Reading a Year of Exits Together
What an annual review of departures should measure, why the aggregate is more useful than any individual case, and the four patterns worth looking for.
What one organisation found when it read twelve months of exits in one sitting
Eleven final-payment queries out of forty-one exits is not eleven mistakes; it is one process problem that nobody had counted. This is one organisation's own review, not a benchmark.
Read all of the year's departures in one sitting, once a year, because individually they are incidents and together they are information. Nobody does this, and the reason is structural: each exit is handled by whoever was available, closed, and forgotten, and no role owns the set.
The pattern described in “Reading a Year of Exits Together” becomes useful only when the source data is consistent and interpreted with context. Teams exploring time tracking with screenshots through a practical route to time tracking with screenshots can use time and project records to locate operational questions, while direct feedback, HR evidence and accountable review remain necessary to explain them.
The review takes a day and is reliably the most useful day anybody in a people function spends. It is also the only way the problems described everywhere else on this site become visible as problems rather than as a series of unlucky weeks.
For an independent reference relevant to “Reading a Year of Exits Together”, consult the Google Security Blog. Use it to test record quality, access, retention, fair process and exception handling against the organisation’s real departure workflow.
What to count
Start with the facts the file already contains, categorised consistently: how many people left, by what route, from which teams, with what length of service, and how long each process took from first conversation to closed file.
Then the quality measures, which nobody records and which matter more: how many final payments generated a query, how many files were closed within a month, how many appeals were lodged, how many had a contemporaneous note before the process started.
Four patterns worth looking for
- Clusters by team or manager, where a disproportionate share of departures comes from one place.
- Clusters by tenure, particularly departures inside the first year, which say something about recruitment or onboarding.
- Resignations that follow a process — somebody who resigned shortly after being spoken to.
- Repeated process failures of the same kind, which indicate a system problem rather than individual carelessness.
The fourth is the one that produces action. Eleven payment queries is not eleven careless payroll officers; it is a process that computes final payments too late with no second check.
Resignations that are not quite resignations
Every review should compare resignations against what happened in the preceding months. Somebody who resigned three weeks after a difficult conversation is a data point of a different kind from somebody who resigned to move abroad.
That comparison is never made because the two facts live in different places — the resignation in the HR system, the conversation in a manager's recollection. Making it is a matter of listing the resignations and asking one question about each, and the answers are frequently uncomfortable.
What not to conclude
The aggregate shows where to look, not what is wrong. A team with high turnover may have a difficult manager, or a difficult market, or a deliberate restructuring, or a manager who is finally addressing problems their predecessor left.
Treating the numbers as findings rather than as questions is how a review produces resentment instead of improvement. The output of this exercise is a short list of places to ask about, and that is all it should claim to be.
Who should see it
Somebody with the authority to change a process, which usually means more senior than the person who compiles it. A review that circulates within the people function and stops there changes nothing.
It also has to be presented in terms of process rather than individuals wherever possible. "Final payments are computed after the last day in most cases" is actionable; "payroll made eleven errors" is a fight.
Doing it consistently
The review is only worth anything in the second year, when comparison becomes possible. That means the categories have to stay the same, which means writing them down rather than deciding them afresh.
A one-page definition of what each category means, kept with the review, is what turns a yearly exercise into a measurement. Without it, every year's numbers are approximately right and mutually incomparable.
Doing it with small numbers
An organisation with nine departures a year cannot do statistics, and attempting to produces nonsense. What it can do is read all nine files in an afternoon, which is not available to anybody with four hundred.
That is an advantage rather than a limitation. The small organisation gets the detail — what actually happened in each case — where the large one gets the aggregate. The questions are the same; only the method differs.
The thing it usually finds
In most organisations the first review finds the same thing: the process is fine when somebody senior is paying attention and absent otherwise, and nobody could have said that before counting.
That is a genuinely useful finding, because it points at the fix — a checklist and an owner rather than more training. It is the audit nobody scheduled, conducted deliberately for once, and it is a question worth asking once a year rather than never.