AI–Human Balance Calculator

1. Enter what you know: people, cost per person, value per person, and the most demand you could serve. These come from payroll and finance — no guessing needed.
2. Enter the AI numbers honestly. The boost: take the vendor's claim and cut it roughly in half — claims are marketing until measured (it can even be negative; that has been measured in real studies). The AI cost: what AI would cost if it did all the work — a per-user licence price is not that number; a licence cannot staff an empty seat.
3. Set your plans: how many people the cut plan would keep and how much work AI takes in each plan. Model the real proposal on the table, not a strawman.
4. The one number nobody knows is "Value lost if AI ran everything, unchecked." Don't enter it once — run it three times: 15% (work is easily spot-checked), 40% (mixed), 60% (judgment-heavy). Click Download PDF after each run. Those three pages are your evidence pack.
5. Read the answers off the pages — you never compute anything yourself.
Two optional inputs sharpen the picture: "Cost to reverse a layoff" prices the one-way door — severance + rehiring + ramp, a rule of thumb is ~6 months of loaded cost per person; it appears in the table as the one-time bill if a cut proves wrong. "Freed capacity redeployed" matters only when you are demand-capped: it is the share of spare capacity that goes to new work instead of idling. Leave both alone and the tool behaves as before.

The strategy table: the Profit row tells you who beats today; Payback tells you when the upfront investment comes back (under ~24 months is healthy; "never" is dead money). The two big tiles answer the two questions every AI debate is really about. Trust the headcount answer only if it is the same on all three PDFs — an answer that flips between pages is a bet on the unmeasured number, and the right response to a bet is a measurement, not a decision.
Sanity rule: if any answer looks absurd — fire everyone, a five-digit ROI — that is not a strategy, it is a broken input. The usual culprits are an AI cost that is really a licence price, or a quality risk nobody measured.

Five strategies (blue/orange bars): blue is value delivered, orange is total cost, the number on top is profit. Each pair is its own world with its own headcount — the sliders labeled "Cut plan" only shape bar 2.
Profit vs how much of the work AI does: the dot is a location, not a gain — "best: 43%" means that policy earns its highest profit when AI does about 43% of the work; the height of the line there is the profit. "AI does X% of the work" is never duplicated work: AI performs that slice of the tasks, and your people do the rest while supervising AI's slice — their role shifts from doing to checking on that portion. At 100% with everyone kept, your team are full-time reviewers of an all-AI production line. A line that only falls means AI hurts this workflow. A green line that dips then recovers is the valley of half-automation: a half-replaced team is the worst place to stand. (If your instinct is "shouldn't freed people build something new instead?" — that is the "Freed capacity redeployed" input, and it activates when you are demand-capped.)
Why do the three lines meet (or nearly meet) at 100%? At full AI share, the lines differ only by how many people each policy kept as reviewers, and each kept person is worth their boost-driven extra output (quality-adjusted) minus their salary. When those roughly cancel, headcount stops mattering — all policies land in the same place, and the only number left deciding your profit is how much value survives unchecked AI. Convergence is the chart telling you: at 100% AI the staffing debate is over, and everything rides on the quality number. In one sentence: each line is a staffing policy, each point is profit if AI does that much of the work, each dot is that policy's sweet spot — and the lines sagging to meet low on the right is the model's core claim: every road to 100% AI ends in the same place.
Profit at each team size: the circled dot is the best team size. A dot at the full team means don't cut. A dot at zero people is a warning, not a plan — audit the AI cost and quality risk that produced it. A curve that is nearly flat on top means the cut question is genuinely contested: measure before anyone is let go.

Rule 1: a strategy that beats today on all three PDFs is safe under every belief — do it now.
Rule 2: a strategy that wins on one page and loses on another is a bet on the unmeasured number — run a 90-day pilot that measures the real boost and the real error/rework rate, then let the measurement decide.
Rule 3: "never" appearing twice in a payback row is what a one-way door looks like in finance — require the measurement before walking through it.
Rule 4: margin, profit, and value delivered rank strategies differently. Margin says how efficient; value says how big the business stays; profit says what you bank. Full automation often wins margin by shrinking the business — say which one you are maximizing before comparing.
The closing sentence for any room: "Only one strategy is green on every page. We do that one today, and we measure the number that decides the rest — before anyone is cut."