There’s work in your business that nobody is doing. You know roughly what it is, you’ve been putting off the decision about whether to hire for it, and somewhere in the last year someone told you AI could cover it instead.
On the default numbers in the calculator I built for that decision, an AI role covers 37% of the gap. A person covers 90%. I sell the AI role, and 37% is the number I chose to publish.
The useful part isn’t the 37%. It’s which 37%.
Coverage is not one number
Most of the hire-versus-automate argument is conducted at the wrong altitude. One side says AI handles it now. The other side says you can’t automate judgement. Both are talking about the gap as a single lump, and it isn’t one.
The calculator breaks the gap into 9 categories, from unchased invoices through to strategy drift, and scores each one separately for how much of it a named AI role can realistically own. Those scores aren’t close to each other:
| Category | What an AI role covers |
|---|---|
| Renewals and contract dates missed | 85% |
| Invoices sitting unchased | 80% |
| Software nobody uses any more | 80% |
| Meeting follow-ups that never went out | 70% |
| Speaking and visibility invites missed | 50% |
| Opportunities that went cold | 40% |
| Strategy drift | 30% |
| A bad hire a tighter process would have caught | 20% |
| Someone good leaving | 10% |
10% to 85% isn’t a rounding difference. It’s the whole decision.
The two shapes of work
Read that table again and the split is obvious. At the top: renewals, invoices, unused licences, follow-ups. Bounded work, with the evidence already sitting in a system, ending in an action anyone competent would take the same way. A date passes, a threshold trips, an email goes out. Nobody needs to be persuaded of anything.
At the bottom: a good person leaving, a bad hire, work drifting off plan. Unbounded work, with the evidence sitting in people’s heads and in the room, ending in a judgement somebody has to own and be wrong about in public.
The test isn’t how hard the task is. Chasing 40 overdue invoices is more work than deciding to let someone go. The test is whether the task ends in a judgement with a name against it. That line predicts the coverage score better than complexity, volume, or how technical the work sounds.
The inversion that decides it
Here’s the part that changes the answer, and it’s the reason “just automate it” keeps disappointing people.
On the default gap in the calculator, the three lowest-coverage categories are the three most expensive. A bad hire, someone good leaving, and strategy drift come to £47,500 of a £102,100 gap. That’s 47% of the money sitting in the rows an AI role barely touches. It covers £8,850 of that £47,500.
Meanwhile the rows it’s genuinely good at, the 70% to 85% band, are worth £15,600 between them. It covers about 80% of those, and 80% of a small number is a small number.
The work an AI role is best at is usually the work costing you least. Not always, and the defaults are only defaults, which is why the tool lets you rewrite every row. It’s the common case though, and it’s why founders install something that works exactly as advertised and still feel like nothing changed. The automation was real. It was pointed at the cheap half of the problem.
What that does to the money
The cost comparison isn’t close, and it doesn’t settle anything.
An AI role, year 1: £6,500 to install, £2,500 a year of care, about £1,200 in model and tool subscriptions. £10,200 all in, dropping to £3,700 in year 2 once the install is paid for.
A £100,000 hire, year 1: salary, plus 13.8% employer national insurance, plus 4% pension, plus a 20% recruitment fee, plus £4,000 of kit and overheads. £141,800 before anyone has done any work. Year 2 settles at £117,800.
That’s roughly 14 times the cost for roughly 2.4 times the coverage, which reads like a straightforward win for the AI role until you carry the residual through. Against that £102,100 gap, the AI role leaves £64,320 uncovered, so the true year-1 position is £74,520. The hire leaves £10,210 uncovered and costs £152,010 all in.
Both are worse than the £102,100 you’re already losing by doing nothing.
That’s not a typo and it isn’t a sales argument. On those defaults, a £100,000 hire against a £102,000 gap doesn’t pay for itself in year 1, and neither option is obviously right. It becomes obvious in either direction once your own numbers go in: a gap concentrated in retention and strategy justifies a person quickly, and a gap concentrated in admin and money-in never will.
3 questions that sort your own gap
Does the work end in a judgement someone has to own? If a real person has to be accountable for the call, and be wrong about it in front of others, that’s a hire. An AI role can hold the rubric, prepare the evidence, and tell you the decision is due. It can’t be the one who decided.
Is the evidence already in a system? Invoices, contract dates, licence usage, calendar and inbox: all of it is already written down somewhere an agent can read every day without being asked. Work whose inputs live only in conversations is work you can’t install yet, whatever the demo showed you.
Would you notice within a week if it stopped? This one decides whether either option survives contact with reality. A hire announces their own absence. An installed workflow doesn’t, so if the answer is no, fix the noticing before you spend anything. I learned that the expensive way on my own install, and the answer for 2 of my 3 workflows was no.
Why the low numbers are low on purpose
The people-shaped rows are scored at 10%, 20% and 30% because that’s what I’ve watched them deliver. It would have been easy to write 60% into those cells. The tool sits on my own site, it recommends my own product, and nobody would have checked.
Inflating them would also have made it a sales prop, and a sales prop is worth nothing to the person trying to make the decision. It is the same reason the hire is credited with 90% rather than something meaner, and the reason the confidence bands stay wide on the rows I am guessing at. Every row is also tagged with whether its figure traces to published UK data or to my own engagements, so a sceptic can see which half to verify and which half they are taking on trust.
The honest version of my pitch is narrow. An AI role reliably owns the bounded, evidenced, judgement-free half of your gap, it costs about a fourteenth of a hire, and on most founders’ numbers that half isn’t where the money is. If the expensive part of your gap is people-shaped, hire the person. I would rather you worked that out from my own tool than from a year of paying me to be the wrong answer.
Run your own numbers, and change every row until they’re yours: the Cost of Not Hiring calculator.
Join the Drift Digest
One short email when a new post lands, plus new AI tells as models change and a quarterly nudge to re-measure your voice. No spam, unsubscribe anytime.
Something went wrong. Please try again, or email web@ctoondemand.co.uk.