Chatbot pricing pages quote one number. Every chatbot has two: what it costs to build and what it costs to run each month. Here is what sits inside each, and what moves them.
Ask ten vendors what a chatbot costs and you will get ten answers, most of them a monthly subscription with the real costs in the small print. The honest answer has two halves. There is what you pay to get a chatbot built, and there is what you pay every month to keep it answering. Small businesses get caught out by the second half far more often than the first.
How much does an AI chatbot cost?
An AI chatbot costs two things: a one-off build and a monthly running bill. The build ranges from an afternoon of setup on a builder tool to a custom project that agency pricing pages we reviewed in October 2026 put at roughly $15,000 to $40,000 for a single-purpose bot. The running cost covers model usage, hosting and channel fees such as WhatsApp, and it grows with the number of conversations.
Which end you land on depends on three questions. Does the bot only answer, or does it also act, such as booking or updating orders? Does it read public web pages, or private documents and systems? And how many conversations will it handle in a month? Get those three answers and the price stops being a mystery.
Chatbot cost to build vs chatbot cost per month
These are separate budgets, paid at different times, and they are moved by different things. Mixing them up is how a cheap quote turns into an expensive year.
| Cost | What it covers | When you pay | What moves it |
|---|---|---|---|
| Build | Design of the conversation, retrieval over your content, integrations, testing, launch | Once, at the start | Channels, integrations, actions, state of your content |
| Model usage | The language model reading each question and writing each answer | Monthly, by usage | Number of conversations, answer length, model choice |
| Hosting | The app, the database and the store your content is searched in | Monthly | Traffic, where it runs, how much content is indexed |
| Channel fees | Charges from the messaging platform, such as Meta for WhatsApp | Monthly, per message | Messages your business sends, their category, recipient country |
| Upkeep | Reviewing conversations, fixing gaps, updating content and prompts | Monthly or as needed | How fast your business changes, how much the bot must get right |
A builder subscription bundles some of these into one fee, usually tiered by conversations or seats. A custom build splits them out, so you see each bill directly. Neither is cheaper in every case. What matters is that you know which lines you are paying for.
The three ways to get a chatbot, and what each costs
Most small businesses choose between three routes. They sit at very different price points because they solve different problems.
| Route | Build cost | Monthly cost | Fits when |
|---|---|---|---|
| Builder tool you set up yourself | Your own time | A subscription, usually tiered by conversations, plus any usage add-ons | The bot answers from your public website and nothing else |
| Builder tool set up for you | A small setup fee or a few days of a freelancer | The same subscription | You want the builder route without learning the tool |
| Custom AI chatbot for business | A project fee, quoted on scope | Model usage, hosting and channel fees, billed to your own accounts | It must read private documents, connect to your systems or follow rules a template cannot hold |
For the custom route, the figures we could verify come from agency pricing pages. One AI development page we reviewed in October 2026 priced a single-purpose agent, such as support triage or lead qualification, at roughly $15,000 to $40,000, and multi-agent systems with custom integrations at $40,000 to $150,000 or more. Another sold a dedicated chatbot developer at around $2,900 for 200 hours a month. Treat these as market signals, not quotes. A website bot over a tidy help centre sits well below the top of those ranges.
What drives the chatbot cost to build
Six factors account for nearly all of the difference between a small chatbot build and a large one.
- Answers only, or actions too. A bot that answers questions is a smaller job than one that books, refunds or edits records, because every action needs permissions, approval steps and testing.
- The state of your content. Clean help articles index in a day. Scanned PDFs, out-of-date price lists and knowledge spread across five tools need cleaning before any bot can use them.
- Integrations. Each CRM, booking system, order database or help desk the bot reads from or writes to adds build and test time.
- Channels. A website widget is one build. WhatsApp, Instagram or your help desk each add a connector, their own rules and their own testing.
- Languages. Every extra language needs its own test questions, and some need different retrieval settings.
- Data rules. Keeping everything inside your own cloud account, or running an open model yourself, takes more setup than calling a managed model API.
Notice what is missing: the chat window itself. The visible widget is the cheapest part of the job. The money goes into making the answers correct, which is why a quote that is mostly design should make you suspicious.
