How Much Does an AI Chatbot Cost in 2026? A Plain-English Guide for UK & US Businesses

“How much will an AI chatbot cost us?” is the first question almost every founder asks us. The honest answer is: anywhere from a few pounds a month to tens of thousands, depending on what you need the chatbot to do. This guide breaks down where the money goes, so you can budget with confidence and avoid paying for things you don’t need.

The short answer

• Off-the-shelf chatbot tool: a monthly subscription, usually the cheapest way to start. Good for simple FAQs on a small website.
• Custom chatbot MVP: a one-off build trained on your own content and connected to one or two of your systems. At Xenova, this starts from £2,500 and takes 3–4 weeks.
• Chatbot that searches your documents (RAG): answers from manuals, policies or contracts, with sources. Our proof of concept starts from £4,000.
• AI agent that takes actions: books appointments, updates your CRM, processes refunds. Our pilot starts from £6,500.

On top of the build, expect a modest monthly running cost for AI model usage and hosting. For most small businesses this is far lower than people expect, because you only pay for the conversations that actually happen.

Off-the-shelf tool or custom build?

Subscription tools are great when your questions are simple and your content lives on a few web pages. You paste in your website, tweak the tone and you’re live in an afternoon.

A custom build starts to pay off when any of these are true:

• The answers live in documents (PDFs, policies, product sheets), not just web pages.
• The chatbot needs to look things up in your systems: order status, stock, bookings, customer records.
• You need control over data: where it’s stored, what the model can see, and audit logs for compliance.
• You want the chatbot to do things, not just talk, such as creating a booking or opening a support ticket.
• Per-seat or per-conversation subscription fees are becoming expensive as you grow.

What actually drives the price of a custom chatbot

1. Where the knowledge comes from

Answering from a handful of FAQ pages is quick. Answering accurately from hundreds of documents needs a proper retrieval pipeline (often called RAG): splitting documents sensibly, indexing them, and making sure the bot quotes the right source. That’s where most of the engineering time goes, and it’s the difference between a bot people trust and one they ignore.

2. Integrations

Each system the chatbot connects to (Shopify, a booking system, HubSpot, a custom database) adds work: authentication, error handling and testing. One or two integrations is typical for a first version.

3. Channels

A website widget is the usual starting point. WhatsApp, Slack, Microsoft Teams or phone (voice) each add effort. Voice is the most involved, because speech recognition, latency and accents all need tuning.

4. Languages

Modern models handle major languages well out of the box. Less common languages or mixed-language speech can need extra work. For example, our BookKaru voice agent handles Urdu and Roman Urdu, and we fine-tuned open models (Llama and Qwen) on around 900 real call-centre recordings to get it right.

5. Safety, privacy and compliance

For UK businesses, UK GDPR applies to any personal data the chatbot handles. Regulated sectors (health, finance, legal) usually need data minimisation, clear handover to a human, logging, and sometimes hosting in a specific region. Building this in from day one costs much less than adding it later.

Ongoing costs to budget for

• AI model usage: charged per amount of text processed. Low for most SMEs; it grows with conversation volume.
• Hosting: a small cloud server or serverless setup for the chatbot backend and document index.
• Maintenance: updating content, reviewing conversations the bot got wrong, and keeping integrations working when other systems change.

A good partner will show you real usage numbers after the first few weeks, so your running cost is based on data rather than guesses.

How to keep the cost down

1. Start with one job. Pick the questions or tasks that eat the most staff time, and automate those first.
2. Run a short pilot. Three to six weeks is enough to prove value with real customers before you commit to more.
3. Use your existing content. Help-centre articles, emails your team already sends and internal documents are the best training material.
4. Measure deflection and conversion. Track how many conversations the bot resolves and how many leads or bookings it creates, then decide what to build next.

A realistic example

For a client in Finland, we built an AI document Q&A system that lets users ask questions in plain language and get answers drawn from their own document library, with the source shown. A project like this typically fits our RAG proof-of-concept scope: a few weeks to a working system on real documents, then a decision on whether to roll it out further.

Frequently asked questions

How long does it take to build an AI chatbot?

A focused custom chatbot MVP usually takes 3–4 weeks. A document-search (RAG) proof of concept takes about 3 weeks, and an AI agent that connects to several systems takes 4–6 weeks.

Will an AI chatbot replace my support team?

No, and it shouldn’t try to. The best results come from letting the chatbot handle repetitive questions instantly, day and night, and passing anything complex or sensitive to a person with the full conversation attached.

Do I own the chatbot after it’s built?

With a custom build you normally should, but always check the contract says so. Owning the code and your data is one of the main reasons businesses move away from subscription tools as they grow.

Get a clear quote in 15 minutes

Tell us what you want the chatbot to do and we’ll give you an honest estimate, including whether an off-the-shelf tool would serve you better. Book a free 15-minute scoping call or see our fixed-scope packages.