Two departments, one attrition rate, opposite problems

Two departments in the same company closed the year at fourteen percent voluntary attrition. The benchmark for the sector was fourteen. Nobody looked further, because there was nothing in the number to look at.

Underneath, they had nothing in common.

The first department had lost four people from the bottom performance quartile and one from the top. Its manager had spent the year making overdue decisions, and the attrition figure was the sound of a system working.

The second had lost four people from the top quartile and one from the bottom. Three of the four had been there between two and four years. All four had reported to the same person.

Same rate. Opposite problems. Opposite responses. And an aggregate that matched the benchmark, which meant neither department got read at all.

An average is a device for making a distribution disappear

That is the whole mechanism, and it is not a failure of anyone's attention. It is what averages are for. A single figure compresses a population into one number so it can be compared to another company's one number, and the compression is the feature. The cost is that everything inside it becomes invisible in the same move.

A rate that comes in above the benchmark at least generates a question. A rate that matches it generates nothing, which makes a healthy-looking aggregate the most expensive number in the reporting pack. It is the one nobody opens.

Right now that problem is worse than usual. The US quits rate has sat at or near 1.9 percent long enough that flat is the honest description, and the Bureau of Labor Statistics called every component of its July 2026 release little changed (Bureau of Labor Statistics, Job Openings and Labor Turnover Survey, July 2026). In a low-quits year, most aggregates look fine. That is a market condition rather than a result, and it does not distinguish between the two departments above.

The cut that separates them

One cut, on data already held, and it takes an afternoon.

Join every voluntary exit from the last twenty-four months to two fields: the manager of record at the time of exit, and the last performance rating on file. Then look at the second field first.

If the departures cluster in the bottom performance band, the number is reporting a functioning system and no further work is needed on it.

If they cluster in the top band, the rate is not measuring turnover. It is measuring selection, and the selection is running the wrong way.

If they cluster under one manager across both bands, that is the finding, and it is the one worth having. Research on culture-related resignations puts poor management at the root of 58 percent of them, which is roughly what this cut tends to return.

Twelve months is not enough window to see it. The cells are too small at manager level and one ordinary departure swings the rate. Twenty-four is the minimum that holds.

What the cut does not tell you

It does not tell you why. A manager with a bad number may have inherited a team mid-restructure, may be running the hardest function in the company, or may have been handed three people who were already leaving. The cut identifies where to look and says nothing about what is there.

It also names individuals, which is the reason it goes unrun in most companies that could run it tomorrow. Decide who sees the output before it is run rather than after. That decision is easier to make in advance and almost impossible to make well once the file exists.

What this is worth

The second department above was losing roughly four people a year more than the first, from a more expensive band. At a loaded replacement cost in the region of $220,000 for a professional hire — an assumption drawn from the published bands, which run 100 to 150 percent of salary for professional roles, rather than a measurement — that difference is most of a million dollars a year sitting inside a number that matched its benchmark.

Scale it. A thousand professional employees and a ten-point spread between the best-retained group and the worst produces around twenty-five departures a year above baseline, or roughly $5.5 million annually. Every company knows how many people left. Almost none know which ones.

That is a difference rather than a total, which is what makes it defensible. The objection to any turnover figure is that some turnover is healthy. A spread has already answered it, because the better-retained group inside the same company, under the same policies, is the proof that the number is reachable.

Two departments, one rate, and the cut that tells them apart takes an afternoon on data already in the system. Most companies have never run it once.

Which of your departments matched the benchmark last year?

The calculator that prices the spread on your own headcount: Employee Turnover Cost Calculator

Peter Allen Mann is the author of The Perception Revolution (Fast Company Press, 2026) and has spent thirty-seven years in leadership across the Navy, Dell, Tech Data and two companies he founded. He writes at peterallenmann.com.

Sources

• Bureau of Labor Statistics, Job Openings and Labor Turnover Survey, July 2026 reference month, released 1 September 2026. Quits rate 1.9 percent; all components little changed from June. bls.gov/news.release/jolts.nr0.htm
• Society for Human Resource Management, published replacement-cost bands: 100–150 percent of salary for professional roles. Gallup: one-half to two times salary, described by Gallup as conservative.
• Mann, P. A. (2026). The Perception Revolution. Fast Company Press. Ch. 2.