Search for what an AI chatbot costs and you will find numbers between a few hundred dollars and six figures, often on the same page. That is not because anyone is being evasive. It is because “chatbot” covers a plugin you install in an afternoon and a system that reads your order database, and those are not the same product.
This is how we price one, what sits at each end of the range, and what the ongoing cost actually is: the part most quotes leave out.
The build cost
A support assistant that answers from your own content sits in our Starter package: $500–$1,500, typically one to three weeks. Payment is split in half, upfront and on delivery, and the first two weeks after launch are covered for fixes and tuning.
What you get at either end of that range is roughly this.
At the lower end. One channel, usually your website. A knowledge base built from the documents you already have — help articles, policies, a product list. Hand-off to your inbox or help desk when the assistant is not confident, with the conversation attached so nobody has to repeat themselves. Answers drawn only from content you have approved, not from the open internet.
At the higher end. The same assistant running on several channels — your website, WhatsApp Business, Facebook Messenger, Instagram — sharing one knowledge base so the answers stay consistent. Live lookups, so it can answer “where is my order” or “is Thursday free” rather than explaining how to check. A dashboard showing which questions were asked, how many were answered, and which topics it could not handle.
What actually moves the price
Four things, in roughly the order they matter.
How many places it has to live. One channel is one integration. Four channels is four, each with its own approval process and its own quirks. This is usually the largest single factor.
Whether it needs to look things up. An assistant that answers from fixed content is a smaller build than one that queries your order system, booking calendar or CRM in real time. The second needs an integration, error handling for when that system is slow or down, and a decision about what it should say when it genuinely does not know.
The state of your content. This one surprises people. If your policies are current and written down, we are mostly organising them. If the real answers live in people’s heads and in two years of support email threads, that has to be gathered and checked before anything can be built on it. We pull your recent conversations, group them by topic and rank them by volume, so the common questions get handled properly first — but someone on your side still has to confirm what the correct answer is.
How much it is allowed to do. Answering is simpler than acting. The moment an assistant can cancel an order, issue a refund or change a booking, you need approval steps, an audit trail, and a clear rule for what happens when it gets something wrong. That is a different scope of work, and it should be.
The cost nobody quotes you
Running an AI assistant is not free after launch, and a quote that does not mention this is incomplete.
Two things cost money every month. AI usage is charged by the provider per conversation, so it scales with how busy you are. A quiet month costs less than a busy one. Hosting is usually small and predictable.
We estimate both up front from your current enquiry volume, before you commit, so the monthly figure is not a surprise in month two. Those accounts stay in your name, so you can see exactly what you are being charged and by whom.
After the first two weeks
The free support period covers fixes and tuning while the assistant meets real customers, which is when you learn what it actually gets asked.
After that, you have a choice. You can take it in-house: you own the code, the prompts and the accounts, and nothing is locked to us. Or you can keep us on a monthly plan — Essential ($300/month), Standard ($700/month) or Priority ($1,500/month), with a three-month minimum — which covers periodic checks, updates where the stack supports them, and small improvements from a list you control.
Neither is the default. If you have someone technical, taking it in-house is a reasonable choice and we will hand over documentation accordingly.
Working out your own number
You do not need a quote to get close. Three questions get you most of the way:
- How many enquiries do you get a month, and what are the top ten questions? If you cannot answer the second part, that is the first piece of work.
- Where do those enquiries arrive? Count the channels. Each one is a cost.
- Does answering any of them require looking something up? If yes, name the system it would have to read.
With those three answers, the range narrows considerably, usually to a few hundred dollars either side of a number, rather than the orders of magnitude you find online.
If it turns out a chatbot is not the right tool for what you are describing, we would rather say so at that stage than after you have paid for one. Plenty of slow-response problems are routing problems, and those are cheaper to fix.
Tell us what you are dealing with and we will give you a fixed quote in writing after a free call. The AI chatbots service page has the full scope, and pricing has every package and plan side by side.