Pay Unfairly

What Bock argues

"Pay unfairly" is a deliberate provocation, and the real claim underneath it is precise: pay should vary much more than most companies allow, because contribution itself varies that much (the power-law of Chapter 8) — so paying two people the same for wildly different output is its own kind of unfairness. Bock's position is that large internal pay inequality is acceptable, even necessary, on one condition — that people can see the process is fair. Plus: not all reward is cash; experiences and recognition move people in ways a bigger number doesn't. The chapter is really an argument that the distribution of pay should track contribution while the process of setting it earns trust.

The provocation works only because of a distinction Bock states casually and the research states precisely — and getting it precise is the whole chapter.

What the research actually says (and where 2015 needs an update)

People do not judge their pay in isolation. Equity theory (Adams, 1965) says they judge it by comparing their own ratio of inputs-to-outcomes against a referent — a peer, a predecessor, the person down the hall — and felt inequity (in either direction) produces tension they act to resolve: withdrawal, reduced effort, exit. So wide pay dispersion is not free; it loads the equity comparison. The naïve read is "minimize dispersion to keep the peace." That is the wrong lever, and here is why.

The fairness people actually respond to splits in two, and the split is the one we drew in Chapter 2. Distributive fairness is whether the outcome (your number) was fair; procedural fairness is whether the process that set it was fair. And the load-bearing finding — the same one that runs through Show Your Work — is that the organization-directed outcomes you care about (commitment, trust, staying, discretionary effort) are driven by procedural justice, not distributive. Pay satisfaction tracks the dollar; trust in the institution tracks the process. That is exactly what makes "pay unfairly" survivable: you can pay unequally — even widely — and keep trust, if the process is legible and consistent. Strip the process out of view and the same dispersion reads as favoritism, and the equity tension turns corrosive. Dispersion isn't the risk. Opaque dispersion is.

Two honest caveats the Guide must add. First, pay-for-performance is fragile: tying money tightly to measured output invites gaming and can crowd out intrinsic motivation (the Chapter 6 / Drive problem) — pay variation works best for clearly attributable contribution, worst for the interdependent, hard-to-measure work that is most of modern knowledge work. Second, "pay unfairly" assumes you can see unfairness you didn't intend. You usually can't, by eye — which is the whole reason this is a measurement chapter.

Where 2015 needs the update: comp is going algorithmic — market-pricing engines, AI-assisted ranges, pay-equity tooling. Run through Chapter 2's lens, that is either the best or the worst thing to happen to pay fairness, depending entirely on legibility. A black-box pay algorithm is the maximum information vacuum pointed at the most equity-sensitive decision in the building; suspicion fills it on schedule. "Pay unfairly" only survives in the AI era as pay-unequally-and-transparently-by-construction.

How you run it

Measure both halves, then read the gap — the Show Your Work move applied to compensation.

• Measured fairness (distributive): the pay-equity residual — what disparity remains after you control for the legitimate drivers (role, level, location, tenure, measured performance), via an Oaxaca-Blinder–style decomposition. This is the disparity you can't explain away.
• Felt fairness (procedural): a procedural-justice-on-pay pulse — do people understand how pay is set, can they trace their own number, do they believe the process is applied consistently.
• Pay-for-performance sensitivity: how tightly pay actually tracks performance (and whether the performance signal it tracks is itself reliable — see Ch 7).

The signal is the gap. A small measured residual sitting next to large felt-unfairness is the most common and most fixable case: the pay is defensible, the process is the problem, and the fix is legibility — not a raise. A large measured residual is a different, harder conversation. Reading the two together is the diagnosis; either number alone misleads.

The analysis you can execute

Near-zero net-new — this is the chapter where the toolbox is most ready. The pay-equity residual is anycomp / pay-fairness, already built; the felt-fairness pulse is the procedural-justice spoke (shared with Ch 2); the pairing — measured residual × felt fairness, the gap as headline — is the same paired-measurement machinery the Show Your Work essay describes. Segment by group through the min-N privacy gate. One bundle, three chapters (2, 7, 10) leaning on it.

The AI-era turn

If pay is going to be set with algorithms, the Chapter 2 requirement becomes a comp requirement: build the pricing and equity tooling so an employee can see why their number is what it is. Legible comp is the operational form of "pay unfairly and keep trust" — and, as everywhere in this book, it's the move a black-box incumbent can't retrofit. Transparency was always how dispersion survives; in the AI era it's also the difference between a pay system people trust and one a plaintiff's lawyer subpoenas.

What to do Monday

1. Compute the pay-equity residual (anycomp) — the disparity left after the legitimate controls. Know your real number before anyone else computes it for you.
2. Run a short procedural-justice-on-pay pulse on the same population.
3. Put them side by side. If the residual is small but felt-fairness is low, fix the process (make pay legible), not the dollars — and resist the reflex to answer a trust problem with a raise.
4. Before adopting any comp algorithm, ask the Ch-2 question: can the person it prices see how it priced them? If not, you're about to automate the equity vacuum.

Cross-refs: content/magazine/show-your-work.md (procedural justice, the measured-vs-felt gap); Ch 2 (same machinery, culture-wide); Ch 8 (the power-law that makes wide dispersion legitimate); Ch 7 (is the performance signal pay tracks even reliable?).