How much does an AI chatbot cost for a small business: platform subscriptions, per-message fees and what a one-off build includes
An AI chatbot for a small business is paid for in one of three ways: a monthly subscription to an AI chatbot platform, a fee for every conversation the bot handles, or a one-off build that is yours once it is finished. A subscription is the easiest to budget, a per-conversation fee is the easiest to start with, and a one-off build is the easiest to live with once it is running. The price page never shows the whole bill: WhatsApp message fees, the AI model behind the answers and the upkeep of those answers are separate lines. We build the third kind, so weigh our numbers with that in mind.
Three ways to pay for an AI chatbot
| Model | You pay for | Where the bill moves |
|---|---|---|
| Platform subscription | Access to the builder by the month, with an allowance of AI conversations | Up when you outgrow the allowance or need a higher plan |
| Fee per conversation or outcome | Each conversation the bot handles, or each one it resolves | Up and down with the number of customers who write |
| One-off build | The work of setting up, loading your documents and testing | Stays flat; changes only if you ask for more work |
A subscription looks like this at Tidio: the Lyro AI Agent add-on starts at $32.50 a month, on a tier that begins at 50 AI conversations, and the base plans are priced separately. The page has a switch between monthly and annual billing, so check which figure you are reading. It also says the first 50 AI conversations on the Free plan are free for life, which is enough to test an idea against your own questions.
Pricing per outcome is what Intercom does with its Fin AI agent: $0.99 per outcome, with separate plans priced per seat per month on the same page. Read Intercom's definition of an outcome before you compare it with anything else.
A one-off build is what we sell. Our AI assistant costs from €480, with no monthly fee, a launch in one to two weeks, and the second half of the payment due after you have accepted the result. What the price includes is further down.
A worked example with made-up volumes
Take a small clinic with an invented but ordinary pattern: 80 conversations handled by the assistant in a quiet month, 300 in a busy one after a promotion. This is an example, not data about any real business.
If each conversation counts as one outcome, per-outcome pricing makes the busy month cost 300 times the unit price and the quiet month 80 times, nearly four times as much. Under a subscription the line is the same in both months, until 300 conversations break through the allowance and you are asked to move up a tier. Under our one-off build there is no monthly line from us in either month.
The WhatsApp layer sits on top of whichever you pick. Say the clinic's customers write on WhatsApp and the assistant sends four replies per conversation: 320 replies in the quiet month, 1,200 in the busy one. Meta's pricing page says it does not charge for messages from users to businesses. From 1 October 2026 it also gives each business phone number 1,000 free service messages a month and charges from the 1,001st; before that date service messages were free. The rate depends on the message category and the market. If all four replies count as service messages (check the category on Meta's page), the busy month runs 200 over the free allowance and the quiet month stays inside it.
Whether that fee is inside the vendor's price or added on top is a question for the vendor. What an assistant can do on the channel itself is in our piece on WhatsApp chatbots for a small business.
The costs that are not on the price page
Behind each answer sits a language model that bills for the text it processes. Google's Gemini API pricing page, for one, says billing is metered per token, with input and output priced separately. The same page treats the content you send differently by tier: on the free tier it can be used to improve Google's products, on the paid tier it is not. That difference deserves a question to any vendor handling your customers' messages: which account does the model run on, under what terms, and who pays for it?
Prices change, opening hours change, a service gets dropped, and somebody has to change the assistant's answers. Ask whether that is a request to the vendor, a form you fill in or a file you edit, and whether it is billed. On our side, bug fixes are free for the first 6 months after launch and maintenance after that is optional.
Anything on a separate line has its own meter. Voice is the clearest case on our own page: the voice assistant is an add-on, and call minutes are billed separately from it.
Tax is a separate question. Our prices exclude VAT. Check whether the quote in front of you includes it.
When to build, and when not to
If you want to try the idea this week, build nothing: use a free plan and your own twenty questions. If someone at your business likes configuring tools, building an AI chatbot yourself on a platform is a fair choice, and our FAQ says as much: you can, if you have time for the setup and testing. If all you need is a menu of five buttons answering five questions, a free plan is enough and we are the wrong call.
A one-off build starts to make sense when conversations are steady or growing, when you do not want a bill that follows your busiest month, and when nobody on the team has the hours to load documents, test the answers and set up the handover to a person.
What a build looks like: Kvant Soft
Our Kvant Soft work is an assistant for a warehouse software company in Tallinn, and the site is in English. It answers from the product documentation, quotes the passage and names the section, and a click opens that section in the docs. When the documentation does not cover the question, it says so and turns the conversation into a ticket with the chat attached. A visitor can also pick a demo slot in the calendar without leaving the chat.
Showing the source under every answer is one of the add-ons on our page, not part of the starting price. What the starting price does cover is the setup around the model: loading the documents, deciding when to say "I don't know" and when to hand over to a person, and testing on your own questions. That setup is what a one-off price pays for. More examples are in our portfolio.
What to do today
- Count your conversations. Open your WhatsApp Business app and your website inbox and note how many chats you had last month and in your busiest month. Without that number no quote can be compared.
- Write down the twenty questions customers ask most. They become the assistant's materials whoever builds it: price list, rules, notes.
- Look up Meta's rate for your customers' country on the pricing page in the sources, and work out how many replies you would send past the 1,000 free service messages.
- Put the same five questions to every vendor. Which account runs the model, and who pays for it? What does the bill look like in my busiest month? Who updates the answers when my prices change, and is that billed? What can I take with me if I leave? Which services process the conversations?
- Try a free plan with your own questions before you pay for anything. Tidio's Free plan, for example, includes a lifetime allowance of 50 AI conversations.
If you want us to build it for you
What we build is an AI assistant for answering customers and taking bookings, not a button-menu chatbot: it understands questions in the customer's own words and answers from your price list, rules and notes, in one channel of your choice: a chat on your website, Telegram or WhatsApp. It costs from €480, with no monthly fee, and the price and timeline are fixed in writing before we start.
In that price: a call before we start; your materials loaded, and the greeting and prompts written by us for you to review; an honest "I don't know" when the answer is not in the materials; a handover to you or your receptionist when a customer types "human"; a label showing that AI is answering; a conversation log; booking requests that reach you as a card on WhatsApp; a check with your own questions before launch; free revisions until acceptance and free bug fixes for 6 months after launch. Prices exclude VAT.
The price is low for one reason: our experienced specialists work faster with AI tools. We reply within an hour during working hours. Tell us what your customers ask and we will come back with a timeline and a price; the first review takes 15 minutes and is free.
Sources
- WhatsApp Business Platform pricing, Meta for Developers. Checked 1 November 2026.
- Intercom pricing, Intercom. Checked 1 November 2026.
- Tidio pricing, Tidio. Checked 1 November 2026.
- Gemini API pricing, Google AI for Developers. Checked 1 November 2026.