How much does AI automation cost a UK small business

For most UK small businesses, AI automation costs sit in one of three bands. Off-the-shelf AI assistants cost roughly £14 to £25 per person per month: Microsoft 365 Copilot Business, for example, is £19.32 per user per month on monthly billing, excluding VAT. Connecting AI into the tools you already use through an automation platform adds a modest monthly subscription plus setup work, usually a few days of someone's time. A bespoke build, where software is written around your specific workflow, starts in the low thousands of pounds and rises with the number of systems it has to touch. UK government research found that businesses adopting AI spent a median of £2,000 on it in 2024, against a mean of £19,000: most adopters spend a little, a small minority spend a lot. The most reliable way to keep your project in the first group is to scope it around one painful task, prove it pays for itself, then expand.

"If you cannot articulate why a subscription or a platform assembly fails for your case, you are not ready to commission a build."

Dean Cookson, founder, Operosus

Why is it so hard to get a straight answer on price?

Because "AI automation" describes everything from a single Copilot licence to a custom-built system that runs your entire intake process. Anyone quoting one number for that whole range is guessing, and three things make the guesswork worse.

First, vendors price what they sell. A software company will tell you a subscription solves it. An agency will tell you it needs a build. Neither is lying, but neither is neutral either.

Second, the expensive part is rarely the AI. Model access is cheap and getting cheaper. The cost lives in the integration work: connecting the AI to your calendar, your CRM, your payment provider and your inbox, and handling all the messy cases where real businesses differ from the demo.

Third, most businesses budgeting for AI are doing it for the first time. The same government research found that among businesses planning to adopt AI within twelve months, 47% had no specific budget allocated. When buyers have no anchor, quotes drift towards whatever the seller thinks the market will bear.

The fix is to understand the three ways you can pay, and which one your problem actually needs.

What are the three ways to pay for AI automation?

What do the subscriptions actually cost?

This tier is the easiest to price because the numbers are public. Microsoft 365 Copilot Business costs £19.32 per user per month billed monthly, excluding VAT, with a discounted annual rate from £13.80 on top of a qualifying Microsoft 365 plan. Claude's Team plan is $25 per seat per month billed monthly, or $20 billed annually. For a five-person business, the whole tier costs less per month than a single hour of most consultants' time, which is why it is the right first step for almost everyone.

What it will not do is run a process for you. An assistant helps the person; it does not replace the workflow.

What does the platform tier cost?

The platform tier connects what you already have. A typical example: a lead arrives through your website form, an AI model classifies and enriches it, the result lands in your CRM, and a tailored reply goes out within minutes. The components are an automation platform subscription, pay-as-you-go AI model usage that for low volumes is usually pennies per run, and the setup work itself.

Setup is where the real cost sits, and it is also where the trap sits. The monthly fees look trivial, so businesses underestimate the project. Budget for the time to map the process properly, build it, test it against real data and handle the exceptions. Done in-house by a capable operations person, that is days of focused work per workflow. Done by an agency, it is a small project fee. Either way, it is finite and it ends with a process that runs without you.

When is bespoke worth paying for?

A bespoke build makes sense when the workflow is the business. If your competitive edge is how fast you respond to enquiries, how smoothly you onboard clients or how efficiently you produce a deliverable, owning that system outright can justify a five-figure investment. The £19,000 mean against a £2,000 median in the government's adoption data tells you exactly who sits in this tier: a minority of businesses making a deliberate, larger bet, pulling the average up while most adopters spend far less.

If you cannot articulate why a subscription or a platform assembly fails for your case, you are not ready to commission a build. That is not a criticism. It is the cheapest piece of advice in this guide.

What are UK businesses actually spending?

The honest picture from primary sources, not vendor surveys. We keep all of these in one maintained table of UK small business AI statistics:

• The ONS found 23% of UK businesses were using some form of AI by late September 2025, up from 9% two years earlier.
• Adoption skews heavily by size: government research puts it at 36% of large businesses against 14% of micro businesses.
• Cost is a real barrier, not an imagined one: 76% of businesses rated high costs as a significant barrier to adoption in the same research.
• Among businesses that have adopted, 75% reported workforce productivity improvements, while most saw no direct revenue change, which fits how this technology earns its keep: it removes cost and time before it adds sales.

