When someone resigns, no invoice arrives. The recruiter's time, the empty seat, the six months the replacement spends getting up to speed — none of it lands on a ledger line, so the retention proposal walks into the budget meeting with a feeling while every competing request walks in with a number. A director can lose four engineers in a year and be unable to say what that cost.

Cascio and Boudreau's Investing in People devotes two full chapters to what they call the high cost of employee separations, and their central claim is an accounting one: turnover has a computable, fully-loaded cost — separation processing, replacement acquisition, and the long tail of lost productivity while the new hire ramps — and that cost is a multiple of anything visible in the ledger. They insist on the distinctions that make the number defensible rather than dramatic: voluntary versus involuntary separations, cost per event versus cost annualized over a segment, and absenteeism as the recurring cousin of turnover that deserves its own line. Their larger point is postural — treat these as investment analyses that inform decisions, not scary numbers that decorate slides.

Mike West's People Analytics For Dummies places the same arithmetic inside the Attraction–Activation–Attrition arc and pushes on the second half of the problem: the cost tells you the stakes, but managing attrition means finding the real reasons people leave — through segmentation and regression on your own data, not exit-interview folklore. Diez, Bussin, and Lee's Fundamentals of HR Analytics supplies the discipline that gets the number heard: make the business outcome the dependent variable, and walk into Finance with turnover expressed as a P&L consequence rather than an HR statistic. The method's honest limit is its assumptions — recruiting cost as a percent of salary, ramp-time productivity loss, knowledge-loss estimates (the softest component) — which is why the serious version of this analysis shows its defaults and its sensitivity, not just its total.

The books hand you the cost model and a blank spreadsheet; here the arithmetic runs deterministically in code — every default rate visible and overridable, missing inputs reported rather than invented — and the write-up flags which assumptions move the total most before you take it to Finance.