AI chatbot cost: what the price actually buys
An AI chatbot costs 15 euros a month at one end and several hundred thousand for a single build at the other. Both ends of that range are real. The gap is not about the technology. It is about which of three quite different things is being bought.
One defined business process, with an agent working inside internal systems, runs €7,000 to €15,000. That is the full price, covering the deliverables, a working process and a conversational interface where one makes sense.
This guide gives the three tiers and what each one buys. It then covers the part nobody puts in a quote, which is that most of the money does not go on software at all.
In this guide
AI chatbot cost, by tier
Three tiers cover almost every case, and the jump between them is not gradual. Each solves a different problem, and the common mistake is buying the third when the second is enough.
The ranges below come from published regional pricing, checked against what we see on our own projects.
| Tier | What it does | One-off | Monthly |
|---|---|---|---|
| 1. Self-service | Answers questions from a fixed list and captures contact details | €0 to €150, mostly internal time | €15 to €100 |
| 2. Built on existing content | Reads the pages and documents, answers questions about specific services and prices | €300 to €2,000 | €50 to €450 |
| 3. Connected to internal systems | Books, checks availability, opens tickets, reads internal tools | €7,000 to €15,000 | €150 to €1,000 and up |
Tier one suits a business with a short and stable list of questions, where the website already covers the ground and people still ask anyway.
Tier two is where most companies land. The assistant reads existing material rather than a script maintained by hand. That is what lets it answer about one specific service or one specific price.
Tier three is a different purchase. It stops being about answers and starts being about actions, and that distinction is covered in the guide on chatbot vs AI agent.
Nothing sits between tier two and tier three, and that is not an oversight. Below €2,000 the purchase is answers. Above €7,000 it is custom development against one process, with a measurable outcome.
Why published prices differ by up to five hundred times
Published figures put custom development at €2,000 on one page and enterprise work at up to €1,000,000 on another. Both figures are honest. They describe different buyers, and both pages were written by somebody selling.
Three published guides, checked in August 2026:
- A regional agency guide puts a standard setup at €300 to €1,200, and custom development at €2,000 to €10,000 and up.
- A development shop lists €3,000 to €15,000 for a question-and-answer bot, and €30,000 to €80,000 for an agent that takes actions.
- A guide aimed at the US market puts an AI-powered chatbot at €75,000 to €150,000, and enterprise work at €200,000 to €1,000,000 and up.
The last one is not wrong. It describes a buyer with a procurement process, a compliance department, and systems that predate the web. A clinic or a webshop is simply not that buyer. Reading it is a common reason a sensible project gets abandoned before it starts.
Paying per resolved conversation
A newer model charges per conversation the assistant actually resolves, rather than per seat or per month. Published rates from the larger vendors run from about €0.45 to €1.75 a resolution, and the wider market spans roughly €0.10 to €3.
The model is worth understanding even when nobody buys it, because it forces the right question. At €1 a resolution, a month with a thousand resolved conversations costs €1,000. Whether that is cheap depends entirely on what those thousand conversations are worth.
Most of the cost is not software
In our experience, the development work is the small part. The cost sits in finding the content, reconciling what it says, and testing that the answers are right before anyone sees them.
Typical timelines from our own projects:
- Two to five days for a straightforward question-and-answer assistant on an existing website.
- Five to ten days for a larger site with a lot of content and priced services, most often a dental practice.
- Four weeks and up once it connects to internal systems, for example when it takes a booking inside the conversation.
Almost none of that is writing software. It is content archaeology, and it is the reason two companies of the same size get quoted differently for the same tier.
What actually drives the price up
Four things, and all four sit on the buyer's side of the table. That is better news than it sounds, because it puts the cost largely within reach.
- There is less written down than most people expect. The most common cause by a distance. A large share of what customers ask has no written answer anywhere.
- The content contradicts itself. Prices are the worst offender. The page, the PDF and the brochure disagree, and somebody has to decide which is right before anything gets built.
- Some of it is out of date. Last year's price list is still on the site and reads as current.
- Test questions arrive slowly. Checking the system needs real questions with agreed answers, and this is where projects stall longest.
None of these is a technical problem. Fixing the first three is worth doing whether or not anything gets bought. The guide on preparing your content for an AI assistant explains how.