Read together, the data says the typical successful adopter is spending hundreds to low thousands of pounds a year, getting productivity gains rather than instant revenue, and expanding from there. That is a sensible benchmark for your own first budget.

What actually drives the cost of a build?

When a bespoke quote comes in higher than you expected, one of these five drivers is usually the reason:

1. The number of systems involved. Each tool the automation has to read from or write to (CRM, calendar, payment provider, accounting software, phone system) adds integration work, and ageing or niche software adds more than modern tools with good APIs.
2. The state of your data. If your customer records are inconsistent, duplicated or spread across spreadsheets, cleaning them up becomes part of the project whether anyone names it or not.
3. The cost of being wrong. An automation that drafts internal notes can fail safely. One that sends payment links to bereaved customers or commits you to a contract needs review steps, fallbacks and alerting, and those guardrails are real engineering work.
4. Edge cases. The happy path is always quick. The expense is in what happens when the customer replies in an unexpected format, the third-party API goes down or two bookings collide.
5. Who maintains it. Models change, APIs change and your business changes. A build with no maintenance plan is a build with a hidden second invoice.

The expensive AI projects are the ones that start with the word "AI" and go looking for a problem. Scope around a task you already pay for in staff time, and the budget conversation almost answers itself.

That is the house view at Operosus, and it is the single filter that separates the £2,000 projects that work from the £19,000 ones that drift.

Which running costs catch people out?

The project fee is not the whole bill. Four recurring costs to put in the spreadsheet on day one:

• Usage fees. AI models are priced per use. At small business volumes this is usually small, but it scales with success, so model it at the volume you hope to reach, not the volume you have.
• Subscription stacking. Each platform, tool and add-on is cheap alone. Audit the stack quarterly, because dead subscriptions accumulate quietly.
• Maintenance. Connected systems break when any one of them changes. Budget a small ongoing allowance for fixes and improvements rather than treating every breakage as a surprise.
• Human review time. Most adopters keep a person checking AI output, and 84% of businesses using AI maintain at least some human checking. That time is a genuine cost, and a worthwhile one. Price it in rather than pretending the system is unattended.

What does this look like on real projects?

Three patterns from our own work show how the tiers play out in practice.

Productising a repeated task. Bidwell, our AI tender-writing product, exists because drafting bid responses is the same structured job for every small firm that does it: hold your real evidence in one place, draft answers against the buyer's questions, keep a human signing off every claim. Because the task is common across thousands of businesses, it belongs in the subscription tier, priced per month rather than built per client. If your painful task is one that many businesses share, check for a product before you commission anything.

Automating a booking flow. For Vets at Home, a home-visit veterinary service, the work was wiring the journey from website enquiry through to a confirmed, paid, correctly attributed booking, with the right notifications reaching the right vet. No single product does that end to end, but most of the components already existed. This is the platform tier: integration work on top of existing systems, where the value is a process that runs reliably without anyone chasing it.

Matching supply to demand with data. For Vivify, which helps schools hire out their facilities, the pattern was using structured data about venues and the organisations that hire them to connect the two sides automatically. That is closer to a bespoke build, because the matching logic is the business, not an accessory to it.

The common thread: the tier follows from the shape of the problem, not from ambition or budget size.

Where to start

Do this in order, and let each step earn the next:

1. List the tasks you pay for in staff hours. Quoting, chasing, booking, onboarding, reporting. Put a rough weekly hour count next to each.
2. Start one or two people on an assistant subscription. At around £14 to £20 per user per month this is the cheapest possible test of whether AI helps your actual work.
3. Pick the single most repetitive task from your list and price the platform tier for it: one workflow, one clear before-and-after measure.
4. Only consider a bespoke build once an automated workflow is already paying for itself and you can name precisely why off-the-shelf is the constraint. Our guide to choosing an AI agency covers how to test the people quoting you.
5. Hold the spend benchmark in mind. The median UK adopter spent £2,000 in a year. If your first quote is ten times that, the burden of proof is on the quote.

If you want a second opinion on which tier your problem belongs in, that conversation is short, and we are happy to have it before any money moves. How we work sets out exactly what an engagement with us looks like, in plain words.