How much does an AI chatbot cost — and what drives the number?

The four things that move the price

1. What it has to know

A bot answering twenty fixed questions is a different project from one that answers from four hundred pages of product documentation. The first is scripted; the second needs ingestion, retrieval and evaluation. This single factor accounts for most of the spread between a $199 build and a $999 one.

2. How many channels

A website widget is the cheapest surface. Adding official WhatsApp means business verification, template approval and a messaging provider. Adding Slack, Instagram or in-app means more of the same. Each channel is incremental, not multiplicative — the knowledge base is shared — but none is free.

3. What it has to do, not just say

Answering is cheap. Booking a slot, creating a CRM record, checking live stock, raising a ticket or taking a payment means integrating with systems that each have their own authentication, rate limits and failure modes. Integration depth is usually the second-largest driver.

4. How much accuracy must be proven

For a marketing site, careful testing before launch is enough. In finance, healthcare or anywhere a wrong answer carries liability, you need an evaluation set, human review workflows and audit logging. That is engineering work, and it is the right work to pay for in those sectors.

Build cost versus running cost

These are separate numbers and it is worth keeping them separate when comparing quotes.

The right-hand column is where most projects land, not a ceiling. A build with several channels, deep integrations, multiple languages or a compliance requirement is quoted above it — which is a conversation to have on a call rather than a number to guess from a page.

Running cost surprises people in both directions. Model usage for a chatbot handling a few thousand conversations a month is usually smaller than the hosting bill. It becomes material only at high volume or with very long context — a system reading fifty pages per answer costs meaningfully more than one reading two.

What a fair quote includes

• Content ingestion and cleaning, not just the bot itself.
• A tested set of real customer questions, agreed before launch.
• Human handoff with the full transcript attached.
• Conversation logging and analytics from day one.
• A month of supervised operation, with transcript review and tuning.
• Handover: the code, the prompts and the knowledge base are yours.

When it pays back

The arithmetic is usually simple. Count the enquiries arriving outside working hours, multiply by the share that never gets a reply, and compare with your average deal value. For most businesses with a steady flow of repeat questions, a chatbot pays for itself within two to four months — and if the numbers do not work, that is worth finding out on a discovery call rather than after a build.