Where the Money Leaks: An AI Teardown of a Typical £2m-£20m Business

The Arithmetic Behind Every Number Here

First, the honesty rules. Nothing below is a case study. There are no named clients and no precise outcomes dressed up as data. These are worked examples for a typical business in the £2m-£20m bracket, and the point of showing the sums is that you can substitute your own figures line by line. Where we do have published results, we link to them.

Second, the cost basis. Manual work is priced at fully loaded cost, not salary. A £32,000 administrator costs the business roughly £40,000 once employer National Insurance, pension and overheads land, and delivers about 1,600 productive hours a year. That is £25 an hour. A £48,000 finance or operations manager works out at roughly £37.50; call it £40 for round numbers. So the two rates used throughout are £25 an hour for admin work and £40 an hour for finance and management work. If a task takes five hours a week, it costs the business that rate times five times 52, whether anyone has ever put it on a budget line or not.

Third, what “AI replaces” means. Not people. The pattern that works in 2026 is that AI produces the first draft, the match, the chase or the summary, and a person reviews it. The saving is the gap between doing the work and checking the work, which for high-volume, format-driven tasks is usually 50 to 80 per cent of the hours.

Finance Operations: The Quietest Leak

Finance is where the manual hours hide in plain sight, because the work is deadline-driven and nobody questions it. Three lines dominate.

Purchase invoice processing. Picture a 50-person distributor handling 250 supplier invoices a month. Keying, coding and matching each one takes six minutes on a good day: 25 hours a month, 300 hours a year. At £25 an hour that is £7,500 a year spent retyping documents the supplier already typed. AI extraction and matching now handles roughly 80 per cent of invoices without human touch, leaving exceptions for review: about £6,000 a year back, plus fewer duplicate payments. We walked through a real version of this pattern in our post on automating invoice processing between Excel and Xero.

Month-end close. A financial controller spending four days a month on close, half of it drafting reconciliations, schedules and variance commentary from scratch, is normal in this bracket. AI drafting compresses that drafting half by around two days a month: 24 days a year at a loaded day rate of about £300 is £7,200 of senior finance time returned to analysis. This is not hypothetical for us: see our month-end close automation case study and the follow-up audit of the results.

Credit control. Five hours a week of chasing debtors is 260 hours a year, £6,500 at admin rates; AI-drafted, prioritised chase sequences halve it. But the hours are the small prize. The big one is cash. A £5m-turnover business that pulls debtor days from 55 to 45 releases roughly £137,000 of cash (£5m divided by 365, times ten days). At 7 per cent borrowing costs, that is about £9,600 a year in interest you stop paying, from chasing consistently rather than when someone remembers.

Sales, Quoting and Retention: The Revenue Side

Cost reduction is only half the brief. The same drafting-and-review pattern that cuts admin hours also puts money on the top line, because the constraint on most SME sales operations is not talent, it is response time and follow-through.

Lead follow-up. A business taking 40 inbound enquiries a month typically gives each one reply and then goes quiet, because follow-up is nobody’s whole job. AI-drafted follow-up sequences, reviewed and sent by a person, mean every enquiry gets chased properly. Model it conservatively: if disciplined follow-up converts just two extra deals a year at a £15,000 average order and 30 per cent gross margin, that is £9,000 of contribution. Our guide to automating email follow-ups with AI includes the workflow.

Quote turnaround. If a quote takes three days to assemble from price lists, drawings and past jobs, AI that drafts it from the same inputs in an hour changes your win rate, because the first credible quote often frames the buyer’s decision. On 300 quotes a year at £8,000 average order value, moving the win rate from 25 to 27 per cent is six extra orders: £48,000 of revenue, around £14,000 of margin at 30 per cent.

Retention. A business with £1m of recurring or repeat revenue losing 8 per cent a year is losing £80,000. AI-monitored account signals (slowing orders, unanswered tickets, sentiment in correspondence) catch a share of that while it is still recoverable. Save a quarter of it and at 60 per cent margin on serviced revenue that is roughly £12,000 a year retained.

Released capacity. In professional services firms the arithmetic is starkest. Four hours a week of admin per fee earner, across five fee earners, is about 920 hours a year. Turn even half of it back into work billed at £100 an hour and you have £46,000 of billable capacity that was being spent on formatting, filing and status updates.

Customer Service

Most service inboxes in this bracket are 60 to 70 per cent repeat questions: where is my order, can I get a copy invoice, how do I return this, what does this error mean. A team handling 600 tickets a month at eight minutes each is spending 80 hours a month, 960 a year, about £24,000 at admin rates. AI that drafts responses from your own help content and order data, with a person approving anything non-routine, reliably takes 40 per cent of that out: about £9,600 a year, and the answers go out in minutes rather than the next morning, which feeds the retention line above. The failure mode to avoid is the unsupervised chatbot; the pattern that works is draft-and-approve, the same as everywhere else in this teardown.

Reporting and Management Information

Every month, someone assembles the board pack, the sales report and the ops dashboard by copying numbers between systems and writing the same commentary structure around them. Ten hours a month at management rates is £4,800 a year, and that is a modest estimate for a business running weekly trading reports as well. AI that pulls the figures and drafts the narrative in your house format, for a person to edit and sign, removes around 70 per cent of it: about £3,400 a year. The bigger win is harder to price: reporting that used to arrive on the 15th arriving on the 8th means decisions get made a week earlier, twelve times a year. Our roundup of AI use cases for finance covers the reporting workflows in more depth.

Want this teardown done on your actual numbers? See the AI Opportunity Blueprint

Admin and Scheduling

The residual category is the one owners underestimate most because it is spread across everyone: scheduling meetings, taking and circulating minutes, chasing documents, re-entering data from one system into another, formatting proposals. Across a 20-200 person business it is rarely less than eight hours a week in aggregate, and usually far more. Eight hours a week at £25 is £10,400 a year; AI note-taking, drafting and structured data entry takes half of it: about £5,200. Small on its own. But this is also the category where the tools are cheapest and adoption is easiest, so it is often where we start a rollout: quick wins here buy the team’s trust for the finance and sales workflows that carry the real money.

Adding It Up

Here is the whole teardown on one table. Illustrative, arithmetic shown above, your numbers will differ.

No single business hits every line. A distributor has no billable capacity; a services firm processes few supplier invoices. Strip it back to the lines that plausibly apply to you and a realistic range for a business in this bracket is £40,000 to £80,000 a year, roughly half from cost taken out and half from revenue added. And notice the asymmetry: even if you cut every figure here by two thirds, you still clear £10,000 comfortably. That is not an accident. It is why we can guarantee it.

This Teardown, On Your Numbers

Everything above is generic by design, because publishing invented specifics helps nobody. The useful version of this exercise is the one done inside your business: your invoice volumes, your debtor days, your enquiry-to-quote-to-order funnel, your team’s actual hours, ranked by payback and sequenced into a plan.

That is exactly what the AI Opportunity Blueprint is: this teardown, run on your operation with your real numbers, in 14 days, for a fixed £999. It comes with the promise that shapes everything we have written here: we identify at least £10,000 in annualised savings, or you don’t pay. If you would rather have the workflows built for you afterwards, our AI consulting engagements do the implementation, and AI for finance training gets your team running the tools themselves.

Frequently Asked Questions