What AI automation actually costs a UK small business

The most common question we get asked is “what does it cost?”, and the most common answer in our industry is “it depends”, followed by a request for a discovery call. That’s not good enough. You can’t decide whether to pick up the phone without a number.

So here are ours. They’re our ballpark figures as of October 2026, for the kind of work we do for small and mid-sized businesses in Yorkshire. Other firms will differ, and a London agency will typically be double. Treat these as a sanity check for any quote you get, from us or anyone else.

The three things people actually buy

A workflow automation: roughly £1,500 to £6,000. One job, start to finish. An enquiry comes in from the website and ends up qualified, logged in the CRM, with a draft reply waiting for someone to check. An invoice arrives by email, gets read, matched to a purchase order and queued in the accounts package. These are built on tools like Make.com or n8n with an AI model doing the reading and drafting in the middle. The spread is mostly about how many systems are involved and how messy the inputs are. Three systems with clean data sits at the bottom; five systems and PDFs scanned sideways sits at the top.

An AI agent with human sign-off: roughly £6,000 to £25,000. The difference between an automation and an agent is that an agent makes decisions. It reads a situation, works out what to do, and either does it or asks. A support agent that answers the routine 70% of tickets and routes the rest. A quoting assistant that builds the estimate from your price book and past jobs. The cost is in the decision logic, the approval design and the testing, because an agent that’s wrong one time in twenty needs catching nineteen times out of twenty. We’ve written about where agents genuinely work in what AI agents can actually do.

Bespoke software with AI in it: £20,000 upwards, usually £30,000 to £80,000. A platform. Something your business runs on, with logins, data, reports and AI woven through it where it earns its place. Our care home governance system, Guardian Care, is this kind of thing. These are quoted properly after a scoping exercise, because the range is too wide to guess from a phone call, but if someone quotes you £8,000 for a platform, they’re quoting for a prototype and you’ll find that out in month four.

The running costs nobody puts in the proposal

Build cost is the headline. Running cost is what you’ll actually be paying in two years. Four lines to ask about.

Model usage. The AI models themselves charge per use. For most small business automations this is genuinely small: pennies per document, a few pounds to a few tens of pounds a month. It only gets large when you’re processing thousands of items a day or using the most expensive models for jobs that don’t need them. A good builder picks the cheapest model that does the job reliably and will show you the maths.

Tool subscriptions. The automation platform, the hosting, any API you’re connecting to. Typically £20 to £150 a month for a workflow or an agent. More for a platform with real users.

Maintenance. Software rots. Suppliers change their APIs, the model you built on gets retired, your accounts package updates its login. Budget something: either a small monthly retainer or an understanding that you’ll pay for fixes as they come. Anyone who tells you it’ll run forever untouched is either new or lying.

Your time. Someone in your business has to check the approvals, answer the questions during the build, and tell the builder when it’s doing summat daft. This is the cost most often underestimated, and it’s the one that decides whether the project works.

What makes the price go up

• Systems without an API. If the thing you want automated lives in a portal, a desktop program from 2011 or a supplier’s extranet, the job involves driving the software by screen, which is slower to build and more fragile to run. Doable now, but it moves you up the range.
• Messy inputs. Handwritten forms, inconsistent spreadsheets, PDFs that are really photographs. The AI can cope; the edge cases take time to find.
• Consequences. Anything that touches money, legal commitments or personal data needs more testing and a proper approval step. Rightly.
• Scope creep. “While you’re in there” is the most expensive phrase in software. Fix the scope, build it, measure it, then do the next thing.

What makes it go down

• Clean data in a system with an API. Xero, HubSpot, Google Workspace, Microsoft 365, most modern SaaS. Quick to connect, quick to build.
• One clear job. The narrower the better. “Draft replies to enquiries” is a week. “Run our sales process” is a quarter.
• A decision-maker who’s available. Projects stall waiting for answers, not waiting for code.
• Starting with the right job. The most valuable automation is usually the most boring one: the thing someone re-types from one system into another every day. It’s also the cheapest to build.

Is it worth it?

Do the sum yourself, because it’s not complicated. Count the hours a week the job currently takes. Multiply by the loaded cost of the person doing it. Multiply by 48. Compare it to the build cost plus a year of running cost.

If a £4,000 automation saves eight hours a week at £25 an hour, that’s about £9,600 a year, and it’s paid for itself by May. If it saves two hours a week, it hasn’t, and we’ll tell you so. Not every job is worth automating, and a quote that doesn’t say which is a quote that hasn’t been thought about.

The honest summary: a single well-chosen automation costs about what a decent second-hand van does, and like the van it needs tax and a service. If you’d like a number for your specific job rather than a range, tell us what the job is and we’ll give you one, free, with no discovery call required.