Chatbot Cost in the UK: SaaS vs Custom AI Agent (2026)

Chatbot cost in the UK comes down to three routes, and each one is billed differently. A software-as-a-service (SaaS) chatbot is a subscription, often with a fee each time it resolves a conversation. Building on a model provider’s API means a small usage bill plus the time to build and maintain the thing. A custom AI agent that acts inside your systems starts, with us, from £5,000, with maintenance and support from £400 a month.

If all you need is a business chatbot that answers questions from your help centre, a SaaS tool is likely the better buy, and we say so below. The rest of this article puts official prices side by side and works through the model usage cost with its inputs shown. It then lists what quotes tend to leave out and covers who is responsible when a chatbot gets an answer wrong.

Key takeaways

• A SaaS chatbot is billed by subscription, by resolution or both. At the time of writing, Intercom lists $0.99 per Fin outcome and Zendesk lists £1.30 per committed automated resolution.
• Model usage is the small part of an AI chatbot cost. In our illustrative example of 2,000 conversations a month it comes to between $5.40 and $108 a month, depending on the model. The build around it is where the money goes.
• A custom AI agent earns its cost when it has to act inside your systems, or when per-resolution fees at your volume keep climbing. Ours start from £5,000 for a first agent and £12,000 for a rollout.
• Assume you are responsible for what your chatbot says. In a 2024 Canadian case, a tribunal held an airline responsible for what its website chatbot told a customer.
• If you are not sure what to automate, price the job before the tool. Our AI readiness assessment is £3,000, fixed, and takes one week.

How much does a business chatbot cost in the UK?

There is no single UK price, because the three routes charge for different things. This table sets them side by side, using the vendors’ own published prices and ours.

Route	How you pay	Price points we verified (October 2026)	Best when
SaaS chatbot or AI agent	Subscription, a fee per resolution, or both	Intercom Fin: from $0.99 per outcome. Zendesk: £1.30 per committed automated resolution. Tidio Starter: $24.17 a month, billed annually	Customers ask questions your help content already answers
Build on a model API	Per-token usage, plus developer time	Claude Haiku 4.5: $1 input and $5 output per million tokens. OpenAI gpt-6-luna: $0.10 input and $0.50 output	You have developers and want control of the experience
Custom AI agent	Fixed build price, then maintenance and usage	CodeLeap: first agent from £5,000, rollout from £12,000, maintenance from £400 a month	The agent has to read and change data in your own systems

All third-party prices are as the vendors’ pages show them at the time of writing (October 2026), and they change. Several vendors publish in US dollars even to UK visitors, so we have kept each currency exactly as shown and converted nothing. Where a page does not say whether VAT is included, we say so below rather than guess.

Chatbot or AI agent: what changes the price?

A chatbot answers questions. An AI agent pursues a goal across several steps and uses your tools to act: it looks up the order, raises the refund, updates the record. Our guide to AI agent examples shows what that looks like in practice.

The price difference follows the access. A bot that answers from a help centre needs content and a model. An agent that acts needs permissions in your systems, error handling when those systems misbehave, and a rule for when a person takes over.

Our work for an energy broker shows the shape of it. For SBU we built an automation layer across Salesforce, an LLM (Anthropic’s Claude) and the client’s email. It classifies replies and drafts responses for approval, and an 85% confidence threshold decides what runs automatically and what waits for a person. The client’s Salesforce edition did not include API access as standard, so access had to be negotiated on the existing plan. That kind of integration work does not show up on a chatbot subscription page, and it is a large part of what separates the routes. If you want the integration side in detail, see our guide to API integration.

What do SaaS chatbots cost?

SaaS chatbots use three billing shapes, and many combine them: a monthly fee, a fee per resolved conversation, or a pool of credits. Here is what the official pages show.

• Intercom. Its pricing page shows “From $0.99 per Fin outcome” and “No seats required”. It says you are charged once per conversation, even if several questions are answered, and that when Fin is used with an existing helpdesk a minimum monthly commitment applies (the example given is 50 outcomes). The price is in US dollars. Intercom counts an outcome when a customer confirms the issue is resolved or does not ask for more help after Fin replies. It also counts one when Fin completes a workflow, including a handoff, so read the definition before you budget (Intercom pricing).
• Zendesk. The UK pricing page shows £1.30 for committed automated resolutions and US$ 2.00 for pay-as-you-go resolutions. Suite Team is £45 per agent per month paid yearly. The page does not say whether VAT is included (Zendesk UK pricing).
• Tidio. The Starter plan shows $24.17 a month with 100 billable conversations, and the Lyro AI agent as a standalone product starts at $32.50 a month. Those prices reflect the annual payment option, which the page describes as “2 months free” (Tidio pricing).
• HubSpot. Its Customer Agent runs on HubSpot Credits, sold at $0.010 per credit beyond what a plan includes, in US dollars (HubSpot pricing).

A quick illustration, with the input stated: if a bot resolves 1,000 conversations a month, Intercom’s headline rate comes to $990 a month and Zendesk’s committed rate to £1,300 a month, before any per-agent plan fees. That is an assumed volume, not a forecast, and you should check your own volume against how each vendor defines a resolution.

SaaS is the right answer more often than agencies like to admit. If the job is answering common questions from content you already publish, and you do not need the bot to touch your systems, a subscription is cheaper to start and someone else maintains it. The costs to watch are the per-resolution fees as volume grows, and the limits of what the vendor’s bot can reach.

Building on a model API: what usage really costs

Model providers charge per token, which is roughly a few characters of text, for what you send in and what comes back. Anthropic’s page lists Claude Haiku 4.5 at $1 per million input tokens and $5 per million output tokens, and Claude Sonnet 5.5 at $2 and $10 (Anthropic pricing). OpenAI’s page lists its gpt-6-luna model at $0.10 input and $0.50 output per million tokens on its Standard tier (OpenAI pricing). Both quote US dollars.

