AI chatbot for business: what it costs and when it pays off
Market prices run from a few tens of euros a month to several thousand for a deployment. The difference is not marketing — it is what the bot actually does. Here is how to tell them apart.
The short answer: off-the-shelf SaaS chatbots cost on the order of tens of euros a month. A custom chatbot that reaches into your systems is a one-off investment in the hundreds to low thousands of euros, plus monthly running costs. With me, a bot for frequent questions starts at €390 and a bot connected to a shop or a calendar at €700.
More important than the price, though, is whether a chatbot pays off for you at all. The deciding factor is the volume of repeated questions. Below roughly 30–50 repeated messages a week the investment usually does not come back, and you are better off answering personally. Below is a calculation you can run on your own numbers.
What actually determines the price
When two suppliers quote €300 and €3,000 for “a chatbot”, they are usually not quoting the same product. The difference tends to be these four things:
- Where the bot gets its answers. Hard-written answers are the cheapest. Answers from your documents cost more. Answers from live data (stock, orders) cost the most, because integration is needed.
- Whether the bot does something or only talks. Writing an enquiry into a CRM, booking a slot, checking a shipment — every action is extra work and another point that has to be tested.
- How much it can get wrong. A bot that answers questions about opening hours has nothing to get wrong. A bot that confirms appointments or quotes prices needs controls — and those cost time.
- Which language and which field. Everyday language in an everyday topic is fine. Specialist terminology needs testing and sometimes a narrowing of scope.
That is why nobody honest will quote you before asking what the bot has to know and what it has to be connected to.
Off-the-shelf SaaS vs a custom bot
This is the main decision, and most companies make it backwards — they pick a supplier first and only then ask what they actually need.
| Off-the-shelf SaaS | Custom bot | |
|---|---|---|
| Entry price | zero or low | one-off investment (hundreds to low thousands €) |
| Monthly cost | subscription, grows with volume | running costs: model API + hosting if needed |
| Time to launch | hours to days | days to weeks |
| Connection to your systems | limited to supported integrations | anything with an API or an export |
| Control over data | data sits with the provider | you decide |
| If the supplier folds or raises the price | a problem — the solution is theirs | the solution is yours, it can be moved |
| When it is the right choice | frequent questions, no integration, fast start | the bot has to reach your data or take action |
An honest recommendation: if you need answers to frequent questions and nothing more, start with an off-the-shelf SaaS product. It is cheaper, faster, and it tells you whether your customers use it at all. A custom build makes sense once you hit its limits — and by then you will know exactly what you need.
The costs people forget
The deployment price is not the whole cost. These are the items proposals often leave out:
- The language model API. Every conversation costs a fraction of a cent up to a few cents depending on length and model. At hundreds of conversations a month that is a small sum; at tens of thousands it is not. Always ask for an estimate at your volume.
- Preparing the source material. The bot needs a current price list, terms and descriptions. If those are scattered and out of date, tidying them is real work — and it usually pays off regardless of the chatbot.
- The first weeks of tuning. After launch the conversations have to be reviewed and the gaps filled. This is the most neglected part and at the same time the one that decides the result.
- Maintenance. Changed your prices or terms? The bot does not know until somebody tells it. Budget for a short regular update.
- Hosting. With a custom build, either a small server or serverless — on the order of a few euros a month, but not zero.
When the investment pays back — a concrete calculation
Instead of percentages you have no way to verify, here is a procedure you can run on your own numbers in two minutes:
- Count the repeated messages. How many messages a week are an answer to the same handful of questions? Not all messages — only the repeated ones.
- Estimate the time each one takes. Even a short answer costs an interruption. Realistically 2–5 minutes including getting back to what you were doing.
- Multiply by the hourly cost. Not the salary — the company's cost for an hour of that person's work.
- Compare with the price of the solution. Investment divided by monthly saving = months to payback.
A model calculation (illustrative, not a client's result): 60 repeated messages a week × 3 minutes = 3 hours a week ≈ 13 hours a month. At a cost rate of €12/h that is roughly €156 a month. A €700 bot with around €20 a month in running costs pays back at that volume in roughly five months.
Run it on your own numbers — if the payback comes out above 12 months, a chatbot is probably not your best investment and the money will work harder elsewhere.
The same procedure for automation in general, calculator included, is in the article How to calculate the payback of automation.
When it is not worth getting a chatbot
When the volume is small. Below roughly 30–50 repeated messages a week, answering personally is both cheaper and better.
When every enquiry needs expert judgement. Design, legal or diagnostic questions. The bot would answer generically and the customer would leave feeling the company does not understand them.
When you have no source material. With no current price list and no written terms, the bot has nothing to answer from. Material first, bot second.
When there is nobody to take an escalation. The bot has to be able to hand a conversation to a person. If nobody is on the other side, the bot only postpones the problem.
When it is about appearances, not usefulness. A chatbot deployed because “everyone has one” is a cost with no return.
What to ask a supplier
Five questions that quickly show whether you are looking at a serious proposal:
- Where will the bot get its answers, and what happens when it has no basis? The right answer: it says so and hands over to a person. The wrong one: “the model will manage”.
- What are the monthly running costs at our volume? If the supplier cannot answer, they have not thought it through.
- Who owns the solution and the data once the engagement ends?
- How is the bot tested before launch? The right answer involves your real questions, not sample sentences.
- What is included in the price after launch? The first weeks of tuning should be part of it, not an extra.
Frequently asked questions
What does an AI chatbot cost a business?
Off-the-shelf SaaS products run on the order of tens of euros a month. A custom chatbot is a one-off investment — with me from €390 for a bot handling frequent questions and from €700 for a bot connected to a shop, calendar or CRM — plus monthly running costs for the model API, by conversation volume.
Is SaaS or a custom bot cheaper?
At low volume and with no integrations, SaaS is almost always cheaper and I recommend starting there. A custom bot pays off when you need a connection to your own systems, control over the data, or when the monthly subscription at growing volume exceeds the one-off investment.
What are the monthly running costs of a chatbot?
With a custom build it is mainly language model API calls, billed by volume of text processed, plus a small amount of hosting if needed. At hundreds of conversations a month that is on the order of single-digit to tens of euros. A precise estimate should be part of the proposal.
From what message volume is a chatbot worth it?
As a rule of thumb, from roughly 30–50 repeated messages a week. Below that, answering personally tends to be cheaper. What decides is not the total number of messages but how many of them repeat.
How long does a chatbot deployment take?
A simple bot for frequent questions is usually ready within a few working days of receiving the material. A bot connected to systems takes longer. On top of that come the first weeks of tuning on real conversations.
Does a chatbot handle less widely spoken languages well?
Current language models handle them well. Differences show up in specialist terminology and in inflected product names, which is why quality is tested on your real questions before launch.
What if the bot invents an answer?
A well-designed bot answers only from the material it has, and when it has no basis it says so and hands the conversation to a person. That is a decisive criterion when choosing a supplier — ask about it directly.
Is a chatbot worth it for a small e-shop?
If a lot of questions come in about availability and order status, yes — especially because a large share of them arrive outside working hours. If the volume is small, probably not. What decides is the number of repeated questions, not the size of the shop.
Sources
- Anthropic — API pricing — model running costs
- OpenAI — API pricing — model running costs
Related services
Related articles
Want to know whether a chatbot pays off for you?
Send me the five questions you get most often and roughly how many messages a week. Free of charge I will work out whether it pays back — and if it does not, I will say so.
Get a free consultation →