AI chatbot cost in the UK: build price, running costs and when a subscription is cheaper

What is the real AI chatbot cost for a UK business?

The real AI chatbot cost in the UK is the build plus twelve months of running costs, not the build alone. Running costs scale with conversations, so a bot that becomes popular costs more each month, which is fine as long as each conversation saves more than it costs.

Build costs cover designing the conversation, preparing your content so the AI can answer from it, connecting any systems, building the chat interface or messaging channel, testing and launch. Once paid, they are done.

Running costs come from three possible places. An AI provider bills token usage. A SaaS chatbot vendor bills per seat, per resolution or a flat fee. Meta bills WhatsApp template messages. On top sit small hosting costs and the time to keep answers current.

The mistake many buyers make is comparing a custom build price with a SaaS monthly fee. Compare like with like: total cost over one or two years at the volume you actually expect.

How much does it cost to build an AI chatbot?

The build cost depends on what the chatbot has to do. Answering questions from your own content is the simplest tier; taking actions in your systems, handling several channels and meeting strict logging requirements each move you up.

Most UK small businesses start at tier 1 or 2 and add channels once the bot has proved it answers well. The AI chatbot development page describes each tier's build steps in more detail.

What does an AI chatbot cost to run per 1,000 conversations?

Running cost per 1,000 conversations is the number of tokens those conversations use, multiplied by your AI provider's price per million tokens, with input and output priced separately. The model you choose changes the result far more than anything else.

A token is a chunk of text. Anthropic's pricing documentation gives a rough rule that one token is about four characters or three-quarters of an English word, and its own customer-support example assumes an average of roughly 3,700 tokens per conversation. At that average, 1,000 conversations use about 3.7 million tokens, split between input (your question, the instructions and the retrieved content) and output (the bot's replies).

Input usually dominates in a chatbot, because every turn resends the instructions, relevant extracts from your content and the conversation so far. That is why keeping prompts tight and retrieving only the relevant passages cuts cost directly.

Providers publish prices per million tokens, and they change as models change, so we price with your current provider rates at quote time rather than printing numbers here. We also show you the tokens per conversation from testing, so you can recalculate whenever prices move.

• Cost per conversation = input tokens × input rate + output tokens × output rate.
• Cost per 1,000 conversations = that figure × 1,000.
• Smaller models cost far less per token and handle most FAQ work well.
• Longer conversations and bigger retrieved passages raise cost per conversation.

How do you keep AI chatbot running costs down?

You keep running costs down by using the smallest model that answers well, caching repeated instructions, retrieving only relevant content, and capping conversation length. These choices are made in the build, which is why a careful build lowers the monthly bill.

Prompt caching is one of the biggest levers. According to Anthropic's pricing documentation, reading a cached prompt prefix costs a tenth of the standard input price on most models. A chatbot resends the same long instructions on every turn, so caching them makes a real difference at volume.

Model choice is the other. A small, fast model handles most “what are your opening hours” and “where is my order” questions. We route only harder questions to a larger model. Batch discounts, which Anthropic lists as half price for asynchronous jobs, do not help live chat, but they suit overnight work such as summarising the day's conversations.

Finally, measure. We log tokens per conversation from day one, so you see cost trends in plain numbers instead of discovering them on an invoice.

How does per-resolution AI chatbot pricing work?

Per-resolution pricing charges you each time the AI agent resolves a conversation, rather than per seat or per token. It aligns cost with outcomes, but at high volume it can cost more than running your own bot on token pricing.

Intercom's Fin is the best-known example. Its pricing page describes a charge per outcome, counted when a customer confirms the issue is resolved, when they do not ask for more help after Fin responds, or when Fin completes a workflow including handoffs, and says you are charged only once per conversation. Fin can be used without buying seats, while bundling it with an Intercom help-desk plan adds per-seat fees.

The model is attractive when your volume is modest and your team already uses that help desk, because you pay nothing for conversations the AI fails to resolve. Watch the definition of a resolution, though: a customer who simply gives up and leaves may count as resolved under “does not ask for more help”.