Here is an illustrative monthly usage bill for a customer-service bot. Every input is our assumption, so replace them with your own:

• 2,000 conversations a month
• 6 model calls per conversation (one per turn)
• 3,000 input tokens per call (instructions, the relevant knowledge-base passages and the conversation so far)
• 300 output tokens per call

That is 36 million input tokens and 3.6 million output tokens a month.

Model	Input price	Output price	Monthly usage cost
OpenAI gpt-6-luna	$0.10 per million	$0.50 per million	$5.40
Claude Haiku 4.5	$1 per million	$5 per million	$54
Claude Sonnet 5.5	$2 per million	$10 per million	$108

Prices are as published in October 2026, in US dollars, with no conversion or VAT applied. Anthropic’s page also lists a cache-read price for repeated prompt content, which is $0.10 per million tokens on Haiku 4.5, so a long unchanging instruction block can cost less than this example assumes. The example also uses the same token counts for every model. Anthropic notes that Claude 4.7 and later models use a newer tokenizer that produces about 30% more tokens for the same text, so treat the Sonnet 5.5 row as a lower bound.

The point of the example is the size of the number. Usage is rarely what makes a chatbot expensive. The cost sits in everything around the model, and it is the same whichever route you take.

What does a custom AI agent cost?

A custom agent is a fixed build price, then running costs. With CodeLeap, scope and price are agreed before work starts, so there is no meter running on the build.

• First agent, from £5,000. One agent, one named job, inside your existing tools, with human sign-off where it matters.
• Agent rollout, from £12,000. Agents across several workflows, with escalation to a person when confidence is low.
• After launch, from £400 a month. Maintenance and support, plus the model usage and hosting that the agent consumes, as in the example above.

The commercial terms are the same ones we use on every build. You pay by milestone as each stage is delivered. You can cancel within the first 21 days and get your money back in full, and all code and IP are yours once the work is delivered and paid for. That last point matters for the cost comparison. A SaaS subscription is rented for as long as you use it, whereas an agent built for you is an asset you own.

On volume, the two cost shapes behave differently. Per-resolution fees rise with every conversation, while an owned build is mostly a fixed cost plus usage. Take the 1,000-resolutions example. Intercom’s rate comes to $11,880 over twelve months. A first agent starts from £5,000, and with maintenance from £400 a month the first twelve months start from £9,800, plus model usage of the kind shown above. The totals are in different currencies, so convert at the day’s exchange rate before comparing them. The shape is what matters: one bill grows with volume and the other mostly does not.

That is not a reason to build at low volume. If your bot is answering a few hundred questions a month from your help centre, buy the subscription. Build when the agent has to do something a vendor’s bot cannot, or when the fees have become a line you notice.

The costs a chatbot quote leaves out

Whichever route you choose, five costs sit outside the headline price.

• Preparing the knowledge base. A bot is only as accurate as the content it draws on. Out-of-date policies, contradictory pages and answers that live in someone’s head have to be found, fixed and kept current.
• Integrations. A bot that checks an order, books a slot or updates a record needs a connection to that system, and that is where access limits and rate limits appear. Our guide to API integration covers the pitfalls.
• Human handoff. Decide what the bot does when it is unsure, who receives the conversation, and whether they get the full context. A bot with no way out turns a small problem into a complaint.
• Monitoring. Models and your own content both change. Someone has to read a sample of real conversations, track the failures and fix them.
• Testing the answers. Before launch, run real questions, including awkward ones, and check each answer against your policies. A bot that sounds confident and is wrong is worse than no bot.

These are the same whether you buy or build. SaaS vendors build some of them into the product, and you pay for that in the fees.

Who is responsible when a chatbot gets it wrong?

Assume you are. A Canadian tribunal has already held a business responsible for what its chatbot told a customer.

In Moffatt v. Air Canada (2024 BCCRT 149), a British Columbia tribunal heard a dispute over what an airline’s website chatbot had told a customer. The airline argued it could not be held liable for information given by its agents or representatives, including a chatbot. The tribunal said that, in effect, this treated the chatbot as a separate legal entity and called it “a remarkable submission”. It said the airline was responsible for all the information on its website (the decision, paragraph 27). It is a Canadian decision and does not bind a UK court, but it is a useful illustration of how a tribunal can approach a chatbot’s answers. This is general information, not legal advice.

If your bot handles personal data, UK GDPR applies as well. The ICO’s guidance on AI and data protection covers topics including transparency, fairness, accuracy and security. Its page says that, due to changes made by the Data (Use and Access) Act, “this guidance is under review and may be subject to change”, so check the current version (ICO guidance).

In practice, that points to a few design rules. Restrict the bot to approved sources. Give every user a route to a person. Log what it says, and review a sample.

Which route should you choose, and what next?

Start with the job you want done.

• Answering common questions from content you already publish: buy a SaaS tool. Check how it counts resolutions and what happens to the bill at twice your volume.
• A bot you want to shape and own, with developers in-house: build on a model API, and budget for the work around the model, not just the usage.
• An agent that reads and changes data in your systems, with a person approving what matters: this is where a custom AI agent pays off, and where our first-agent and rollout prices apply.
• A back-office process rather than a conversation: look at AI process automation instead. A chat window may not be the right shape for the job at all.

If you cannot yet say which of those describes you, that is the question to answer first. Our AI readiness assessment takes one week at a fixed £3,000 and is built for buyers who are not yet sure what to automate.

Write down the one job you want the chatbot or agent to do, how many times a month it happens, and which systems it would need to touch. With those three answers, we can talk through which of the three routes fits. Book a call about AI agents and automation, and bring those three answers with you.