An AI chatbot costs between roughly $5,000 and $80,000 to build, and between a few dollars and several thousand a month to run. The build number depends almost entirely on whether it answers from your own content, and the running number on how many people use it and how much context each answer needs.
Key Takeaways
- Retrieval over your own documents is the feature that separates a $5,000 bot from a $40,000 one.
- Running cost scales with tokens, and context length drives tokens more than message count.
- Budget for evaluation — an assistant nobody measures will confidently give wrong answers.
The three tiers of chatbot
A scripted assistant follows decision trees you define. It never invents an answer, costs $3,000-$8,000, and is genuinely the right choice for narrow tasks like booking or order status.
A retrieval-augmented assistant answers from your own documents. This is what most businesses actually want, and it costs $15,000-$50,000 because the work is in the data pipeline, not the prompt.
A fully custom agent that takes actions in your systems starts around $50,000, because every action needs permissions, auditing and a failure path.
- Scripted flows: $3,000-$8,000
- Retrieval over your content: $15,000-$50,000
- Agent with tool access: $50,000+
What running it actually costs
Model providers charge per token, roughly four characters each, for both input and output. A support answer that retrieves three document chunks might use 2,000 input tokens and 300 output tokens.
At current mid-tier pricing that is a fraction of a cent per answer. Ten thousand conversations a month lands in the tens of dollars, not the thousands — until you add long context windows or an expensive reasoning model to every request.
- Retrieval keeps prompts short, which keeps cost down
- Vector database hosting: $0-$100/month at small scale
- Use a cheaper model for routing and an expensive one only when needed
The cost nobody quotes: evaluation
The difference between a chatbot that helps and one that quietly damages trust is measurement. You need a test set of real questions with known-good answers, run against every prompt or model change.
Teams that skip this find out about hallucinated answers from customers. Budget 10-15% of the build for evaluation and expect it to be ongoing, not a one-off.
How to keep the number down
Narrow the scope. A bot that answers questions about one product line, well, is more useful and dramatically cheaper than one that attempts everything and is unreliable at all of it.
Start with retrieval over your existing documentation before considering fine-tuning. Fine-tuning is expensive, slow to iterate on, and rarely the right first answer to a knowledge problem.
Frequently asked questions
How much does it cost to build an AI chatbot?
A scripted chatbot costs roughly $3,000 to $8,000, a retrieval-augmented assistant that answers from your own documents costs $15,000 to $50,000, and an agent that takes actions in your systems starts around $50,000. The main cost driver is whether it must answer from your content.
What are the monthly running costs of an AI chatbot?
Running costs are driven by token usage. A typical retrieval-based answer uses around 2,000 input and 300 output tokens, costing a fraction of a cent. Ten thousand conversations a month typically costs tens of dollars in model usage, plus hosting for the vector database.
Should I fine-tune a model or use retrieval?
Use retrieval first. Fine-tuning teaches a model style and format, not facts, and it is expensive and slow to update. If the goal is answering questions from your documentation, retrieval-augmented generation is cheaper, more accurate and far easier to keep current.
How do I stop an AI chatbot giving wrong answers?
Ground it in your own content through retrieval, instruct it to say it does not know rather than guess, cite the source document in the answer, and maintain an evaluation set of real questions with known-good answers that you re-run on every change.