Insights · 8 min read
What it actually is, what it genuinely costs, what it can and can't do — and how to make a good decision about whether it's right for your business.
If you run a UK small business and you've spent any time looking at AI automation in the last six months, you've probably been told three contradictory things by three different people in three different sales calls. Most of them were trying to sell you software. None of them asked what your team actually does on a Tuesday afternoon.
This is the version of that conversation written down. What AI automation actually is, what it costs for a real UK small business, what it does well, what it can't do, and how to make a sensible decision about whether it's worth pursuing. No jargon, no hype, real numbers.
AI automation is not a product you buy. It's not a chatbot you add to your website, a dashboard that tells you things you already know, or a subscription that promises to save you hours before you've actually used it.
For a small business, AI automation is the process of identifying the repetitive, time-consuming work your team does every day — the same steps, the same data, the same outputs — and building systems that handle those tasks automatically. The AI part is what allows those systems to deal with variation: a booking enquiry worded differently each time, an invoice that doesn't quite match a purchase order, a review that needs a personalised response rather than a template.
The most useful question isn't "how do I use AI?" It's "which specific tasks is my team doing manually that a system could handle more reliably, faster, and without someone's attention?"
For most UK small businesses, the answer to that question sits somewhere in: booking and enquiry management, invoice processing, supplier communication, reporting, scheduling, or customer follow-up. Not usually in customer-facing conversations, creative work, or decisions that require judgment about unusual situations.
The honest version of this list is shorter than most AI companies would have you believe.
What automation handles well:
To make that concrete: take a 12-person accounting practice. Every quarter end, two people spend three days each chasing missing receipts, manually matching invoices, and emailing clients for clarification. That's 96 hours per quarter — 384 hours a year. At a senior accountant's hourly cost, you're paying roughly £15,000 a year for highly trained people to do data-chasing. Automation handles the chasing, the matching, and the routine clarification emails. The £15,000 stays in the business and the senior accountants get back to the work clients actually pay them for.
That kind of compounding time loss — small daily tasks adding up to hundreds of hours annually — is where AI automation pays back fastest. Where it doesn't pay back is the work that requires actual judgment.
What automation doesn't replace:
Most AI consultancies won't publish a price. Here is ours, because transparency is how you make a good decision.
Discovery — £1,500 fixed. A week of mapping how your operation actually works, identifying where time and money are leaking, and producing a written roadmap of what's worth automating, in what order, and what it will cost to build. Includes a fixed-price quote for implementation. The Discovery fee is refundable if you proceed to implementation.
Implementation — £3,000 to £20,000. Fixed price, agreed upfront. Most UK small businesses fall in the £5,000–£12,000 range. The variation depends on how many systems need connecting, how complex the logic needs to be, and how much data needs to flow between tools. A single-process automation (for example, booking enquiry handling) sits at the lower end. A multi-process system connecting several tools across an operation sits at the higher end.
Ongoing costs — under £200/month for most builds, covering the API costs of running the systems. You own everything we build — no monthly licence fee, no vendor lock-in. If you want to modify the system later, you can do that yourself or hire anyone to do it. The code is yours.
For comparison: a part-time admin assistant at the UK National Living Wage costs roughly £13,000 per year before employer NI and pension. A full automation system that handles the equivalent workload costs £5,000–£12,000 to build and under £2,400 per year to run. The full cost comparison is here if you want to run the numbers for your business.
There are five questions worth asking honestly before committing to a build.
1. Is there a real, repetitive task eating time? Automation compounds. A task that takes 30 minutes a day is 130 hours a year. At a senior staff member's day rate, that's £3,000–£5,000 in labour cost. If you can't identify a task that costs at least that much, the maths probably don't work yet.
2. Is the process consistent enough to automate? If the same task happens differently every time — different people, different rules, no clear pattern — automation will fight you. The process doesn't need to be perfect, but it needs to be stable. Automation can improve a messy process; it can't replace a process that doesn't exist.
3. Do you have the data in a usable format? Automation connects systems. If the data you need is in people's heads, in WhatsApp messages, or in paper records, the first step is getting it somewhere a system can access it. That's not a barrier to automation, but it affects the scope of the build.
4. Does someone have the authority to make this change? Automation changes how people work. If implementing it requires buy-in from multiple people and nobody has the authority to make the call, you'll spend the budget on planning conversations rather than building systems.
5. Is the business stable enough to handle a build period? A build takes 2–6 weeks. If you're in the middle of a restructure, a rebrand, or a significant operational change, wait until the process you're automating is settled. Automating a process that's about to change is expensive.
If you can answer yes to the first two and have a reasonable answer to the rest, it's worth having a conversation. If you can't — if you're not sure where the time is going — that's what a readiness assessment is for.
The businesses that get the most from AI automation tend to share a few characteristics: they have a team of 3–20 people, they've been operating long enough to have settled processes, and they're doing the same types of tasks every week in a volume that makes manual handling genuinely costly.
In practice, across UK small businesses, this tends to mean:
If your business doesn't fit these categories but you have staff doing repetitive admin work, the conversation is still worth having. The industries above are where we have the most experience — but the principle of automating repetitive work applies to any business.
This is where most guides stop being useful, because most guides are written by the people selling the service. Here is the honest version.
Ask any consultant you're considering these questions before you commit to anything:
A consultant who can't answer the first two clearly, or who builds systems you don't own, is selling a dependency rather than a solution. The market has enough of those.
At On The Hill AI, the answers are: yes, fixed-price, you own it, anyone can modify it, and yes — ten years managing hospitality and logistics operations before building automation systems. That last one matters more than it sounds. The systems we build work in production because we've been in the operations they're designed to support.
It depends on the volume of repetitive work, not the headcount. A restaurant with six staff processing 50 booking enquiries a week has a stronger case for automation than a 20-person professional services firm whose admin is genuinely varied and low-volume. The question is whether there's a consistent, repetitive task costing enough time to justify a build. If there is, the size of the business doesn't determine whether automation is worth it.
Discovery takes approximately one week. Implementation takes 2–6 weeks depending on complexity. A single-process automation — for example, booking enquiry handling or invoice matching — can be live within 2–3 weeks of starting the build. More complex multi-system integrations take 4–6 weeks. You'll know the exact timeline before any build work begins.
Off-the-shelf tools like Zapier and Make are excellent for simple, linear automations: if this happens, do that. They're fast to set up and cheap to run. Custom AI automation is appropriate when the task involves variation, judgment, or context — when a canned response won't do and when the process doesn't fit neatly into a standard workflow tool. Most small businesses benefit from both: off-the-shelf tools for simple tasks, custom builds for complex ones.
Not the people — the tasks. AI automation replaces the repetitive, time-consuming work that stops your best people from doing the work that actually grows the business. A restaurant manager who stops manually chasing booking confirmations doesn't lose their job — they get their time back for customer relationships, team development, and the operational decisions that need a person's judgment.
For most small business builds, ongoing running costs are under £200 per month — covering the API costs of the AI services the system uses. You own the system outright, so there's no ongoing licence fee or consultant retainer required to keep it running. Costs scale with usage: a system processing 100 items a month costs less to run than one processing 10,000.
No. Custom AI automation is built to work with whatever tools you already use — your booking system, your accounting software, your email, your spreadsheets. The automation sits between your existing tools and connects them, rather than replacing any of them. You don't need to change platforms or retrain your team on new software.
Book a free 30-minute call. I'll ask about your operation, where the time is going, and what you've already tried. Then I'll tell you honestly whether automation makes sense — or whether the answer is wait, hire, or fix the process first. No pitch, no pressure.
Book a free call