with how advanced it gets. That’s a chapter of its own, and we cover the pricing in more detail in a separate article.
What vendors would rather not discuss: the handover to a human
Here’s the part almost always missing from the sales brochures. What happens when the bot can’t answer?
A good handover is called warm. The bot says it’s passing you on, gathers up what’s been said, and sends the whole conversation along to the person taking over. A bad handover is cold. The customer is shunted on without warning, lands in a queue, and has to start over from scratch.
You probably recognise the worst pattern as a customer yourself. You’ve explained your whole issue to the bot, get handed over, and then the human asks the exact same thing again. Everything you just wrote is gone.
When should a bot hand over? When the customer asks for it outright. When it has failed twice in a row. When the matter is sensitive or expensive, like a refund, a cancellation, or a complaint. And when you notice the customer starting to get irritated.
One thing isn’t negotiable. There must always be a visible path to a human. A bot measured on how many customers it manages to keep away from staff is built wrong. Before long someone’s sitting there typing HUMAN into a box and getting nowhere, and that’s not a customer you want to leave there. The US Consumer Financial Protection Bureau has a name for it, doom loops, and warned in a report on chatbots in consumer finance (June 2023) about customers stuck in repetitive answers with no way through to a human.
When you don’t need a chatbot (yet)
Sometimes the most honest answer is to wait. A chatbot doesn’t solve everything, and in some situations it’s the wrong investment right now.
If you have low traffic, the bot rarely does much good. If your site gets a few hundred visitors a month, it’ll barely be used. For a small local business, this is the most common reason to hold off.
If your enquiries are simple, a contact form is often enough. A name, a phone number, a short description, and you reply within an hour or so. A bot adds nothing there, and the form is also what catches the customer if the bot ever plays up.
And if you have a weak website, a bot is the wrong place to start. It’s an add-on to a site that works, not a plaster on one that doesn’t.
Some questions should never sit behind a bot either. Medical advice, legal matters, disputes over money. There a human should take care of the customer directly.
A simple check before you decide. Do you have steady visitor volume? Recurring questions? Are you losing business to late replies? Do you have time to set it up properly? And does your site already work as it should? If you can’t say yes to most of that, wait a little longer.
Two legal points to keep an eye on
Two rules are worth knowing before you get started.
The first is GDPR. When a bot handles your customers’ personal data, you need to sort out the roles. Usually your business is the data controller for how the bot is used, while the vendor may be a data processor, depending on how the service is set up. If the vendor is a processor, GDPR requires a data processing agreement between you, as the Swedish Authority for Privacy Protection (IMY) explains in its guide to processor agreements (in Swedish). Bear in mind, too, that “EU region” doesn’t always mean the data is actually processed within the EU. Ask where the data is processed, not just where it’s stored. IMY also has guidance on GDPR and AI (last updated May 2026, in Swedish) that’s worth reading.
The second is about transparency. Since 2 August 2026, users in the EU must as a rule be told when they’re talking to an AI system, under Article 50 of the EU AI Act. For a chatbot, that means it should be clear right there in the chat, not buried in the terms. What that means in practice we cover in a separate article on the AI Act and chatbots.
Frequently asked questions
Roughly what does an AI chatbot cost?
It depends on the route you choose. A simple bot you build yourself costs essentially just a monthly subscription. A ready-made service costs a monthly fee that grows with volume. A custom solution costs considerably more. We go through the full pricing picture in a separate article.
Can I build a chatbot myself?
Yes, a simple assistant inside an AI tool you already pay for is something you can build yourself in an afternoon. It works best internally. If it needs to sit on your website and meet customers, connected to your systems, you’ll need something more.
Do I have to tell people it’s an AI?
As a rule, yes. Since 2 August 2026, users in the EU must be told when they’re interacting with an AI system, under the EU AI Act. For a chatbot, it should be clear right there in the chat.
What happens when the bot can’t answer?
Then it should hand over to a human smoothly and pass along what the customer has already written. Always having a visible path to a real person is the single most important thing in the whole setup.
Where do you start?
Which route suits you depends on your question volume, the kinds of questions you get, and how your business actually works. That’s hard to answer in an article, but easy to sort out in a short conversation before you spend a krona.
Want to know whether a chatbot is genuinely the right step for you? Get in touch, and we’ll do a simple review of your situation first.