What does a chatbot cost per month?
After launch, you pay for what the bot uses. For a custom AI chatbot that means four lines, each billed by a different provider.
- Model usage. Model providers charge by the amount of text processed, both the question plus the documents the bot reads to answer it, and the reply it writes. More conversations, longer documents and larger models all raise it.
- Hosting. The app that runs the chat, the database of conversations and the vector store your content is searched in. For a small business this is usually a modest line, and it barely moves with traffic until volume gets large.
- Channel fees. Free on your own website. On WhatsApp, Meta charges for certain messages, covered in the next section.
- Upkeep. Someone reviewing the questions the bot could not answer and adding the missing content. You can do this yourself or pay for it as a retainer.
You can estimate model usage before you build. Take the number of conversations you expect in a month, multiply by the average number of messages in each, and multiply again by a typical message length plus the passages the bot pulls from your documents. Run that through your chosen provider's published per-token rates. It is rough arithmetic, but it puts a realistic range on the chatbot cost per month before you commit.
One thing that surprises owners: the size of the documents the bot reads per question often costs more than the answer it writes. Tuning retrieval so it fetches three relevant passages instead of fifteen loosely related ones is one of the easiest ways to cut the bill.
WhatsApp chatbot for business price: what Meta charges on top
A WhatsApp chatbot has the same build and running costs as a website bot, plus one more: Meta's fees on the WhatsApp Business Platform. These are set by Meta, not by whoever builds your bot, and they change from time to time.
Meta charges per delivered message, only for messages your business sends, at rates set by message category and by the country of the person receiving them. Template messages your business starts, such as reminders, order updates or marketing, are charged. Since 1 October 2026, so are ordinary replies inside the 24-hour customer service window: each business phone number gets 1,000 of these service messages free per month, and Meta charges from the 1,001st. Chats that start from a click-to-WhatsApp ad open a free window instead. The exact rules and rates are on Meta's WhatsApp Business Platform pricing page, checked in October 2026, and they are worth reading before you plan any outbound campaign.
If you connect through a business solution provider rather than directly through Meta, that provider may add its own platform fee. Ask for it in writing. And avoid unofficial WhatsApp tools that automate a personal number. They are cheap until the number gets banned.
Custom AI chatbot for business or a builder tool?
A builder tool is the right answer more often than agencies like to admit. If your bot only needs to answer from public pages on your site, a builder will get you live this week for the price of a subscription.
Custom work earns its cost when the bot needs something a builder cannot give it:
- Private documents, such as policies, contracts or internal guides, with control over who can see which answers.
- Live data from your own systems, such as order status, stock levels or appointment slots.
- Actions, such as booking, refunding or creating tickets, with a person approving the risky ones.
- Data that must stay in a particular country or inside your own cloud account.
- Volume high enough that paying model providers directly costs less than a per-conversation subscription.
That last point is easy to miss. Builder subscriptions are priced for convenience. At low volume they are good value. At high volume, a custom AI chatbot for business that bills model usage to your own account can cost less each month, and you own the code instead of renting it.
RAG chatbot cost: does answering from your own documents cost more?
A RAG chatbot searches your own content before it answers, which is what stops it inventing things about your business. It adds work to the build: your documents have to be cleaned, split into passages, indexed and tested against real questions. It also adds a small hosting line for the store those passages live in.
It is worth it for almost any bot that talks to customers. A bot answering from a model's general memory is cheaper to build and expensive the first time it quotes a refund policy you do not have. Most of the RAG build cost is a one-off, and the monthly addition is small next to model usage. It is why our custom AI chatbot development work starts with retrieval over your content, before any chat window exists.
Hidden costs that show up after launch
These are the lines that never appear on a pricing page and always appear in year one.
- Content upkeep. Prices change, policies change, products launch. If nobody updates the source documents, the bot keeps quoting the old ones.
- Reviewing conversations. The questions your bot could not answer are the most useful data you have. Someone needs to read them every week or two.
- Model changes. Providers retire older models and release new ones. Each switch needs a short round of testing so answers do not drift.
- Staff handover time. A good bot passes hard conversations to a person. Those conversations still need answering, by someone, during the hours you promised.
