Case study
August 2026
In use
What hiring plan reaches the number?
A single-file planning model that turns a revenue target into a hiring plan, then grades how much of its own answer you should trust. Built at Stratus, used by the CRO and CFO for 2027 headcount.
- Sales structures modelled
- 5
- Inputs, each graded
- 36
- Libraries used
- 0
- Decision owners
- CRO + CFO
Context
The 2027 plan needed a seller count. The usual method is a spreadsheet: type a target, divide by a quota, read the headcount off the bottom. That produces an answer, but it doesn't produce a decision, because nobody can see which inputs the answer rests on or what happens when one of them moves.
Six sellers today against a target that implied materially more. Underneath that sat questions a spreadsheet cannot reach: how many accounts can one rep actually carry, does the account base even support the plan, and which of these numbers did anyone measure rather than assert.
The problem
Capacity models tend to be quietly circular. An earlier version of this one computed capacity as the smaller of quota capacity and base yield. The win rate cancelled out on both sides, so capacity always came back equal to the quota ask. It was the target with extra steps, and it would have passed review.
The second failure is flatter and more common. Planning models present every input with the same confidence, whether it came out of the CRM, got said once on a call, or was somebody's reasonable guess. The output looks equally solid all the way across, which is exactly when a number stops being useful for an argument.
What I built
One HTML file. 4,579 lines of vanilla JavaScript, no libraries, no build step, no backend. It opens off a file share and runs in a meeting on a locked-down corporate network, which is the actual deployment constraint.
Five named sales structures act as complete baselines. Pick one and it loads that structure's headcount, per-layer economics, and every global assumption. Around 35 sliders sit underneath. Change any of them and a deviations strip records what moved, so the room can always see how far the current view has drifted from the preset it started from.
The capacity engine works per seat, not per team. It divides the account base into books, subtracts the servicing load, and reports whether each seat is limited by bandwidth or by supply. That distinction carries the weight: two plans can miss the same number for opposite reasons and need opposite fixes.
The account base it plans against is not the raw CRM. I rebuilt it first in Clay: a full TAM and SAM scrape to expand the universe, rules-based Enterprise and Commercial segmentation, and an agent that visits each company's site to check for the service lines that define fit. The supply side of this model is only meaningful because that ran first.
Three things that make it a decision tool
- It ranks its own recommendations
- Every structure runs through the capacity engine and sorts by gap, so “best on the numbers” is computed rather than argued.
- It grades its own inputs
- A trust ladder splits all 36 inputs into measured, stated, and assumed. The sheet asks you to open it before reading any verdict below.
- It solves backwards
- Given a target it searches headcount vectors until capacity clears, and separately inverts the funnel to show what each input would need to reach for the current plan to close.
Drive it
The real model, running here. Switch structures, drag the assumptions, watch the verdict move. Start with the guided tour on the Coverage sheet if you want the short version. Account names, rep names, and the company are anonymised; the figures are the real measured values.
Loading the live artifact…
Three things to look at
The opening sheet states the gap before anything else: the ask, the capacity the current structure produces, and the shortfall between them. Below it, two auto-ranked structure recommendations.
The trust ladder. Every input is sorted into measured, stated, or assumed, and the count is computed from the model itself rather than maintained by hand.
The supply side. A qualification funnel with a dashed line marking how much of it the books can actually reach, plus a ranked list of the data problems that would move the answer.
Outcome
The CRO and CFO use it to make 2027 headcount decisions. It also became a supporting artifact in the decision on whether to launch a new product, because the same account-base analysis answers whether there is enough addressable supply to justify the build.
The comments in the source are a running record of that reasoning: which model got replaced and why, where two inputs disagreed by ten points, which sections were cut and what was wrong with them. That was deliberate. A planning model exists to be argued with, and the argument goes better when the reasoning is sitting in the file.