What does an AI chatbot cost a small business?
Real numbers for building and running an AI chatbot in 2026 — setup costs, monthly usage, the hidden expenses nobody quotes, and when it is genuinely not worth it.
Every quote you get for an AI chatbot will be a different number, and none of them will explain why. Here is the actual cost structure, including the parts that usually get left out of the proposal.
The three costs, separated
Most confusion comes from bundling three genuinely different things into one figure.
Build cost is one-off. Somebody connects the bot to your content, decides what it is allowed to say, tests it, and puts it on your site. This is where nearly all the money goes.
Usage cost is monthly and depends on traffic. Every conversation sends text to a language model and gets text back, and you pay per unit of text. For a typical small-business website this is genuinely small — often single-digit dollars a month, sometimes low double digits.
Maintenance is the one nobody quotes. Your policies change, your products change, and a bot answering from last year’s return policy is worse than no bot.
What the build actually costs
A subscription widget you configure yourself runs somewhere in the tens of dollars a month with no build cost. These match questions against an FAQ list you type in. If your questions are genuinely simple and few, this is the correct choice and you should not pay anyone to build anything.
A custom bot grounded in your real documents typically runs a few thousand dollars to build. That money buys the boring parts: getting your policies into a form the bot can retrieve from, deciding what happens when it does not know, and testing it hard enough to trust.
The gap between those two options is not really about capability. It is about whether the bot is wired to your business or just wired to a text box.
The cost everyone forgets
The expensive failure is not the bill. It is the bot confidently telling a customer something untrue.
If your bot invents a return window, you either honor a policy you did not agree to or you argue with a customer holding a screenshot. Either outcome costs more than the entire build, and it costs it in trust rather than dollars, which is harder to get back.
This is why grounding matters more than model choice. A cheap model that only answers from your documents and says “I don’t know” otherwise beats an expensive model free-associating about your business.
When it is not worth it
Be honest about volume. If you get five inquiries a week, a chatbot is a toy. The time you spend maintaining it exceeds the time it saves, and you should answer those five emails yourself.
Be honest about variety, too. If every inquiry is genuinely different — bespoke quotes, complex diagnostics — there is no repeated question to deflect. A chatbot helps when the same fifteen questions arrive over and over. That is the whole mechanism.
And if your policies are not written down anywhere, fix that first. It is the actual bottleneck, it costs nothing, and you will need it regardless of whether you ever build a bot.
A reasonable way to decide
Count how many inquiries you get a month and how many are one of your top ten repeated questions. If that second number is under a hundred, start with a good FAQ page. If it is in the hundreds, a custom bot will pay for itself inside a year. If it is in the thousands, it already has.
Working out whether a chatbot makes sense for your situation? Book a free 30-minute call — I will run these numbers with you, including the part where the answer is no. Or read more about how I build chatbots.