Quick reference: AI chatbot cost by approach
| Approach | Build cost | Ongoing cost/month | Timeline |
|---|---|---|---|
| No-code platform (Intercom Fin, Tidio, Freshdesk) | £2,000 – £8,000 | £200 – £2,000 | 2–6 weeks |
| Custom LLM chatbot (existing product) | £8,000 – £25,000 | £100 – £800 | 4–10 weeks |
| Chatbot with deep system integrations | £20,000 – £50,000 | £200 – £1,500 | 8–16 weeks |
| Internal AI assistant (staff tool) | £15,000 – £40,000 | £100 – £600 | 6–12 weeks |
| Autonomous AI agent (multi-step actions) | £40,000 – £80,000+ | £300 – £2,000 | 12–20 weeks |
Build cost covers scoping, design, development, testing, and deployment. Ongoing cost is LLM API usage plus hosting; platform-based chatbots substitute this with a SaaS subscription. Neither is typically included in initial quotes.
No-code platform vs. custom build: what actually differs
The most important decision is whether to configure a platform or build custom. Both use LLMs under the hood — the difference is who controls the architecture.
No-code platforms (Intercom Fin, Tidio, Freshdesk, Drift)
You’re buying a product and configuring it. The AI model is provided by the platform, your knowledge base is uploaded via a CMS, and integrations are limited to those the platform supports. Setup is fast — 2–6 weeks for a reasonably complex configuration. The tradeoffs:
- Monthly fees continue indefinitely, typically £200–£2,000/month depending on volume and tier
- Limited control over how the AI reasons or responds
- Integration depth depends entirely on the platform’s connector library
- You can’t change the underlying LLM or fine-tune behaviour beyond what the platform exposes
- Your conversation data sits in a third-party system
No-code is the right choice when your use case is standard (customer support Q&A, lead qualification, FAQ deflection) and you need to be live in weeks rather than months.
Custom LLM chatbot
You own the code, the architecture, and the data. You choose the LLM (GPT-4o, Claude, Llama 3), design the conversation logic, and integrate directly with your own systems. The build cost is higher upfront but the ongoing cost is just API usage and hosting — typically a fraction of platform subscription fees at scale.
Custom build is the right choice when: your integration requirements go beyond what platforms support, you have data privacy requirements that prevent sending conversations to a third-party platform, you need the chatbot to take actions in your systems rather than just answer questions, or your volume is high enough that per-conversation SaaS pricing becomes expensive.
Total cost of ownership matters more than build cost. A no-code platform at £800/month costs £19,200 over two years before any setup fee. A custom chatbot at £18,000 to build with £200/month in API costs totals £22,800 over the same period — and you own it outright at the end. Run the maths for your expected volume before assuming the platform is cheaper.
Cost by chatbot type
Customer service chatbot (£8,000–£30,000)
Handles inbound queries: product questions, order status, returns, troubleshooting. The cost range is driven by integration depth. A chatbot that answers questions from a static knowledge base sits at the low end. A chatbot that queries your order management system, retrieves live account data, and initiates refund workflows in real time sits at the top.
Key engineering components:
- Knowledge base ingestion and retrieval (RAG pipeline for large corpora)
- Intent classification and escalation logic
- Live system queries (order status, account lookup)
- Human handoff with full conversation context transfer
- CSAT collection and conversation logging
Internal staff assistant (£15,000–£40,000)
Helps employees answer questions from internal documentation, policies, runbooks, or institutional knowledge. These are often the highest-ROI chatbots because the audience is small (your team), the knowledge base is controlled, and the quality bar for responses is high — staff will notice wrong answers immediately.
Common integrations: SharePoint, Confluence, Notion, internal databases, HR systems. The RAG pipeline for ingesting and querying unstructured internal documents is usually the most complex part of this project type.
Sales qualification chatbot (£10,000–£25,000)
Qualifies inbound leads before human sales involvement. Asks discovery questions, collects contact details, assesses budget and timeline, and either books a meeting directly or routes to a sales rep. These require tight CRM integration (HubSpot, Salesforce, Pipedrive) and careful conversation design — a chatbot that feels like a rigid form loses leads.
Autonomous AI agent (£40,000–£80,000+)
Goes beyond answering questions to taking multi-step actions: booking appointments, processing orders, updating records, sending emails, triggering workflows. This is the most expensive category because the engineering requirement is qualitatively different — you need robust error handling, audit logging, safety guardrails, human oversight hooks, and extensive testing for edge cases where the agent could take a wrong action.
What drives chatbot development cost
1. Integration depth
A chatbot that answers questions from a PDF is a very different project to one that queries your live inventory, creates support tickets in Zendesk, looks up customer order history in your database, and sends confirmation emails. Every integration adds engineering time for API design, authentication, error handling, and testing. A single deep integration (CRM, ERP, custom internal API) typically adds £5,000–£15,000 to the project.
