Salary compression: how it starts and how to unwind it

Compression is rarely one bad decision. It is a series of individually reasonable ones, viewed together for the first time when someone resigns.

Compression is the gap between two people that has become too small to make sense — a senior engineer earning barely more than the person they onboarded last spring, a manager within a few thousand dollars of their most junior direct report. It is one of the most common structural problems in growing companies, and one of the least visible until it becomes a retention event.

How it forms

Compression is almost never the result of a single bad call. It forms through the accumulation of decisions that each looked correct in isolation.

• External market rates rise faster than internal increase budgets. New hires come in at current market; incumbents move at 3-4% a year.
• Hot-skill premiums get applied to new hires and not retroactively to the people who already have those skills.
• Counteroffers and retention adjustments are given case by case, moving individuals without moving their peers.
• Promotions are granted with a title change and a modest increase, leaving the newly promoted person below others at their new level.
• A single high-pressure hire sets a de facto new floor for a role that nobody updates the band to reflect.

Notice that each of these is a defensible action. That is exactly why compression is hard to prevent with policy alone: there is no rule you can write that catches "reasonable decision that becomes unreasonable in aggregate." You need visibility across decisions, not stricter rules for individual ones.

Finding it before it finds you

The diagnostic is straightforward once you decide to look. For each level and job family, compare the pay of adjacent levels and of peers within a level. Two indicators matter most: the differential between levels, and the spread within a level relative to tenure and performance.

A healthy structure has a consistent, deliberate differential between adjacent levels — enough that moving up is materially meaningful. When that differential drops below what your philosophy says it should be, you have compression whether or not anyone has complained yet.

The related failure mode is inversion: a subordinate or a less experienced peer paid more than the person above them. Inversion is compression that has already gone past the warning stage, and it tends to become known internally faster than teams expect.

Compression is discovered in one of two ways: by a report you ran, or by a resignation you did not see coming.

Unwinding it

Correcting compression is a budgeting and sequencing exercise, and it is worth being honest that most companies cannot fix all of it at once.

Start by separating true compression from apparent compression. A small gap between a high-performing junior and an average-performing senior may be entirely defensible. Compression that needs correcting is where the gap is unexplainable by level, scope, or performance.

Then triage by risk and severity. Inversions first, since they are indefensible and highly visible. Then cases involving people whose loss would be materially damaging. Then the broader population, in a planned cycle with a stated multi-quarter path if the full correction does not fit one budget.

Communicate carefully. An off-cycle correction that arrives with no explanation reads as either an apology or a reward, and both interpretations create problems. Framing it as a structural correction — the band moved, we are aligning people to it — is accurate and avoids implying that everyone else should ask for one.

Preventing the next round

Prevention comes down to running the compression check at the point of decision rather than at the point of annual review. Every offer, every promotion, every retention adjustment should show its effect on adjacent pay relationships before it is approved.

The other half is band maintenance. Compression frequently signals that a band no longer reflects the market you hire in. If you find yourself approving repeated exceptions to hire into a range, the range is the thing that needs fixing — and fixing it is cheaper than the sequence of individual corrections that follows from not fixing it.