Guide · Global

AI Chatbot Development Cost: What You Pay to Build and Run One

What an AI chatbot really costs: the one-off build, the monthly running costs, and how a custom chatbot compares with per-resolution tools like Intercom Fin and Tidio Lyro, with a worked example using published vendor prices.

"How much does an AI chatbot cost?" has two honest answers. There's the one-off cost of building it, and there's the monthly cost of running it. Most quotes only talk about the first, but for a busy customer service chatbot the second often matters more over two or three years.

This guide explains both in plain language for businesses in the USA, UK, Canada and Australia. It uses the AI vendors' own published prices (checked on 17 November 2026, in US dollars) so you can build a realistic budget and compare a custom chatbot with an off-the-shelf tool. It's general information: your costs depend on your systems, your volumes and the vendor terms you sign.

If you're still deciding whether you need a chatbot that answers questions or an agent that also takes actions, read AI chatbot vs AI agent first. Agents cost more, and we cover them separately in our AI agent development cost guide.

What are the two ways to get an AI chatbot?

Subscribe to a help desk or chatbot tool. Intercom Fin, Tidio Lyro, Zendesk and similar products give you a chatbot that learns from your help centre. You pay a subscription, and often a fee per AI conversation or "resolution". Setup is quick and there's little to build.

Have a custom chatbot built. A developer builds a chatbot on top of an AI model (such as OpenAI's GPT models or Anthropic's Claude), connects it to your content and systems, and hosts it for you or in your own cloud account. You pay for the build, then for hosting and the AI model's usage, which is billed per token (a token is roughly three-quarters of an English word).

Neither is always cheaper. The rest of this guide shows where the money goes in each case.

What does it cost to build a custom AI chatbot?

The build cost is mostly people's time. These are the stages that drive it:

  1. Discovery. Deciding which questions the chatbot should answer, which it must hand to a person, and how success will be measured (for example, the share of conversations resolved without a human).
  2. Content preparation. Collecting and cleaning the FAQs, policies, product data and past tickets the chatbot will answer from. Out-of-date or contradictory content is the most common hidden cost.
  3. Retrieval set-up. Splitting that content into searchable pieces and setting up search so the chatbot finds the right passage before it answers (often called retrieval-augmented generation, or RAG).
  4. Integrations. Each connection adds work: your website, Shopify or WordPress, WhatsApp, a CRM, a booking system or a help desk for hand-over.
  5. Guardrails and hand-over. Rules for topics it must not discuss, what it must never promise, and a clean transfer to a person with the conversation attached.
  6. Testing. Running the chatbot through a set of real past questions and tricky edge cases before launch, and again after every change.
  7. Launch and tuning. Watching the first weeks of real conversations and fixing gaps in the content.

A simple website FAQ chatbot that answers from a few dozen pages needs far less of this than a chatbot that checks order status, books appointments and speaks two languages. When you compare quotes, ask each developer to price these stages separately so you can see what is and isn't included.

What does an AI chatbot cost per month to run?

A custom chatbot has four running costs:

  • AI model usage (tokens). You pay for the text sent to the model (input) and the text it writes (output). Published prices on 17 November 2026: OpenAI's GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens, and its GPT-6.1 Sol model costs $2.00 and $10.00. Anthropic's Claude Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens (for prompts up to 100,000 tokens), and Claude Sonnet 5.5 costs $2 and $10.
  • Hosting and search. A small server or serverless functions, plus a database or vector store for your content.
  • Monitoring. Logging conversations, reviewing failures and alerting when answers go wrong.
  • Upkeep. Updating content when prices or policies change, and re-testing when the AI vendor updates or retires a model.

Prompt caching can cut the input cost further when the same instructions are sent with every message. Anthropic and OpenAI both list cheaper "cached input" rates on their pricing pages.

How do per-resolution tools charge?

Subscription tools usually combine a seat price for your human agents with a fee for each AI conversation:

  • Intercom Fin is priced at $0.99 per outcome. Intercom says you're charged at most once per conversation and not at all if Fin can't answer or the customer asks for a human. Fin Voice, for phone calls, is listed at $1.99 per voice outcome. Seat prices for the Intercom help desk are on top.
  • Tidio lists a free plan and paid plans starting at $24.17 a month, with its Lyro AI agent conversations limited by plan (the first 50 Lyro conversations are free) and higher limits on the Growth and Plus plans.