2. Knowledge base complexity
Small, well-structured knowledge bases (under 50 documents, consistent format) can be embedded directly into the system prompt or a lightweight vector store. Large, unstructured, or frequently-changing corpora need a proper RAG pipeline with embedding generation, vector indexing, retrieval tuning, and a freshness pipeline. This adds £8,000–£20,000 to the project compared to a simple prompt-based approach.
3. Conversation complexity
A chatbot that answers single-turn questions is simpler than one that maintains context across a multi-turn conversation, remembers user preferences, or picks up mid-conversation where a previous session ended. Stateful conversation management — storing and retrieving context correctly — is one of the trickier engineering problems in chatbot development and is frequently underestimated in quotes.
4. Channel requirements
A web widget is the cheapest deployment channel. Adding WhatsApp Business, Slack, Microsoft Teams, SMS, or voice (via Twilio or similar) each adds integration overhead. Multi-channel deployment with consistent behaviour across all channels is significantly more complex than single-channel.
5. Evaluation and safety
Any chatbot talking to customers needs an evaluation framework: automated tests that check response quality, a way to monitor for hallucinations or off-topic responses, and a mechanism to catch quality degradation when the underlying model updates. Building this properly adds 2–3 weeks to a project and is frequently the first thing cut when budgets are squeezed — resulting in a chatbot that silently starts giving wrong answers after a model version update.
Which LLM to use in 2026
| Model | Strengths | API cost (approx) | Best chatbot use case |
|---|---|---|---|
| OpenAI GPT-4o | Strong reasoning, tool use, multimodal | £0.003/1K tokens | Complex multi-turn support, agents |
| Anthropic Claude 3.5 Sonnet | Long context, careful responses | £0.003/1K tokens | Document-heavy Q&A, careful tone |
| Anthropic Claude Haiku 4.5 | Fast, cheap, simple tasks | £0.0003/1K tokens | High-volume simple FAQ, classification |
| OpenAI GPT-4o mini | Cheap, fast, good for structured output | £0.0002/1K tokens | Lead qualification, routing |
| Self-hosted Llama 3 / Mistral | No per-token cost, data stays on-prem | Compute only | High volume, strict data residency |
For most UK businesses, a combination works well: use a fast, cheap model (Haiku, GPT-4o mini) for simple classification and routing tasks, and a more capable model (GPT-4o, Claude Sonnet) for complex responses that require reasoning. This hybrid routing can reduce API costs by 40–60% compared to sending every message to the top-tier model.
Hidden costs nobody mentions upfront
LLM API costs after launch
At £0.003 per 1,000 tokens, a customer service chatbot handling 500 queries/day with average conversation length of 500 tokens costs approximately £225/month in API fees. At 2,000 queries/day, that’s £900/month. These figures are almost never in the build quote.
Content management
Someone needs to keep the knowledge base updated. Product changes, policy updates, new FAQ entries — an unmaintained knowledge base produces outdated answers. Budget ongoing staff time or a content management retainer (£500–£2,000/month for a complex knowledge base).
Model update maintenance
When OpenAI or Anthropic releases a new model version, your prompts may produce different outputs. Regression testing and prompt tuning after model updates is a real ongoing cost: typically 5–10% of your initial build cost annually.
GDPR compliance
Chatbots collecting personal data (names, emails, order details) require privacy notices, data retention policies, and subject access request processes. If you’re sending user data to OpenAI or Anthropic APIs, you need a Data Processing Agreement with them. Legal and compliance setup is rarely included in chatbot build quotes and can cost £2,000–£8,000 to do properly.
Red flags in chatbot quotes
Question the quote if it:
How to scope your chatbot project
Before getting quotes, be clear on three things:
- What are the three most common queries it needs to handle? This defines the core knowledge base and integration requirements better than a long feature list.
- What actions should it take vs. what should it escalate? Define the boundary between autonomous action and human handoff explicitly. This is the most important design decision in the project.
- Where does the data live that the chatbot needs to access? Existing API, database, SharePoint, PDF library — the state and accessibility of your data will affect cost more than the chatbot logic itself.
Summary
AI chatbot development in the UK costs £2,000–£8,000 for no-code platform setup, £8,000–£25,000 for a custom LLM chatbot on an existing product, and £20,000–£80,000+ for deep system integrations or autonomous agents. The key cost drivers are integration depth, knowledge base complexity, conversation statefulness, and how much autonomous action the chatbot takes.
For most UK businesses, the decision between no-code and custom comes down to integration requirements and data control. If your use case is standard customer support Q&A with no unusual integrations, start with a platform. If you need deep system access, strict data residency, or long-term cost efficiency at volume, a custom build is the right investment.
