Chatbot pricing is one of the most searched-for topics in business AI right now — and one of the most confusing. Quotes range from £0 to £80,000 for what sounds like the same thing. This guide breaks down what you actually get at each price point.
The three tiers of business chatbot
Every business chatbot in 2026 falls into one of three categories, and your budget determines which tier is realistic:
| Tier | What it is | Typical cost |
|---|---|---|
| Platform chatbot | Hosted tool (Intercom, Tidio, Drift) | £50–£800/month |
| Configured LLM chatbot | GPT/Claude wired to your data via prompt engineering or RAG | £3,000–£15,000 build + hosting |
| Custom AI agent | Multi-step agent with tool use, integrations, and feedback loop | £15,000–£80,000+ |
Platform chatbots: £50–£800/month
Tools like Intercom, Tidio, Freshchat, and Crisp give you a chatbot out of the box. You configure rules, add FAQs, connect to your knowledge base, and deploy in a day. They handle the hosting, security, and infrastructure.
What you get: reliable FAQ deflection, lead capture, and handoff to human agents. What you don't get: the ability to answer nuanced questions about your specific products, handle multi-step requests, or take actions in your backend systems.
These platforms make sense if your chatbot goal is "handle the top 20 FAQs and route everything else to a human." They do not make sense if customers ask questions your FAQ doesn't cover, or if you need the bot to actually do things — book appointments, check order status, update CRM records.
Configured LLM chatbots: £3,000–£15,000 build
This tier connects a large language model (GPT-4o, Claude Sonnet, Gemini) to your documentation, product data, and knowledge base. The model can answer nuanced questions in natural language — not just match keywords to FAQ answers.
The build involves:
- Ingesting your content into a vector database (chunking, embedding, indexing)
- Prompt engineering: system prompt, context injection, guardrails
- Retrieval-augmented generation (RAG) pipeline to pull relevant content per query
- Chat UI or API endpoint that integrates with your website or app
- Fallback and escalation logic
Ongoing costs: LLM API fees (£80–£400/month at medium volume) plus hosting (£30–£80/month). Use our chatbot ROI calculator to model the economics.
This tier is right for companies that have a substantial knowledge base (support docs, product catalogue, policy documents) and want the chatbot to answer from it accurately — without the risk of hallucinated answers.
Custom AI agents: £15,000–£80,000+
An AI agent is not just a chatbot — it can take actions. Book a meeting in Calendly. Create a support ticket in Zendesk. Look up order status in your WMS. Update a customer record in Salesforce. Trigger a refund workflow.
Building a reliable agent requires tool definitions, error handling, retry logic, memory (session context plus long-term user context), and safety guardrails. Every integration adds engineering time. The testing burden is also higher — you're not just checking answer quality, you're checking that the agent doesn't take wrong actions.
Large enterprises commissioning complex multi-agent systems can pay well over £100,000 for the initial build. See our full breakdown in AI Chatbots for Business in 2026.
What drives cost within each tier?
Within any tier, these factors push costs up:
- Number of integrations: Each system you connect (CRM, ticketing, booking, WMS) adds 15–40 hours of engineering work.
- Data volume and freshness: A 10,000-page knowledge base that updates daily requires a more sophisticated ingestion pipeline than a 200-page static FAQ.
- Multi-language support: Handling 5 languages well typically adds 20–30% to build cost.
- Compliance requirements: Healthcare (HIPAA), finance (FCA), or GDPR-sensitive use cases require additional architecture and review time.
- Custom UI: A branded embedded widget built to your design system is a separate workstream from the underlying chatbot logic.
Ongoing costs to budget for
Build cost is only part of the picture. Budget separately for:
- LLM API fees: GPT-4o costs roughly £1.25 per 1M input tokens. A medium-complexity conversation uses 1,000–3,000 tokens. At 10,000 conversations/month, expect £12–£38/month in model costs alone.
- Vector database hosting: Pinecone, Weaviate, or pgvector on Postgres — £20–£80/month depending on index size.
- Infrastructure and monitoring: Logs, error tracking, latency monitoring — £30–£100/month.
- Ongoing tuning: Most chatbots need prompt adjustments and knowledge base updates quarterly. Budget 4–8 hours/month of developer time.
Red flags in chatbot quotes
Watch for these signs that a quote will lead to disappointment:
- No mention of RAG or how the bot will handle questions outside the FAQ
- Pricing based purely on number of messages (per-message pricing penalises you for success)
- No line items for testing, QA, or guardrail implementation
- Quotes that exclude ongoing API costs — ask for TCO not just build cost
- No escalation path defined — what does the bot do when it can't answer?
Use our Chatbot ROI Calculator to model payback period before committing to any build, and read our guide on building vs buying a chatbot to decide which approach fits your business stage.