Having an AI chatbot built for your business: an honest guide

A real business chatbot versus a toy widget

Most people picture a chatbot as that scripted bubble that pops up with “Hi, how can I help?” and then fails the moment you ask something off-script. That is a decision tree dressed up as intelligence. It can only do what someone explicitly programmed it to do, and it breaks on the first question phrased in a way nobody anticipated.

A modern AI chatbot works differently. It runs on a large language model and understands intent, nuance and context, even when a customer types messily, switches languages or asks two things at once. The difference is not cosmetic. One takes work off your team because it genuinely resolves questions; the other quietly trains your customers to skip the bot and ask for a human straight away.

When you decide to have an AI chatbot built, this is the line that matters most. You are not buying a widget, you are buying a layer of judgement that sits between your customer and your business. That layer is only as good as what it knows and how carefully it was set up.

Why training on your own content changes everything

A language model out of the box knows a lot about the world and nothing about your business. It does not know your prices, your return policy, your delivery times or the name of the product you launched last month. Ask it directly and it will either refuse or, worse, invent a plausible-sounding answer. For a customer-facing tool, a confident wrong answer is far more dangerous than no answer.

The technique that solves this is called retrieval-augmented generation, or RAG. In plain terms: your own documents, your website, your FAQ, your product catalogue and your policies are indexed and stored. When a customer asks a question, the system first retrieves the relevant passages from your material, then asks the model to answer using only that source. The model becomes a careful reader of your knowledge rather than a guesser.

This is the single biggest factor in whether a chatbot feels trustworthy. A well-built RAG setup can point at where an answer came from, stays current when you update a document, and refuses gracefully when something falls outside its knowledge. That is the foundation we build every AI chatbot on.

Where a chatbot actually pays for itself

Three use cases carry most of the value. The first is customer service: a chatbot trained on your knowledge base handles the repetitive forty to sixty percent of incoming questions instantly, at any hour, leaving your team the cases that genuinely need a person. The second is lead qualification: instead of a static contact form, the bot asks the right follow-up questions, understands what the visitor needs and passes a warm, structured summary to sales. The third is booking and scheduling: the bot checks availability, proposes slots and confirms appointments directly in your calendar.

What these have in common is volume and repetition. If something happens dozens of times a day and follows a recognisable pattern, a chatbot is a strong fit. If it happens twice a month and every case is unique, automation rarely earns its keep. The honest test is not “could a chatbot do this” but “does this happen often enough to be worth automating”.

The most valuable chatbots rarely stop at the conversation. They tie into the systems behind them: your CRM, your inbox, your booking tool. That handover between talking and actually doing is where a chatbot becomes part of your AI automation rather than a standalone novelty.

What good actually looks like

A good chatbot knows its limits. It answers what it can defend from your material and hands off cleanly to a human the moment it is uncertain, rather than bluffing. It keeps the context of the conversation, so a customer never has to repeat themselves. It speaks in your tone of voice, not in generic support-bot phrasing. And it leaves a trail: you can see what people asked, where it succeeded and where it struggled.

Equally important is what a good chatbot does not do. It does not promise things you cannot deliver, it does not guess at prices, and it does not pretend to be human when asked directly. These guardrails are not a limitation, they are what makes the tool safe to put in front of real customers. A chatbot that occasionally says “let me connect you with a colleague for this” earns far more trust than one that answers everything and is sometimes wrong.

The build process and a realistic timeline

A serious build follows a clear arc. It starts with discovery: which questions come in, where the volume sits, which systems need to connect, and what tone fits your brand. Then comes knowledge preparation, where your content is gathered, cleaned and indexed, because the quality of the answers depends entirely on the quality of the source. After that the conversation design and the guardrails are set, the integrations are wired up, and the bot is tested against real historical questions before it ever meets a customer.

For a focused chatbot with a single clear goal, expect roughly two to four weeks from kickoff to a live pilot. A broader build with multiple integrations, several departments or complex workflows typically runs six to ten weeks. The first two weeks after launch matter most: you watch the real conversations, find the gaps and refine. A chatbot is not a project you finish, it is a system you tune.

