Why pay decisions break down before they reach approval

Most bad pay outcomes are not bad judgement. They are decisions made without the context that already existed somewhere in the company.

Ask a Head of Total Rewards where compensation goes wrong and you rarely hear "we picked the wrong number." You hear something closer to: the recruiter had one version of the range, the hiring manager had another, the finance partner was working from last quarter's budget file, and by the time anyone looked at the two people already sitting in that job, the offer had been verbally extended. The decision did not fail at the moment of approval. It failed in the twenty conversations before it.

That is the pattern worth naming, because it changes what you fix. If pay decisions failed because people had poor judgement, the answer would be training. They mostly fail because the context needed to make a good call is scattered across a benchmarking tool, a spreadsheet of ranges, an HRIS, a philosophy doc nobody has opened since it was written, and the memory of whoever handled the last similar case. Judgement is fine. Assembly is the problem.

The four failure points

In practice, decisions break at four recognisable points, and each one has a different tell.

• Context arrives too late. The equity check happens after the candidate has a number in their head, so the conversation becomes about walking something back rather than getting it right.
• Context arrives incomplete. The range is checked; the two peers hired eight months ago at the top of that range are not.
• Context arrives unexplained. Someone says "that is outside policy" without saying which policy, so the exception gets argued on volume rather than on merit.
• Context is never preserved. The decision is made well, and six months later nobody can reconstruct why, so the next similar case starts from zero.

The last one compounds. Every decision made without a preserved rationale increases the cost of the next one, because your team is re-litigating settled questions instead of building on them. Lean teams feel this fastest: when three people handle every offer, promotion, and adjustment across the company, the "institutional memory" is literally their memory.

Speed is not the enemy of defensibility

There is a common assumption that rigorous pay decisions are slow ones. That is true when rigour means adding an approval step. It is not true when rigour means making the relevant facts visible at the moment the decision is being framed.

A recruiter who can see, before making the offer, that the proposed number sits at the 78th percentile of the band, that two peers with comparable scope are at 62 and 65, and that the last three exceptions of this shape were declined, does not need a longer process. They need a shorter one, because the negotiation they were about to open is already answered.

A decision is defensible when the person who made it can state, in plain language, what they checked and what they weighed.

What "defensible" actually means

Defensible does not mean unassailable. Reasonable people can disagree about whether a hard-to-fill role justifies a placement above the midpoint. Defensible means the decision can be explained to three audiences without changing the story: the employee, an internal reviewer, and — if it ever comes to it — someone outside the company examining a pattern of pay outcomes.

That standard has a useful property: it is testable before the fact. You can look at a proposed action and ask whether you could explain it to all three audiences today. If the honest answer is "only if nobody asks about the other two people on that team," you have found the problem while you can still fix it.

Rebuilding the sequence

Fixing this is less about new policy and more about resequencing. Three changes tend to do most of the work.

• Move the checks to the front. Range placement, internal equity, compression risk, and policy fit should be visible when the number is first proposed, not when it is submitted for approval.
• Make the checks explain themselves. A flag that says "internal equity risk" is noise. A flag that says "two peers at the same level and scope are 11% and 14% below this proposal" is a decision input.
• Capture the why with the what. When the call is made, record the reasoning in a sentence or two alongside the number. This costs almost nothing at the time and is the single highest-return habit a comp function can build.

The human stays in the seat

None of this argues for automating pay. Compensation decisions involve tradeoffs — retention risk against budget, market pressure against internal consistency, a specific person's circumstances against a general rule — that a system should surface but not settle. The goal is not to remove the human from the decision. It is to stop asking the human to reassemble the company's context from scratch every single time.

When the context is assembled for them, two things change. The decisions get better, because nothing relevant was missed. And the decisions get faster, because most of the debate in a pay conversation is really a debate about facts that were never established.