AI Chatbot Development Cost in 2026: Build vs Buy, Tiers and Running Costs

How much does an AI chatbot cost?

Build or buy?

Helpdesk vendors now include AI agents. Intercom, for example, lists its Fin AI agent from $0.99 per resolved conversation on top of seat plans from $29 to $132 per seat a month. For a standard support use case on a platform you already use, that can be the fastest option.

The two also combine well: a custom assistant inside your product for account-specific tasks, with a handoff into the helpdesk your team already uses.

What drives chatbot cost

• Knowledge sources: one help centre is simple; several systems with permissions need more retrieval and access-control work.
• Actions: every order lookup, booking or account change needs authentication, an integration and careful testing.
• Channels: each of web, WhatsApp, SMS and voice adds its own interface, limits and testing.
• Languages: current models handle most languages, but each language needs its own evaluation questions.
• Quality bar: an evaluation set and guardrails add time up front and save far more after launch.

Running costs

Model usage is billed per token by providers such as OpenAI, Anthropic and Google, and differs widely between models. A support bot handling a few thousand conversations a month usually costs tens to low hundreds of dollars in model fees when simple questions go to smaller models and responses are cached. Add hosting, vector storage for your content and a retainer for monitoring and content updates. We estimate the monthly model bill during discovery from your real conversation volumes.

What a good chatbot budget includes

1. Discovery with 100 or more real questions from tickets and chats.
2. Retrieval over approved content with citations, refreshed automatically.
3. An evaluation set run on every change, measuring accuracy and groundedness.
4. Clear handoff rules to a person, with the full conversation passed along.
5. Guardrails against prompt injection, off-topic use and personal data leaks, tested before launch.
6. Analytics on containment, handoff reasons, unanswered questions and cost per conversation.

A typical chatbot project timeline

1. Week 1: collect real questions, agree what the bot may and may not do, choose channels.
2. Weeks 2 to 3: connect content sources and measure whether the right passages come back.
3. Weeks 3 to 5: conversation design, tone, handoff rules, guardrails and any account actions.
4. Weeks 5 to 8: pilot on a share of traffic, review transcripts daily and tune.
5. After launch: monthly reviews of unanswered questions, cost per conversation and satisfaction.

How to measure the return

The simplest model multiplies three numbers: conversations per month, the share the bot resolves without a person (its containment rate), and what a human-handled ticket costs you today. Subtract the monthly model, hosting and retainer costs, and compare the result with the build cost to get a payback period. Track containment and customer satisfaction together; a bot that deflects questions without answering them improves the first number while hurting the second.

How to keep the cost down

1. Start with the top 20 question types, which usually cover most of the volume.
2. Launch on one channel and add others once the bot performs well.
3. Route by difficulty: small models for classification and simple answers, larger ones only when needed.
4. Fix content gaps the bot reveals; better help articles improve answers for free.

See AI chatbot development, RAG development and the ChatGPT-style app cost guide for full AI products.

Sources

• Innovation Insight AI chatbot development and pricing
• Intercom pricing: https://www.intercom.com/pricing
• OpenAI API pricing: https://openai.com/api/pricing/
• Anthropic pricing: https://www.anthropic.com/pricing
• OWASP Top 10 for LLM applications: https://genai.owasp.org/llm-top-10/