Who it pays for, and who it does not
No volume threshold decides this, and anybody offering one is guessing. What decides it is the value of a single resolved conversation, and there are two separate ways for that to work.
When one conversation is worth a lot
Some businesses need very little volume. Think of a dental practice that captures one implant consultation, a supplier that captures one qualified enquiry, or an agency that books one discovery call. A single conversation can cover a year of either of the first two tiers.
For these, the assistant is not a cost-saving tool. It answers at eleven at night while somebody is still deciding, and it captures the enquiry that would otherwise go to whoever replied first.
When one conversation is worth almost nothing
The opposite case, and it is just as valid. A webshop answers "Where is my order?", "What are the delivery terms?" and "Do you ship to this address?" Individually those are worth cents. A few hundred of those a week take up most of a person's working time.
Here the assistant genuinely is a cost-saving tool, and the arithmetic is easy. Count the repeat questions in a week, multiply by the minutes each one takes, and price that against the tiers above.
When it does not pay
If a single conversation is worth little, and there are not many of them, none of this pays back. A small local business gets a handful of enquiries a week, and the owner is there to answer them. Buying a tool for that is buying a tool nobody needs.
One more case is worth naming. Where every question is genuinely different, and answering needs judgement rather than lookup, the assistant will not answer it. It can still route the question and take the details, which is worth something, but do not buy it expecting answers.
The costs nobody puts in a quote
Four recurring costs sit outside the setup fee, and they are the reason a cheap quote sometimes ends up expensive.
- Usage. Every conversation costs something to run. Where pricing is usage-based the bill rises with the number and length of conversations rather than the number of users, and some platforms charge per seat on top.
- Content upkeep. Prices change, and the source the assistant reads has to change with them.
- Re-testing after changes. A change to prices or terms changes what the system is entitled to say, so the test questions get run again.
- Extra languages. Another language raises both the setup cost and the running cost, modestly rather than dramatically.
Content upkeep is the expensive one, and the fix is a person rather than a budget line. Somebody has to own the content the assistant reads. We have seen a system relay a price for months after it had changed. The page it read was never updated, because nobody had that job.
Most of what any supplier charges is for the content work, not for code. We built the platform and the evaluation framework before the first client. The days on a project go into mapping the sources, reconciling the conflicts, and running the test questions with the client before go-live. The shape of it by sector is visible on our AI assistant in healthcare page.
Frequently asked questions
Short answers to the questions we get asked most.
How much does a chatbot for a website cost?
A chatbot built on existing content costs between €300 and €2,000 to set up and €50 to €450 a month to run. Setting one up on a ready-made platform costs €15 to €100 a month plus internal time.
How much does an AI agent that takes actions cost?
One clearly defined business process runs €7,000 to €15,000. That is the full price, covering the deliverables, a working process and a conversational interface where one makes sense. Running costs sit on top, from €150 to €1,000 a month and up, depending on conversation volume.
Why does one agency quote €2,000 and another ten times more?
The two agencies are describing different work for different buyers. The high figures come from guides written for large organisations with procurement and compliance requirements. They do not describe what a clinic or a webshop needs.
How long does it take to build?
A straightforward assistant on an existing website takes two to five days. A large site with priced services takes five to ten days. Once it connects to internal systems, expect four weeks or more.
What makes a chatbot expensive?
Content is what makes a chatbot expensive, almost always. Scattered information, prices that contradict each other between pages, and out-of-date documents all add days, and none of them is a technical problem.
Is a chatbot worth it for a small business?
It depends on what one conversation is worth, not on how many arrive. If a single enquiry is worth hundreds of euros, very little volume justifies it. If enquiries are worth very little, volume has to make up for it.
What are the ongoing costs after launch?
The ongoing costs are usage, content upkeep, and re-testing after a change to prices or terms. Usage rises with the number and length of conversations rather than with the number of users. Budget for somebody owning the content, and the guide on accurate AI answers explains why testing does not stop at launch.
AI chatbot cost is the wrong first question, and it is the one everybody asks first. The figures are knowable and the ranges above are real, but what decides it is the value of one resolved conversation. Once that is known, the rest is arithmetic. Then budget for the part nobody puts in a quote, which is the week or two somebody spends deciding what the content actually says.