These tools are excellent value at low and medium volume, especially if you already use the help desk. The per-conversation fee is what grows as you get busier.

A worked example: 5,000 conversations a month

Here's an illustration, not a quote. Assume a store's website chatbot handles 5,000 conversations a month, each with 4 back-and-forth messages. Each message sends about 4,000 tokens to the model (instructions, retrieved help content and the conversation so far) and gets about 300 tokens back.

  • Input: 5,000 × 4 × 4,000 = 80 million tokens a month.
  • Output: 5,000 × 4 × 300 = 6 million tokens a month.
  • Custom chatbot, efficient model: Model or pricing: Claude Haiku 5.5 or GPT-6 Luna (80M × $0.10 + 6M × $0.50 per million); Approximate monthly AI cost: about $11
  • Custom chatbot, larger model: Model or pricing: Claude Sonnet 5.5 or GPT-6.1 Sol (80M × $2 + 6M × $10 per million); Approximate monthly AI cost: about $220
  • Intercom Fin: Model or pricing: $0.99 per outcome, if 60% of conversations are resolved (3,000); Approximate monthly AI cost: about $2,970, plus seats

The custom chatbot's AI bill is far smaller, but it isn't the whole cost: you also pay for the build, hosting, monitoring and someone to maintain it. The subscription includes all of that in the price. That's the real trade-off: a higher running cost in exchange for no build and no upkeep.

As a rule of thumb, the busier your chatbot and the more it needs to connect to your own systems, the more a custom build pays off. At a few hundred conversations a month, a subscription tool is usually the sensible choice.

What makes an AI chatbot cost more?

  • Integrations with order systems, CRMs, booking tools and payment systems.
  • More channels: website, WhatsApp, Facebook Messenger, SMS and phone each add work.
  • More languages, which need separate testing.
  • Regulated topics such as health, finance or legal questions, which need stricter guardrails, privacy reviews and audit logs.
  • Data residency rules. Both OpenAI and Anthropic list higher prices for in-region processing (OpenAI shows +10% for data residency and Anthropic 1.1× for US-only inference).
  • Messy content that has to be rewritten before the chatbot can answer from it.

How can you keep chatbot costs down?

  • Start with the 20 to 30 questions that make up most of your support volume.
  • Use a small, fast model for routine questions, and send only hard ones to a larger model.
  • Keep retrieved content short and relevant. Fewer tokens per message means a smaller bill.
  • Use prompt caching for the instructions you send with every message.
  • Set spending limits with your AI vendor and alerts in your monitoring.
  • Measure resolution rate and hand-over quality, not just the number of conversations.

What should a chatbot quote include?

Before you sign, check that the quote names: the stages above and what each includes; who owns the code, prompts and content index; which AI model is used and who pays for it; hosting costs; how the chatbot is tested before launch; what monitoring you get; and what support costs after launch. A good developer will also tell you when an off-the-shelf tool would serve you better.

If you'd like a second opinion on a quote or want a custom chatbot scoped, our AI chatbot development team works with businesses in the USA, UK, Canada and Australia.

Sources

FAQ

Questions we hear often.

How much does an AI chatbot cost per month?

For a custom chatbot, the AI model fee can be a few dollars to a few hundred dollars a month depending on volume and model, plus hosting, monitoring and maintenance. Subscription tools charge seats plus a fee per AI conversation, such as Intercom Fin's $0.99 per outcome.

Is it cheaper to build a custom AI chatbot or use Intercom or Tidio?

At low volume, a subscription tool is usually cheaper because there's nothing to build. At high volume, or when the chatbot must work with your own systems, a custom build often costs less over two to three years because token fees are low compared with per-resolution fees.

What is the biggest hidden cost of an AI chatbot?

Content. If your FAQs, policies and product data are out of date or contradict each other, someone has to fix them before the chatbot can give reliable answers, and keep them up to date afterwards.

Do AI chatbot prices change?

Yes. AI vendors change model prices and retire models regularly, so check the current pricing pages before you budget and ask your developer how model updates are handled.

Great products start
with a conversation.