Skip to content
allanninal.dev/build
Collaborate

Part 5 of 7 · Exit interview collector series ~5 min read

How the retention report reads

The report is short, boring, and carefully stripped of everything that would make it more interesting, because every one of those things is a way of identifying somebody. What is left is still enough to change how a business runs.

Key takeaways

  • Counts and directions only. No quotes, ever, however anonymised they look.
  • No department, role or manager breakdown. In a small business those are names.
  • A coverage line at the top, so the report is not read as more than it is.
  • Quarterly, not live. A dashboard updating after one response is a name with a delay.
  • The unmatched count is reported, because it says the theme list is out of date.

What the page says

Q3 2026 — leavers

  • Coverage. 7 leavers. 5 answered at exit, 4 answered at six weeks, 3 answered both.
  • Scheduling — 4 people, rising. Up from 1 last quarter. Two said they raised it while employed.
  • Progression — 3 people, flat. Same as the last two quarters.
  • A better offer — 5 at exit, 2 at six weeks. The gap is the largest recorded so far.
  • Unmatched — 3 answers. Above the usual 1. Worth a read of the theme list.
  • No individual response is shown. Themes need three people before they appear.

Five lines and a footer. The coverage line is first because it determines how much weight the rest deserves: three people answering both asks out of seven leavers is a real signal and a small one, and putting that at the top stops the report being quoted in a board meeting as though it were a survey.

Everything it refuses to include

Four report features that are deliberately refusedA vertical chain of five steps entered by a box labelled A tempting addition, asked for regularly. Step one is a representative quote, described as just one and anonymised; it exits to Refused, because a sentence is a fingerprint. Step two is a breakdown by department, to target the fix; refused, because a department of four is four names. Step three is a breakdown by manager, framed as for their development; refused, and noted as the one people ask for. Step four is a live dashboard, so it is seen sooner; refused, because a count that moves is a person. Step five is Counts and directions, which is what remains. A note says each of these is a reasonable request and each is a way to identify somebody.AWS ACCOUNTA tempting additionasked for regularlyA representative quote'just one, anonymised'Refuseda sentence is a fingerprintnoBreakdown by department'to target the fix'Refuseda department of four is four namesnoBreakdown by manager'for their development'Refusedthis is the one people ask fornoA live dashboard'so we see it sooner'Refuseda count that moves is a personnoCounts and directionswhat is leftEach of these is a reasonable request and each is a way to identify somebody.
Fig 1. The four additions that get requested and are refused. None of them is unreasonable and all four defeat the property the whole system depends on.
  • App integration
  • Security & identity
  • Analytics
  • Front-end & mobile

Why a quote is worse than a name

A quote is presented as the safe compromise: no name attached, just what somebody actually said. In a business of thirty people, a sentence about a specific incident, in somebody’s own phrasing, identifies them more reliably than a name would — because everybody knows who was involved, and now they also know what that person said about it. There is no way to anonymise a quote that survives contact with colleagues.

Why the manager breakdown is the one people ask for

And it is the one that must never exist, for a reason that has nothing to do with protecting managers. The moment a report can show what a manager’s leavers said, everybody in the business understands that answering honestly is an accusation against a named person. Answers become diplomatic within one quarter and the system produces nothing useful ever again.

The thing to do with a scheduling theme is fix the scheduling. If the fix requires knowing where, that is a conversation somebody has with people who still work there, which is a better conversation anyway.

Why quarterly

Why exit themes are reported quarterly rather than on a live dashboardA horizontal row of five boxes. Live count: moves on a single answer. Who left: everybody knows. The join: is trivial. Quarterly: several people are reported at once. Safe: and still timely enough. A note says a live count and a known leaving date are the same thing as a name.WHY THERE IS NO DASHBOARDLive countmoves on one answerWho lefteverybody knowsThe joinis trivialQuarterlyseveral people at onceSafeand still timely enoughA live count and a known leaving date are the same thing as a name.
Fig 2. Why the report is quarterly rather than live. A count that moves the week somebody left is attributable by anyone who noticed they left.
  • Machine learning
  • Management
  • Analytics
  • People

A quarter is late enough that several responses are pooled and early enough that a rising theme can still be acted on within the same year. It is not a compromise on speed so much as a recognition that this data is about patterns, and a pattern that is only visible within a week is not a pattern.

Next: what all of this costs to run.

All posts