.png)
Where to begin when the board says “AI” and means “cut costs”.
There’s a scene playing out in boardrooms up and down the country. Someone on the board has been to a conference.
They’ve seen a slide with a robot on it. Now the minutes say “explore AI-driven efficiency opportunities”, and the responsibility usually sits with the Head of Finance.
The brief, translated from board into English: automate something, with AI if at all possible, and make a cost go down. By Q3 would be lovely.
You nod. You go back to your office. You look out at a business with four hundred processes, a dozen systems and a spreadsheet with fourteen tabs that nobody dares open, and you ask the only sensible question: where on earth do you begin?
That question is the whole problem. Not “whether”; the answer to that sits somewhere between “probably” and “obviously”. Not “how”; there are firms for that, and we’re one of them. The hard part is “which”. Which task first, how much to spend on it, and what to tell the board when they ask what they got for the money.
Here’s the good news. You already own the answer. “Where to begin” isn’t a technology question. It’s an investment question, and those are yours.
Stop looking for AI. Start looking for a number.
The most expensive way to begin is with the technology. The AI strategy deck. The transformation programme with its own logo.
The chatbot somebody’s nephew can knock up over a weekend. These all start from “what’s exciting” and hope a saving turns up later. I
t rarely does, and when the board asks what the money bought, the answer is usually vague and hidden in an unimpressive slide deck.
Flip it round. An automation, AI-powered or otherwise, is a capital decision like any other.
It replaces a cost you’re paying today, it returns something, it pays back over some period, and there’s a sensible ceiling on what to spend.
Seen that way, “where to begin” becomes a sorting problem: take the candidate tasks, work out what each costs you now,
work out what a realistic automation would hand back, rank them, start at the top.
The metrics below are the sorting key. “AI” is the word the board uses. “Payback” is what they mean. You’re the translator.
Metric 1: the cost of doing nothing
Every investment case starts with a baseline, and the baseline here is what the task costs you today. Three ingredients:
-
Heads. How many people touch the task. Not “Sarah does it”, because Sarah is on holiday in August, and then it’s Sarah’s colleague, Sarah’s manager and one very confused temp.
-
Hours. How long each of them spends on it per week. “Half an hour” is almost never half an hour once you count the checking, the re-checking, and the strong cup of coffee they need before starting.
-
Hourly cost, fully loaded. Salary plus employer’s National Insurance, plus pension contributions, plus the desk, the laptop, the licence and the biscuits. Take-home pay is what your staff think they cost. Fully loaded is what you know they cost. Home turf.
Multiply the three together, then multiply by working weeks. We use 46, not 52. Holidays exist, flu exists, and a business case built on 52 weeks is one you’d take apart in your sleep.
Napkin example: invoicing, two to three people, a few hours a week each, at £25 to £40 an hour fully loaded.
About £320 a week, or £15,000 a year, into a task that isn’t on any budget line. It never shows up on the P&L because it’s spread across several salaries.
It’s still £15,000, and there are a dozen tasks like it.
Metric 2: the cost of being human
Humans make mistakes. Every mistake gets paid for twice: once to make it, once to find it and fix it.
A mis-keyed invoice becomes a credit note, a reissue, a phone call, and a supplier who now checks everything you send them.
Count the time spent fixing errors as part of the task’s cost, because it is. In the invoicing example, one weekly slip that takes an hour and a half to unpick adds about £2,200 a year, so the true figure is nearer £17,000. Robots don’t fat-finger invoices. They have other flaws, but not that one.
One honest caveat: our calculator counts the labour of fixing mistakes, not the direct losses, like the refund you shouldn’t have paid or the penalty for a late filing. If your task has those, your real number is higher than ours. We’d rather understate the case than oversell it.
Metric 3: the automatable share
Nothing is 100% automatable. There’s always an exception, an edge case, a supplier who sends invoices as photographs of invoices.
A realistic automation removes about three quarters of the work. We use 75% because that’s what our own projects typically deliver, and if a vendor promises 100%, ask them how automated their own expense claims are.
So of that £17,000 a year, roughly £12,750 is what a sensible automation hands back.
