The Compensation Decision Thesis — plain-language explainer

Compensation has no direct line to firm performance. Its only performance lever is how it shifts the exit rate by performance characteristic over time — pay competitively enough and the people who matter face fewer viable external alternatives, so the right people stay. Treating pay as a religion (match the market because everyone does) is a false premise; the real job is to spend comp where it cuts exit risk for your key roles and sustained high performers, and let the risk ride elsewhere. The dose-response curve behind it is calculable in principle and must be calibrated honestly — that calibration is the open work.

A People Analytics Toolbox capability. Built to the portfolio Explainer Standard v1.0. Grounded in the real code — the AnyComp decision spoke (src/spokes/anycomp/, the measures + actions + scenario loop), its BLS-prior exit-risk layer, the segmentation-studio attrition-risk segment, and the positioning canon in docs/POSITIONING/. The exit-rate dose-response model is named honestly as not-yet-calibrated.

1. What good looks like

Good compensation strategy, stated as the thesis rather than the ritual:

Pay is aimed at exit risk, not at the market for its own sake. Good looks like spending compensation to address the exit risk of the people who matter most — prioritized key roles and sustained high performers — and consciously letting the risk ride where retention isn't strategically worth the dollars.
Every dollar is traced to a stated priority. Good is a small set of distinct, scored scenarios — never a single take-it-or-leave-it number — each one a way to spend the same budget, each scored on the same measures, so the comp leader chooses defensibly among several rather than rubber-stamping one.
The impact is communicated in the language leaders speak. Good looks like a stated, honest claim of the form "moving this segment from the 50th to the 75th percentile reduces their viable external alternatives by a calculable margin → lower exit rate" — not an arcane band-math table.

The plain version: there is no direct translation of what you pay people into organizational performance. Where pay produces a performance lift is through how it changes who leaves, by performance characteristic, over time. You either control that or your competitors do.

2. Why keep checking

Markets move, and a percentile you set last year is a different percentile this year. Pay position relative to the market distribution is what drives the count of viable external alternatives an employee faces — and that distribution drifts. The exit-risk picture is a moving target, so the read has to be standing, not annual.
Exit risk concentrates and shifts. A high performer in a key role who was safe at last cycle's pay can become a flight risk as the market re-prices or as their performance rises. Checking continuously is how you catch the concentration before the departure, not after.
The dose-response must be re-estimated as data accrues. The "50th → 75th cuts exit by X%" relationship is a hazard curve estimated per performance/role segment — and it sharpens only as the cross-client, outcome-linked data flywheel grows and Principia priors firm up. Keeping it under estimation is the difference between an honest curve and a stale guess.

3. What the problem is — and why it matters

The pain it removes: most companies manage compensation like a religion — something to "get right" for its own sake, where "right" means "what everyone else pays" (the market). That is a false premise on multiple levels: matching the market neither targets the people who actually matter to the firm nor connects, in any traceable way, to performance. It just sets a number and calls it discipline.

Why it matters: comp is one of the largest controllable line items, and spent as a market-matching ritual it buys very little of what leaders think it buys. Re-rooted in exit risk, the same budget becomes a precise instrument — retain the high performers and key roles whose departure actually costs you, and stop overpaying to retain people whose exit you could afford. The mechanism is rigorous, not hand-wavy: pay percentile vs. the market distribution → the share of viable better-paying external alternatives an employee faces → exit hazard → exit rate by performance segment → retention of the right people → performance.

The shift, stated plainly:

FROM comp managed as faith — match the market, treat the number as the goal, hope it helps performance.
TO comp managed as a lever on the exit rate by performance characteristic — spend where it cuts the hazard for your key roles and high performers, let it ride elsewhere, and communicate the predicted exit-rate impact of a change.

How it differs: a generic comp tool prices you to the market and stops; this thesis treats the market price as an input to a retention calculation, not the answer. The output isn't "you're at the 48th percentile" — it's "here's what moving the segment that matters would do to your exit rate, and here are several scored ways to spend the budget to get there."

4. Where it fits in the toolbox

Data flow and dependencies:

Exit-risk model per employee — built on the existing anycomp exit-risk priors (the BLS labor-metrics layer, keyed by SOC × geography × period), the segmentation-studio attrition-risk segment, and the firm's HRIS plus a small set of survey questions → P(exit) per person.
Comp → exit-rate dose-response — pay percentile → viable-alternatives count → exit hazard, grounded in the market distribution (wage-benchmark) and external priors (principia-connector), estimated per performance/role segment.
Strategic retention allocation — the AnyComp decision layer's optimizer, with exit-rate-by-performance as a first-class outcome: spend comp to cut exit risk where strategic value is highest, and produce several scored scenarios rather than one option.
Communicate the impact — the explanation/reporting layer: the predicted exit-rate change by performance segment from a comp change, in narrative and visualization a leader can act on.

The honest rail. The mechanism is calculable in principle — it is survival/hazard modeling over a real market distribution, not a slogan. But the dose-response must be estimated honestly: causal, not naive correlation; survival modeling, not a line fit. Today the exit-risk priors and the market distribution are real, while the precise per-segment exit-rate curve is not yet calibrated — it sharpens as the outcome-linked data flywheel and the priors mature. We will not claim a precise exit-rate prediction before it is calibrated; honest uncertainty is the central feature, not a disclaimer in the footnotes.