AI chatbot pricing in the UK is one of the most opaque corners of software development. You'll get quotes ranging from £900 to £80,000 for what sounds like the same brief, and most agencies won't put a number on a page until you've had three calls with a sales lead. So here's the version we'd give a friend over coffee — three honest tiers, what each actually includes, what makes the number move, and the running costs nobody mentions until contract day.
For context, we built Lexa — Pakistan's first AI legal chatbot, live at lexa.lawyer — and we ship AI chatbots into other products as part of our AI integration work. This post is the conversation we wish more buyers had before signing.
Starter — from £2,000
Who it's for: founders and small teams who need a working AI chatbot in front of users in two or three weeks. Typically a single use case — answering FAQs grounded in your existing content, qualifying leads, triaging support, or onboarding a new user.
What you get: an LLM-backed chat interface wired into your product (web or in-app), prompts tuned to your tone and scope, basic conversation memory within a session, one or two safe-refusal patterns for off-topic questions, and a simple admin view so you can see what users are asking. Hosted on your preferred cloud (Vercel, AWS, your existing infra) with a clean, on-brand UI.
Timeline: 2–3 weeks from kickoff to live, depending on how clear the content and brand input is on your side.
What it isn't: a full RAG pipeline over thousands of documents, multi-channel deployment (WhatsApp + web + Slack + CRM), or custom training. Those move you into the next tier — but a surprising number of "AI chatbot" needs land here cleanly.
Standard — £5,000 to £15,000
Who it's for: product teams shipping AI as a real feature, not a marketing checkbox. This is the bracket most of our chatbot work lives in.
What you get: a production custom chatbot grounded in your data via retrieval-augmented generation (RAG) — embeddings, a vector store, a thoughtful chunking and retrieval strategy so answers come from your sources, not the model's memory. Real guardrails: input/output validation, prompt-injection defences, safe refusal for high-stakes questions, traceable citations so a user (or your legal team) can verify an answer. A custom UI that matches your product, a small ops dashboard for your team to see what's being asked, and an evaluation pipeline that catches regressions when you change a prompt or model.
This is the tier Lexa was delivered in. Read about how we built it for the full story.
Timeline: 6–10 weeks from kickoff to live.
Enterprise — from £20,000
Who it's for: organisations with regulated content, multi-channel deployment, multiple languages, complex integrations, or a roadmap that goes beyond a single chatbot into agents and copilots.
What you get: advanced RAG (hybrid search, rerankers, query routing across multiple corpora), multi-channel deployment (web + WhatsApp + Microsoft Teams + Slack + voice), agent capabilities so the chatbot can take actions — book a meeting, raise a ticket, query a CRM — and the full evaluation, observability, and audit-trail stack required for serious production AI in regulated industries. Add-ons like custom fine-tuning, custom model hosting, or PII/data-residency controls live here too.
Timeline: 10–16 weeks, sometimes longer when integrations are deep.
Ongoing / monthly maintenance
AI chatbots are not "ship and forget." The model providers change pricing and capability monthly; your content changes; your users find edge cases nobody anticipated. Expect:
- From £250/month: light maintenance — dependency updates, small prompt tweaks, monitoring, monthly health report.
- £500–£1,000/month: production cadence — eval pipeline runs, monthly retrieval-corpus refresh, security patching, small feature work.
- £1,000+/month: active product team — multi-channel, ongoing prompt iteration, model migrations, performance work.
Separate from this is the inference cost — what you pay your LLM provider per call. That's usually a small operating line for chat workloads (a few pounds per thousand conversations on hosted models), but it scales linearly with usage so plan for it.
What actually moves the price
The variables that matter, in roughly the order they hit the budget:
- Knowledge surface size and complexity. A chatbot grounded in 50 pages of clean content is fundamentally cheaper than one grounded in 50,000 pages across multiple formats.
- Integrations. Each external system the chatbot reads from or writes to (CRM, ticketing, calendar, internal APIs) adds engineering time.
- Channels. Web only is the cheapest target. Every additional channel — WhatsApp, Slack, voice — is a new deployment surface to test and maintain.
- Languages. Multi-language adds testing and content overhead, especially for nuanced domains like legal or medical.
- Compliance and data residency. Regulated content, GDPR-sensitive PII flows, or audit-trail requirements add meaningful engineering work.
- UI custom design. A polished, animated, on-brand chat UI takes more time than an off-the-shelf widget.
The hidden costs nobody quotes you
Three running costs that catch first-time buyers off-guard:
- LLM inference. Cheap on hosted models for chat, but it scales with usage and long-context retrieval can be expensive — worth modelling at expected volume.
- Vector storage and embeddings. Tiny at the starter tier; meaningful at enterprise scale (hundreds of thousands of documents).
- Evaluation and prompt iteration over time. Quality is a number you have to keep an eye on. A small monthly eval cadence is what separates a chatbot that gets better from one that quietly drifts.
Why our starter is £2k
Because we don't run an agency middle layer. The senior engineers who scope your project are the ones who build it — no account manager, no junior-bench markup, no "discovery phase" that exists to bill hours. We've shipped this exact pattern across 11+ products and we know what the smallest useful version looks like. That's how we can responsibly start a custom chatbot from £2k where most agencies start at £15k.
How Lexa was built
Lexa — Pakistan's first AI legal chatbot — sits inside the Standard tier. The work that mattered: a real RAG pipeline grounded in the actual legal corpus so answers cite their basis; safe-refusal behaviour so the system declines questions it shouldn't answer; a calm, plain-language conversation design for non-lawyers under stress; and a fast Next.js front end that makes the assistant feel instant. We wrote the full build story here — it's the most honest case study of what production AI chatbot work actually looks like in 2026.
FAQ
Why do AI chatbot quotes vary so wildly? Because "AI chatbot" describes everything from a £400 widget that wraps an OpenAI call to a £100,000 enterprise platform. Scope, channels, grounding, and accuracy requirements decide where you land. A clear scope is the cheapest cost-control tool you have.
Can I really get a production-grade AI chatbot for £2k? At the Starter tier — yes, for the right scope. Single use case, single channel, content you already have, no deep integrations. We've delivered exactly this. What you can't get for £2k is an enterprise platform with multi-channel deployment and advanced RAG — and we'll tell you up front when your brief is bigger than the Starter tier.
Do I need to choose a specific LLM provider? No. We build with a provider-agnostic layer so you're not locked into OpenAI, Anthropic, Google, or any single vendor. Model choice is a tuning decision, not a foundational one.
How long does an AI chatbot take to build? Starter: 2–3 weeks. Standard: 6–10 weeks. Enterprise: 10–16 weeks. We can usually have an internal demo in your hands within the first 7–10 days regardless of tier.
Who owns the code, the prompts, and the data? You do. Code in your repos, prompts and pipelines documented, data in your accounts, no licensing back to us. The engagement ends; the system stays yours.
Building an AI chatbot and want a real price for your specific brief? Get a project quote for your AI chatbot — you'll walk away with an honest tier recommendation and a written estimate, not a sales deck.
