AI chatbot cost per month for a small business

Most small businesses pay for a website AI chatbot in two parts: a fixed amount for the platform or the build, and a variable amount for usage that rises with the number and length of conversations. On top of that sit one-off setup work to prepare your own content, and a modest but real monthly cost in someone’s time to keep it accurate. Quotes differ so much because they bundle these layers differently, and some leave the costly ones out.

Below: each cost layer, a simple way to estimate your own usage, the realistic options side by side, the risks to plan for, and when a chatbot is not worth paying for at all.

Why chatbot quotes differ so much

Two quotes for “a chatbot on our website” can sound identical and still be far apart, usually because they cover different layers of the work. When you compare quotes, check whether each one includes:

• The platform or build: the software the chatbot runs on, either a subscription you rent or something built and connected for you.
• Usage: the language-model processing behind each reply, which may be included up to a limit, billed separately, or passed through at cost.
• Setup: gathering and organising your services, prices, policies and FAQs so the chatbot answers from your information rather than general knowledge.
• Handover and lead capture: what happens when a visitor needs a person, and where their details go.
• Maintenance: keeping its knowledge current and reading transcripts to catch mistakes.

A low monthly figure that leaves out setup and maintenance is not cheaper; it moves the work to you. A higher quote may simply be the first one that admits all five layers exist.

Layer one: the platform or the build

An off-the-shelf subscription

Many chatbot products are sold as a monthly plan. You add a small piece of code to your site, upload your content or point the tool at your pages, and adjust the settings in a dashboard. Plans are commonly tiered by the number of conversations or messages, the number of staff seats, and features such as integrations or removing the vendor’s branding.

Read the tier limits closely: what matters is what happens when you exceed them, whether a hard stop, an automatic upgrade or an overage charge. Off-the-shelf suits fairly standard questions. You are renting it, so if you leave, the setup usually stays behind.

A custom integration

A custom chatbot is built on a language-model provider’s API and connected to your own website and systems. You pay for the build once, then for usage directly from the provider, plus hosting and upkeep. The extra cost buys control: which sources it may answer from, how it hands over to staff, and what it can do, such as checking a booking calendar or creating a lead in your CRM.

The trade-off is that someone has to build and look after it. If the main reason for going custom is a connection to another system, that is integration work in its own right; our page on API and CRM integrations explains what it involves.

Layer two: usage, and how tokens work

Behind most website chatbots is a large language model, usually billed by the volume of text processed, measured in tokens. A token is a chunk of text, often part of a word. What goes into the model and what comes out are both counted, often at different rates.

The part people miss is how much goes in with each reply. To answer one visitor question, the chatbot typically sends the model:

• its standing instructions: what it is, and what it may and may not say;
• the passages from your content that it has looked up for this question;
• the conversation so far, so it can follow the thread;
• the visitor’s new message.

So a late message in a long conversation can cost noticeably more than the first, because much of what came before is sent again. Some providers reduce the cost of repeated content, and a well-built chatbot keeps its instructions and passages tight, but the principle holds: cost scales with how many conversations you have and how long they run.

Estimating it: an illustrative method

You can reach a sensible estimate before you buy anything. The method is the same for any provider:

1. Conversations per month (C): count the enquiries you get now by form, phone and email, and assume a chatbot will see at least that many.
2. Exchanges per conversation (E): one visitor message and one reply count as one exchange.
3. Tokens per exchange (T): everything sent in plus the reply that comes out, including instructions, retrieved content and history.
4. Monthly volume = C × E × T.
5. Monthly usage cost = monthly volume × your provider’s published rate, split into input and output if they price them separately.

For illustration only, with made-up round numbers: if a chatbot handled 300 conversations a month, at 6 exchanges each, with 2,000 tokens processed per exchange, that is 300 × 6 = 1,800 exchanges, and 1,800 × 2,000 = 3,600,000 tokens a month. If your provider’s published rate is R per million tokens, usage for the month is 3.6 × R. These inputs are not benchmarks; replace every one with your own figures and your provider’s current rate.

