How much does an AI customer service chatbot cost in the UK?

Off-the-shelf support bots charge per resolved conversation, plus a monthly seat or plan fee. The prices below are the US dollar list prices on each vendor's own pricing page on 7 November 2026. This guide did not check their UK pricing pages.

Read the definition before you compare prices. Intercom's outcome can include a customer who stopped replying after Fin answered, and a conversation Fin hands to a person. Both count at $0.99.

The model costs about a cent per conversation. Hosting, support and the work of keeping answers current cost more.

Worked example with 2,000 conversations a month

Assume each support conversation uses 3,000 input tokens (the question, the instructions and the retrieved help articles) and 500 output tokens. That is an assumption, and longer chats with more history use more.

• On Claude Sonnet 5.5, at $2 input and $10 output per million tokens, 3,000 input tokens cost $0.006 and 500 output tokens cost $0.005. That is $0.011 a conversation, or $22 a month.
• On Claude Haiku 5.5, at $0.10 and $0.50, it is $0.0003 plus $0.00025, or $0.00055 a conversation. That comes to $1.10 a month.
• On Claude Opus 5.5, at $4 and $20, it is $0.012 plus $0.01, or $0.022 a conversation. That comes to $44 a month.

Outcome pricing looks different at the same volume, in the same currency. If 1,000 of those 2,000 conversations count as Fin outcomes, Intercom charges $990 a month before seats. At Zendesk's committed rate, 1,000 automated resolutions cost $1,500. Per conversation, Fin's $0.99 is 90 times Sonnet 5.5's $0.011, though a custom build adds the build fee, hosting and maintenance.

Softomate's own running estimates for a UK SME chatbot are higher than the raw token maths. It puts hosting at £50 to £200 a month, AI usage at £20 to £80 a month for 500 conversations and £200 to £500 for 5,000, and an optional support retainer at £300 to £800. Its total is about £370 to £1,080 a month. Measure your own token counts in a pilot before you budget.

Yarigai builds support agents that answer from a client's own docs, inside the client's own helpdesk and accounts. The client owns the code, the data and the accounts from day one. AI and hosting bills are paid at cost, so the token maths above is what the client pays the model provider.

Every project starts with a free 30-minute scope call on Google Meet. A blueprint follows, which takes about a week and is credited to the build. It includes a clickable prototype, the monthly running cost, a fixed price and the payback maths against what support costs today, such as seats and per-resolution fees. A person approves anything that sends money, emails a customer or deletes data, so refunds stay with a person.

Yarigai quotes a fixed price after the one-week blueprint. Its guarantee reads "If the build isn't profitable for you, we won't build it." If the build doesn't come out ahead within the first year, Yarigai doesn't take the job. More on AI customer support.

A chatbot is the wrong buy when the questions are few, complex or emotional, or when the real problem is a broken process.

• At a handful of tickets a day, a £5,000 build plus £299 a month takes a long time to pay back. Tidio's free plan or Fin's per-outcome pricing costs less to try first.
• Complaints, disputes and anything with legal weight need a person, not a bot.
• Refund rules, prices and policies that change every week need one strict source of truth, or the bot will contradict your own policy pages, as Air Canada's did.
• If most tickets ask where an order is, better tracking emails remove the question.
• A bot that answers from your docs is only as good as the docs, so fix the help centre first.

Phone calls are a separate build. See what an AI receptionist costs.