Reading a leaver cohort without blaming a single manager

Someone in the room will want a manager ranking. A team of seven that lost two people will look like a disaster; a team of ninety that lost twelve will look like discipline. The arithmetic is true and the story is often false.

We draw leavers by tenure first. A pile of exits before six months is a different problem from a pile at year three. Then we draw site or unit, especially when one coastal shop or one night shift is carrying the company’s “high turnover” reputation.

Manager splits come last, and only with a stable identifier. Even then we suppress teams below a size we agree in scoping — usually ten people — because a single resignation becomes a percentage you cannot put in a board appendix without a fight.

Reason codes deserve suspicion. “Personal” in Georgian files often means the form had no better box. Recoding last year’s exits from memory is not analysis; it is a new dataset. If you want better reasons next year, change the form now, not in the briefing.