Then run the sum again at double your expected volume, so a busy month does not surprise you, and ask for a monthly spending cap or alert, so a fault or a bot hammering the chat window cannot run up a bill unnoticed.

Layers three to five: setup, handover and upkeep

Preparing your content

This layer decides whether the chatbot is useful, and it is the one most often under-quoted. A chatbot can only answer from your information if it exists in a clean, current form. Expect to gather and check:

• your services, and what each one includes and excludes;
• your service area, hours and how to reach you;
• your policies on cancellations, refunds, deposits and warranties;
• the questions customers actually ask, with your real answers;
• the topics the chatbot must never answer and should pass to a person instead.

This often shows the website itself is missing answers. Fixing those pages helps every visitor.

Human handover and lead capture

A chatbot that cannot pass a visitor to a person ends conversations at exactly the wrong moment. Decide in advance:

• When it hands over: a request for a quote, a complaint, anything urgent, or simply whenever the visitor asks.
• How it hands over: live chat to staff during opening hours, a click-to-WhatsApp or call button, or a short form outside hours.
• What it collects: usually a name, a contact method and a one-line summary of the need, and nothing more.
• Where the lead goes: your inbox, a shared mailbox or your CRM, with a named person responsible for replying.

This is often where a chatbot earns its keep: an enquiry at nine in the evening that is captured properly and answered the next morning.

Maintenance

Budget time every month, even when the software cost is fixed. Someone should update its content whenever services, hours or policies change, and read a sample of transcripts for wrong answers, missed handovers and unanswered questions. Each unanswered question is either a gap in your content or something the chatbot should not be handling.

Off-the-shelf, custom or no chatbot

“No chatbot” is a genuine option here: for many small businesses, a clear FAQ page and a click-to-WhatsApp button does the job.

Risks, and the safeguards that cost little

Language models are fluent, which makes their mistakes persuasive. Plan for these risks before launch:

• Confident wrong answers. A model can state something false in a perfectly reasonable tone. Restrict it to your own content, instruct it to say it does not know when that content does not cover the question, and test it with awkward questions before launch.
• Answers beyond your business. Without limits it may offer general legal, medical or technical advice you would never give. Keep it to your services and have it hand off anything else.
• Quoting prices or making promises. A price, completion date or guarantee given in chat creates an expectation you may have to honour or argue about. Do not let it quote prices or commit to anything; have it explain what affects the price and offer a person to confirm.
• Privacy. Visitors type things into a chat box they would never put in a form. Ask them not to share sensitive information such as card numbers, health details or identity documents, collect only what you need to reply, check where transcripts are stored and for how long, and mention the chatbot in your privacy policy.
• Not saying it is an AI. Visitors should know when they are talking to software. Label it clearly in the chat window and make reaching a person easy.

None of these safeguards costs much. Leaving them out can cost more than the chatbot.

When a chatbot is not worth it

A chatbot is not the right spend for every business. It is usually the wrong choice when:

• Your enquiry volume is low. If you get a handful of enquiries a week and answer them promptly, a chatbot adds cost and upkeep without saving much time.
• Your questions are few and predictable. A clear FAQ page answers them faster, costs nothing per visit and can be read by search engines too.
• Every enquiry needs a person. If each job needs a site visit or a conversation about scope, the chatbot would mostly hand over, which a WhatsApp or call button does directly.
• Your website does not answer the basics yet. Fix missing service areas, hours and pricing approach first. A chatbot built on thin content gives thin answers.

If you are still working out the cost of the website itself, our guide to what a small business website costs sets out typical market ranges for orientation. At Tecnomation, AI chat is one of the specialist services quoted individually, in writing, before work starts, and it never changes the monthly website price described on our pricing page.

If you are weighing up a chatbot, start with the content: write down the questions your customers ask and your real answers, then decide whether a chatbot, an FAQ page or both is the right home for them. Our AI chat and automation service covers building one that answers only from your information, hands over to a person and captures leads properly, and you can talk to us about whether it is worth it for your business before anything is quoted.