How much does an AI agent cost a UK business?
The short answer: anywhere from £40 a month to six figures, which is another way of saying the question needs sharpening. A subscription chatbot and a custom agent wired into your job system are both called AI agents, and they have as much in common as a stapler and a production line. Most UK businesses commissioning their first custom agent should expect a build in the £3,000 to £25,000 range, plus modest running costs. This article breaks down where those numbers come from, what moves them up or down, and what the evidence says you get back.
A note on why the ranges are wide. Market surveys of agent pricing, such as this 2026 pricing breakdown and this small business cost guide, quote everything from £40 per month to £80,000 per project, and both are accurate. The price of an agent is mostly the price of connecting it to your business. The model is the cheap part.
The three routes, and what each costs
Off-the-shelf tools: £40 to £400 a month
Website chatbots, meeting schedulers, email drafters and the AI features appearing inside software you already pay for. Configuration takes days, not months, and no engineering is required. The limitation is reach. These tools cannot see your job data, your pricing rules or your order history, so they stop at the edge of your systems. For answering common website questions or summarising meetings they are good value. They will not run a process for you.
A custom agent inside your existing systems: £3,000 to £25,000
This is the route most businesses actually need, and the one our agentic AI consultancy exists for. An engineer connects an agent to the software you already run, through controlled interfaces, so it can read the enquiry, check the history, draft the quotation from your rules and hand the result to a person for approval. No migration, no new platform for staff to learn. The build cost covers integration work, permission design, testing against real cases and deployment. Where the first process is narrow and the systems are tidy, the low end of the range is realistic. Complex integrations, messy data or heavier governance push it up.
An application built around AI: £25,000 and up
Sometimes the opportunity justifies building the software around the agent rather than fitting the agent into the software. That is a custom development project with AI at its centre, and it is priced like one. Few businesses should start here. Prove the return on a smaller integration first, then let the evidence argue for the bigger build.
What actually moves the price
Four factors explain most of the variance between quotes, and none of them is the AI model itself.
Integration depth. An agent that reads from one system costs less than an agent that reads from three and writes back to two. Each connection needs building, securing and testing. Systems with clean APIs are quick. Systems without them need more engineering, which is often the moment a business discovers the real cost of software that was never built to talk to anything.
The state of your data. Agents reason over your records. If pricing lives in a spreadsheet nobody fully trusts, or customer history is split across systems that disagree, part of the budget goes on making the data dependable first. This work is unglamorous and it is also where much of the lasting value sits, because clean data pays back with or without AI on top.
Governance and approvals. Deciding what the agent may do alone and what routes to a person takes design work. Permissions, audit logging and human approval steps are not optional extras for anything touching money or customers. We covered our approach in the consultancy launch announcement, and budgeting for it at the start is far cheaper than retrofitting it after an incident.
Running costs. A deployed agent incurs model usage fees, typically £40 to £400 a month per agent for most business workloads, plus hosting and monitoring. Providers reduce model prices regularly, so this line tends to fall over time, but it never reaches zero and a quote that omits it is incomplete.
What the evidence says about returns
The UK government's AI adoption research, published by the Department for Science, Innovation and Technology in 2025, found that only 16 per cent of UK businesses currently use AI at all, while 65 per cent plan to increase AI budgets. Among businesses that have adopted it, 75 per cent report improved workforce productivity. Cost was cited as a barrier by 76 per cent of non-adopters, which is part of why we think plain public pricing information matters.
The pattern inside larger organisations is instructive too. McKinsey's State of AI research finds that while nearly nine in ten organisations now use AI somewhere, only a small minority report meaningful financial impact, and the businesses that do are the ones that redesigned specific workflows around the technology rather than buying subscriptions and hoping. The lesson for a smaller business is encouraging. Returns follow from picking a real process and engineering the agent into it, which is exactly the part a smaller business can do faster than a large one.
On the question every owner asks quietly, the evidence is also reassuring. US Census Bureau data on AI-adopting firms found the most common outcome was no change in headcount. The time agents free up gets reallocated to the work that was never getting done, from chasing overdue invoices to the competitor round-up nobody had a spare Monday for.
How to spend well
The businesses that get burned buy broad. The businesses that do well buy narrow. One process, with a measurable before and after: cost per quotation, hours to produce the month-end report, time from enquiry to response. Deploy one agent against it, compare the numbers, then expand on evidence. This is how we structure engagements, and it is also simply how you keep the first cheque small while you find out what agents are worth in your business rather than in a vendor's case study.
It also means the right first invoice is rarely the £25,000 one. An assessment that maps your processes and produces a costed shortlist gives you a number for your business, not a range from an article. Ours takes about a day, and the first candidates usually surface within the morning.