What it costs: one-off and monthly

Cost splits into two parts, and any builder who hides one of them should give you pause. The first is the one-off build. A focused chatbot with a tight scope and a clean knowledge base typically lands between 2,500 and 6,000 euro. A more substantial build with multiple integrations, custom workflows and a larger knowledge set runs from roughly 7,000 to 20,000 euro and up, depending on complexity.

The second part is the monthly running cost, and it is real. It covers the model usage (behind the scenes you pay per conversation), the hosting of your knowledge index, monitoring and maintenance. For most small and mid-sized businesses this sits somewhere between 150 and 600 euro a month, scaling with the number of conversations. A bot that handles thousands of conversations costs more to run than one handling dozens, simply because the underlying model usage scales with traffic.

The point of these ranges is not precision but realism. The right way to judge the cost is against the work the chatbot removes. If it takes even a few hours of repetitive support a day off your team, the monthly fee is modest against an employee doing the same. If it qualifies leads that would otherwise leak away through a dead contact form, it pays for itself in conversions long before it pays for itself in saved hours.

When you should not build one

Honesty here separates a partner from a vendor. A chatbot is the wrong tool when your information is genuinely thin: if you have no documented answers, no FAQ and no consistent policies, there is nothing for it to learn from, and you would be automating chaos. It is also wrong when your interactions are few in number but high in stakes, such as bespoke legal or financial advice, where every answer has to be human and accountable.

And sometimes the better answer is automation without any chat interface at all. If the real bottleneck is internal, like processing orders or routing email, a quieter workflow behind the scenes serves you better than a bot on the website. The right question is never “do we want a chatbot” but “what is the actual bottleneck, and is a conversation the right shape for solving it”. We would rather tell you a chatbot is not the answer than sell you one that disappoints.

How to choose who builds it

Ask any builder three questions. First: how do you handle wrong answers and hallucination? A serious answer is about grounding in your content, clear guardrails and graceful handover, not about “the model is very accurate”. Second: what happens after launch? A chatbot needs monitoring and tuning, so a one-and-done delivery with no aftercare is a warning sign. Third: who owns the knowledge and the data? You should keep control of your own content and conversations, and know where the data flows.

Look for a builder who talks about your business before they talk about technology, who is comfortable saying no, and who can show you the difference between a chatbot and a broader strategy. If you want to dig deeper, our pieces on implementing AI agents and on whether a chatbot is a gadget or a gamechanger are a good next read. The right partner leaves you with a tool you understand and can grow, not a black box you are afraid to touch.

A calm next step

You do not need to have everything figured out to start. The most useful first move is a short, honest conversation about where your real volume sits and whether a chatbot is the right shape for it. Sometimes the answer is yes and we map a clear path, sometimes it is a quieter automation instead. Either way you leave with clarity rather than a sales pitch. When you are ready, get in touch and we will think it through with you, with no pressure attached.

Frequently asked questions

How long does it take to have an AI chatbot built?

A focused chatbot with a single goal is usually live as a pilot within two to four weeks. A broader build with multiple integrations runs six to ten weeks. The first two weeks after launch are spent tuning it against real conversations.

What does an AI chatbot cost?

A focused build typically costs between 2,500 and 6,000 euro one-off; larger builds run from 7,000 to 20,000 euro and up. On top of that, expect a monthly running cost of roughly 150 to 600 euro for model usage, hosting and maintenance, scaling with the number of conversations.

Will the chatbot make things up?

A well-built one will not, because it is grounded in your own content through RAG and only answers from that source. When something falls outside its knowledge, it hands off to a human rather than guessing. Guardrails against invented answers are the core of a serious build.

Can the chatbot connect to our systems?

Yes. The most valuable chatbots connect to your CRM, inbox, calendar or booking tool, so they can actually do something rather than only talk. The specific integrations are scoped during discovery, based on where your real workflows live.

Do we keep ownership of our data and content?

You should, and it is a fair question to ask any builder. You keep control of your own content and conversation data, and you should know exactly where that data flows. If a builder is vague about this, treat it as a warning sign.

What if a chatbot is not the right solution for us?

Then we say so. If your real bottleneck is internal, or your interactions are few in number but high in stakes, an automation behind the scenes often serves you better than a bot on the website. The goal is to solve the bottleneck, not to sell a chatbot.