Metric 4: payback, and why we don’t say “ROI”
Here’s the rule of thumb our calculator is built on: an automation is worth doing if half the investment comes back within the first year.
Notice what we didn’t say. We didn’t say “50% ROI in year one”. You know exactly what that phrase means, and it means something considerably stronger: half the investment as profit, on top of getting the money back.
We’re not claiming that, and we’d rather you caught us not claiming it than caught us claiming it.
“Half back in year one” means the whole project pays for itself within two years, and from year three it’s simply earning. Unlike the company car, it doesn’t lose a fifth of its value the moment it leaves the forecourt. It does want the occasional service, and any honest supplier will say so.
Metric 5: the number you take back to the board
Flip the payback rule around and you get the number the board actually wants to hear: a ceiling. If the first year’s savings are about £12,750, and half the investment should be back by then, the most you should sensibly spend is twice that: £25,500. Call it £25,000; a ceiling is worth rounding down.
That’s not a quote. It’s the envelope. “We can spend up to £25,000 automating invoicing and it pays for itself within two years” is a sentence a board understands, approves and remembers. It also lets you hand the operations team a number and say “come back with something under this”, which is, let’s be honest, most of the job.
Metric 6: capacity value versus cash
The board will hear “saves a few hundred hours a year” and quietly think “saves a salary”. It doesn’t. Not by itself, and this is the bit your auditors would want said out loud.
Freed hours are capacity, not cash. They become cash-releasing only when one of three things happens:
-
the freed hours go into work that earns or saves money;
-
an actual expense shrinks: overtime, contractors, agency staff;
-
the next hire you were about to make quietly doesn’t need making.
In the United Kingdom, HM Treasury’s business-case guidance draws precisely this line between cash-releasing benefits and redeploying existing staff, and so does our calculator: it calls the savings “capacity value”, and it means it.
So decide, before go-live, which of the three it’s going to be, and write it down. That decision is the difference between a cost saving and a nice story about one. It’s also the thing to track afterwards: not hours saved, but where they went. Most projects forget this, which is why most boards end up asking “so where did the money go?”
Metric 7: the human ledger
Some costs don’t sit neatly in a cell. The task everyone dreads is the one that gets done late, done badly, and done by whoever lost the argument. It’s the task people procrastinate around; other jobs suddenly become fascinating.
And occasionally it’s the task people leave over. Nobody has ever handed in their notice because the ERP was too automated.
You know the cost of replacing someone in your business better than we do. We didn’t dare put it in the money maths, because a headline number that includes “morale” is a headline number nobody trusts.
So the calculator keeps the human stuff separate, in a report card graded A to F, and lets it do exactly one thing: tip a borderline verdict. If the numbers say “maybe” and your team’s faces say “please”, the answer is yes, and we say so openly.
Metric 8: the verdict, including “not this one”
Put it all together and each task gets one of three answers.
-
Automate. Yesterday. Expensive enough that a proper project earns its keep: half back in year one, all of it by year two, and then it keeps paying without sighing once. This is where you begin.
-
Borderline. Worth a conversation, and worth starting small. A modest fix now, built up later, beats a grand programme that never gets signed off.
-
Don’t automate. (Yet.) Sometimes the honest answer is “don’t bother.” Below a couple of thousand pounds, the overhead of running a project eats whatever you’d save - so we say so, and move on to the next thing on your list. An automation agency telling you not to automate. We’ve checked with our accountant. It’s fine.
So, where do you begin?
Not with the AI strategy. With one task. The one everyone groans about. Give it a number. Then the next task, and the one after that.
An afternoon of this and you have a ranked list: each task with a cost of doing nothing, a payback and a ceiling, and the one at the top is where you begin.
Take that list to the board. It’s the “AI cost-cutting plan” they asked for, and it’s built on the numbers you’d have used to argue with them.
Nine questions, about a minute, no email gate
We’ve turned all of the above into a calculator. Nine multiple-choice questions, about a minute of your time, and none of that “enter your email to see your results” business; the results are simply there.
It tells you what the task costs, how much you could sensibly invest in automating it with half of that back within the first year, and whether it’s worth automating at all. Start with the one with fourteen tabs. Then do the next one.