Every AI conversation I have with an owner lands in one of two ditches. The first is hype: someone on a stage has promised them a 10x business if they just "embrace AI," no arithmetic supplied. The second is fatigue: they've spent eighteen months and a drawer full of subscriptions discovering that nothing 10x'd, and they're quietly done with the subject.
Both ditches have the same cause. Nobody did the math.
So let's do it — on one page, with numbers you're welcome to challenge.
Start With What a Head Actually Costs
Your people cost more than their salaries. Take gross pay, add employer CPF at 17%, bonuses, insurance, the software stack each person carries, the laptop, the recruitment fee amortised over their tenure, and the share of rent keeping a desk under them. For a Singapore services business — agency, studio, consultancy, professional practice — a blended S$6,000 per head per month fully loaded is a fair planning figure. Your senior people cost double or triple that; your juniors cost half. Blended across the team, S$6,000 is honest.
A 50-person firm is therefore carrying roughly S$300,000 a month — S$3.6 million a year — of human capacity. That's the single largest asset most services firms own. Almost nobody manages it like one. If you've never computed your own blended figure, it's an afternoon's work with last quarter's payroll report — and it's worth doing before you read the next section.
The Only Other Number You Need
A 15–20% capacity unlock is the credible range. Not 50% — that's the conference number. Watch where a services team's hours actually go: status updates, meeting preparation, first drafts, reconciliations, research summaries, formatting, version-chasing. That routine layer is comfortably a fifth of most professional jobs, and often more. AI deployed properly — against measured processes, not as a toy — returns most of that layer to the business.
Note what this is not. It isn't headcount removed. It's capacity returned: hours that flow back into pitching, delivering, and selling.
Multiply those two numbers together and you get a table most owners have never drawn:
At the 50-person scale, that's the productive capacity of seven to ten additional full-time hires — with no recruitment fees, no onboarding ramp, and not one new desk. This is the same arithmetic we model when scoping a programme for a mid-sized Singapore services firm; every figure it commits to is a target to be measured against monthly, not a result anyone is claiming. If your team is twenty people, the numbers scale down. The logic doesn't change.
Where the Value Actually Shows Up
The reason owners miss AI value isn't that it doesn't exist — it's that they look for it on the wrong lines. The unlock rarely appears as a deleted cost. It appears in four places, and only one of them sits anywhere near the expense rows.
Cycle time. Month-end close is the cleanest example: a real baseline of 8 days across 600–700 transactions, targeted down to 2–4. Every day saved means management sees the numbers earlier — and every decision downstream of those numbers moves earlier too.
Rework. In the creative and professional teams I've sat inside, rework quietly consumes 15–25% of delivery capacity — briefs that were wrong, context that went missing, versions that multiplied. AI skills that integrate inputs across functions attack the cause, not the symptom.
Win rate. A firm pitching four to six times a month at roughly a 45% win rate has an enormous lever hiding in small numbers. A few points of win rate — or simply the senior capacity to fight one more pitch a month properly — is worth more than every subscription you've ever bought. AI doesn't win the pitch. It returns the senior hours that do.
Close speed. Not just the books. Quotes out faster, proposals out faster, client questions answered in minutes instead of meetings. Speed compounds invisibly — which is why nobody budgets for it and everybody feels it.
Why Your CFO Should Own This Number
If AI value lives in capacity, cycle time, and dollars, then it lives in finance's native language — and it should be measured by the function whose entire job is measurement.
Finance already owns the machinery. Baselines, targets, variance against model, a monthly cadence — this is ordinary budget discipline pointed at a new asset class. No new ceremony required.
Whoever claims the savings shouldn't be selling the tool. Vendor dashboards measure usage; usage is not value. The P&L is the one scoreboard that can't be flattered.
A published model creates accountability in both directions. Write it down — 15–20%, S$6,000 a head, your actual headcount — and measure against it monthly. If the unlock is real, it shows. If it isn't, you stop paying for it. Either outcome beats hope.
One more line for the Singapore version of the model: a properly scoped transformation programme can qualify for up to 50% EDG co-funding — which halves the denominator of every return you've just calculated. That trap-laden process deserves its own essay, but the headline belongs in the math.
The hype number is whatever the stage speaker said. The fatigue number is zero. The real number has been sitting in your payroll ledger all along, multiplied by a percentage you can test inside ninety days — or in the next two minutes, if you run your own numbers on the capacity model.
Hype is a multiple somebody quotes you. Capacity is a number you can audit.