To compare, work out your expected monthly resolutions, multiply by the per-outcome price, and set that against a custom build plus token costs at the same volume. We will do that sum with you using your real ticket numbers.

Are flat-fee chatbot subscriptions cheaper than a custom build?

Flat-fee chatbot tools are often the cheapest option at low volume and for simple FAQs, and a custom build becomes better value as volume grows or the bot needs to act in your systems. There is no single cheaper answer; there is a crossover point.

Flat-fee tools charge a monthly plan with limits on conversations, messages or AI credits. Below those limits, cost is predictable. Above them you move up a plan. They rarely connect deeply to your booking system, stock or CRM, and conversation data lives on their platform.

A custom build costs more upfront but runs on token pricing you control, connects to whatever you need, and keeps content, prompts and logs in your accounts. If the tool you are considering already fits your needs and your volume is small, use it. If you are bending your process to fit a template, that is the signal to price a custom bot.

What does a WhatsApp AI chatbot cost in the UK?

A WhatsApp AI chatbot costs the build, the AI usage, and Meta's fees for template messages. Since 1 July 2025, Meta charges per delivered template message rather than per conversation, and replies within an open customer service window are free.

According to Meta's WhatsApp Business Platform pricing, when a customer messages you a 24-hour customer service window opens, and your non-template replies inside it cost nothing. Utility templates (order updates, appointment reminders) sent inside that window are also free. Marketing templates are always charged. Authentication and utility templates outside the window are charged, with rates set by the recipient's country code and volume tier.

There is also a free entry point: when someone reaches you through a Click to WhatsApp ad or a Facebook Page button, and you reply, a 72-hour window opens in which any message is free.

For a customer-service chatbot that mostly replies to incoming messages, Meta fees can therefore be small. For a bot sending promotions or reminders to people who have not messaged recently, they are a real line in the budget.

What does an AI chatbot cost to maintain each month?

Monthly upkeep covers updating the knowledge the bot answers from, reviewing conversations for wrong or weak answers, adjusting prompts, watching cost and handling changes in AI models. A chatbot left alone gradually drifts out of date.

Knowledge updates matter most. When prices, opening hours, stock, policies or staff change, the bot must learn the new facts, or it will confidently give old ones. Where possible we connect it to the live source so updates happen automatically.

Conversation reviews catch the questions it cannot answer well, which often point to gaps in your website content too. AI providers also retire older models over time, so occasionally the bot needs moving to a newer model and retesting.

With us, the first two months after launch are free support. After that, chatbot care starts at US$120/mo. The website maintenance cost guide explains how care is scoped more generally.

UK GDPR, transparency and an AI chatbot: what the build must support

If your chatbot handles personal data, and most do, UK GDPR applies to how it collects, uses and stores it. The ICO publishes specific guidance on AI and data protection, including “Explaining decisions made with AI” and an AI and data protection risk toolkit, which your adviser can use to assess your set-up.

The build supports your obligations in practical ways: telling users they are talking to an AI, linking your privacy notice in the chat, collecting only the details needed, choosing AI providers and hosting regions you are comfortable with, setting how long transcripts are kept, and giving staff a way to find and delete a person's data on request.

A human handover route matters too, both for customer experience and for anything significant. We design bots so they do not make decisions with legal or similarly significant effects on people; those go to a person.

Compliance itself is your responsibility, confirmed by your own adviser. We do not give legal advice or claim certification; we build the controls and document how data flows so your adviser can review it.

Is an AI chatbot worth the cost for a small UK business?

An AI chatbot is worth it when it answers enough repetitive questions, or captures enough out-of-hours enquiries, to save more than it costs. If you get a handful of enquiries a week, a good FAQ page and a fast reply on WhatsApp may serve you better.

Signs it will pay back: staff answer the same questions many times a day; enquiries arrive in the evening and go cold by morning; customers ask for order status or availability that a system already knows; or you run a shop where quick answers turn browsing into buying.

Signs it will not: questions are all different and need expert judgement, volumes are low, or your website content is thin, because the bot can only be as good as what it answers from. Sometimes improving the content is the better first spend.

We will tell you honestly if a chatbot is not the right investment yet. It is cheaper for both of us than building something you switch off.