- Vendor price changes. A subscription that fits at launch can be re-tiered a year later. Owning the accounts protects you from this more than any contract.
How to keep chatbot running costs down
- Use a smaller, cheaper model for simple routing and greetings, and a larger one only for the questions that need it.
- Limit how much text the bot retrieves per question. Fewer, better passages cost less and usually answer better.
- Cap answer length. Customers want short replies on a phone screen anyway.
- Cache answers to the questions you get fifty times a day, so the model is not paid to write them fresh each time.
- Set clear handover rules, so the bot stops trying when it is out of its depth instead of burning through long, unhelpful exchanges.
Is an AI chatbot worth the cost for a small business?
Work it out from your own inbox, not from a vendor's case study. Count how many messages a week are the same handful of questions: opening hours, prices, delivery times, how to book. Time how long your team spends on them. If that is several hours a week, a chatbot that handles even half of them pays its running cost quickly.
There is a second return that is harder to count: the enquiries that arrive at night and on weekends and currently wait until morning, or go to whoever answered first. A bot that replies instantly, captures the details and books the call keeps those leads. If your business gets few messages and every one needs a careful human reply, a chatbot is the wrong spend, and a faster inbox routine will do more. Chatbots also sit within a wider picture of AI automation for small businesses; our guide on how to automate your business covers how to decide what to automate first.
Want a fixed quote for your chatbot build?
Send us the questions your customers ask most and where the answers live today. We will tell you whether a builder tool is enough, and if it is not, quote the build of your custom AI chatbot as a fixed price, with the monthly running costs estimated from your real volume. You can start a project here, and we reply within one business day.
FAQ
Frequently asked questions
No. A ChatGPT subscription is for a person using the chat app. A business chatbot calls a model through the provider's API, which is billed separately by usage, based on how much text goes in and comes out. You can use models from OpenAI, Anthropic, Google or open models you host yourself. A well-built bot can switch between them, so you are not tied to one provider's pricing if it changes.
Not entirely. The WhatsApp Business app on a phone is free and offers simple greeting and away messages. A custom AI chatbot connected to your own content and systems runs on the WhatsApp Business Platform instead, where Meta charges for delivered messages your business sends, at rates that depend on the message type and the recipient's country. Since October 2026 that includes replies to customers after the first 1,000 each month per number. A provider may add its own fee. On top of that you pay the usual build, model and hosting costs.
A focused website or WhatsApp chatbot over content you already have is a matter of weeks. Most of that time goes into collecting real customer questions, cleaning documents and testing answers, not into the chat window. A bot that takes actions in your CRM or booking system takes longer, because every action needs permissions and testing. Bigger builds are best split into stages, so a useful version goes live early.
You can get started for free. Many builder tools have free tiers, and you can set up a basic bot over your website in an afternoon. Free tiers usually limit conversations, features or branding, and the bot will only know what you feed it. For a handful of common questions on a small site, that can be enough. Once you need private documents, integrations or higher volume, you will be on a paid plan or a custom build.
It depends on how much you do yourself. The essential work is reviewing unanswered questions, updating the documents the bot reads, and retesting after model or content changes. A business with stable information might spend an hour or two a month on it. One that changes prices and products often will spend more. Many businesses pay for this as an optional monthly retainer, so someone is responsible for answer quality after launch.
A chatbot answers questions and collects details. An AI agent takes actions in other systems, such as booking appointments, updating orders or raising refunds. Agents cost more to build because each action needs permissions, error handling, approval steps for anything risky, and much more testing. A sensible path is to launch a chatbot first, prove it answers well, then give it actions one at a time as the business case for each becomes clear.
That depends on the contract, so check before you sign. You should own the code, the conversation data, and the accounts for the model provider, hosting and messaging platform, all set up in your name. If an agency keeps the accounts, your monthly bill passes through them and leaving becomes hard. With custom builds, ownership is one of the main reasons to pay more than a subscription, so make sure you actually get it.
Written by
Kavion Solutions Team
The senior team behind Kavion Solutions. We build websites, apps, AI tools and cloud setups, and run the SEO and ads that bring in customers, so everything here comes from work we have actually shipped.
Want this done for you?
Start a project