Building an AI chatbot with a team in India from the UK

You work directly with the three of us on WhatsApp and video calls. The chatbot runs in your cloud account with your AI provider account, so everything you pay for is in your name.

India is four and a half hours ahead of the UK in British Summer Time and five and a half hours in winter. Our afternoon overlaps your late morning onwards, which is when we hold demos and test sessions. Between calls, you test the bot on a staging link and send notes as messages.

Quotes are in USD; you can pay from a GBP account by Wise, bank wire or PayPal, with invoices issued from India. Nothing is billed until you approve the written quote. AI provider and Meta fees are billed to you directly by them, with no markup from us.

How to compare AI chatbot cost across UK quotes

Compare chatbot quotes on the full cost over a year, how the bot handles questions it cannot answer, and who owns the accounts. A cheap build on someone else's AI key, billed back to you with a markup, can cost more than a dearer build in your own name.

• Is the AI provider account in your name, billed to you directly?
• Is there an estimate of tokens or cost per conversation from testing?
• What happens when the bot is unsure: does it hand over, or guess?
• Which of your systems does it connect to, and is each priced separately?
• Where are transcripts stored, and for how long?
• Who updates the knowledge base when your information changes?
• Can you take the code, prompts and content elsewhere?

Does a chatbot help your business show up in AI search?

Not directly. A chatbot on your site answers visitors who are already there; it does not make ChatGPT, Gemini or Google's AI Overviews recommend you. That depends on the content on your public pages.

There is a useful side effect, though. Building a chatbot forces you to write clear, factual answers to the questions customers really ask. Put those answers on your website as well, and you improve your chances of being quoted by AI search tools.

Conversation logs are also a list of real questions in customers' words, which is valuable input for content. Our AI search optimisation service uses exactly this sort of material, alongside monthly SEO from US$150/mo. Nobody can guarantee AI inclusion or rankings.

Worked example: costing a chatbot for a Sheffield MOT garage

This scenario is hypothetical and only illustrates the arithmetic. Say a two-bay MOT and servicing garage in Sheffield gets many repeat questions by web chat and WhatsApp: prices for a service, whether a car is ready, how to book an MOT, and opening hours.

Build: a tier 2 bot answering from the garage's price list and FAQs, checking job status from its workshop system, and offering MOT slots from the booking calendar, on the website and WhatsApp. Two integrations take it above our US$600 starting price.

Running: testing shows the average tokens per conversation. The owner multiplies by their expected monthly conversations and the provider's current rates for a small model, with caching on the long instructions. Most WhatsApp traffic is replies to customers who messaged first, so it falls inside the free 24-hour window. The only regular Meta charge is utility reminders sent the day before an MOT when no window is open.

Upkeep: price list updates flow from the workshop system automatically. After two free months, care from US$120/mo covers monthly conversation reviews and prompt tuning. The owner compares all of this with a per-resolution tool at the same volume and chooses with full numbers in front of them.

Red flags that inflate AI chatbot cost

Be cautious if a supplier cannot explain what the bot costs to run, or will not put the AI account in your name. Those two gaps cause most chatbot budget surprises.

• “Unlimited AI conversations” on a flat fee with no mention of fair-use limits.
• No handover to a human when the bot is unsure.
• The bot runs on the supplier's AI key, rebilled with a markup.
• No plan for updating the bot when your prices or policies change.
• Claims that the chatbot is “GDPR certified” or “fully compliant”.
• Promises of guaranteed sales uplift or specific percentage savings.
• No test period with your real questions before launch.

AI chatbot cost UK: budget checklist

Before you approve any chatbot, make sure you have a figure or an owner for each line below. It turns a vague AI project into a normal budget.

• Build: tier, channels and each integration priced.
• AI usage: tokens per conversation from testing, and expected monthly volume.
• SaaS alternative: per-seat, per-resolution or flat-fee cost at the same volume.
• WhatsApp: expected template messages by category.
• Hosting for the bot's back end.
• Care: knowledge updates, reviews and model changes.
• Data: where transcripts live and how long they are kept.
• Accounts: AI provider, cloud and Meta all in your business's name.