Retention Analytics: Decoding Attrition Before It Becomes a Crisis
Attrition is the most reported and least understood HR metric. This guide separates regretted from non-regretted, builds early-warning indicators, and turns exit data into a system that prevents the next wave — not just counts the last one.
If your only attrition metric is an annualized percentage, you are flying blind. Two companies at '15% annual voluntary attrition' can be in completely different situations — one losing its top performers in year 2, the other losing low performers in month 4. The number is the same; the diagnosis is different; the prescription is the opposite.
Frame the question correctly
- Total attrition is a vanity metric. Split into voluntary, involuntary, and reorg.
- Voluntary attrition splits into regretted and non-regretted (your judgement, your call, recorded).
- Regretted attrition is the only number worth reporting to a board.
- Tenure matters more than total — losing people in year 1 vs. year 5 are different diseases.
Definitions that hold up
| Metric | Formula | Notes |
|---|---|---|
| Annualized voluntary attrition | (Voluntary leavers in period / avg headcount) × (12 / months in period) | Use trailing 12m for board reporting |
| First-year attrition | Voluntary leavers <12m tenure / new hires in same window | Diagnostic for hiring + onboarding |
| Regretted attrition rate | Regretted voluntary / total voluntary | Manager + skip-level codify regret |
| High-performer attrition | Voluntary leavers rated top 25% / total top 25% | The most important leading indicator |
| Manager attrition | Voluntary leavers reporting to a given manager / their direct reports | Surface in talent reviews |
Cohort and survival analysis
Cohort survival curves (borrowed from product analytics and biostatistics) are the most underused tool in people analytics. Plot the percentage of each hiring cohort still employed at month 3, 6, 12, 18, 24. The shape tells the story.
- Steep early cliff (months 3–6): onboarding / wrong-hire problem.
- Cliff at month 12–18: 'bonus cliff' or growth-stagnation problem.
- Steady decline: cultural / compensation drift.
- Long tail flat: a healthy senior cohort.
Early-warning indicators
| Signal | Where to find it | Lift |
|---|---|---|
| Decline in calendar / commit activity | Workplace analytics (Viva, Cultureamp) | 2–3x |
| Recent missed promotion or rating drop | HRIS performance data | 2x |
| LinkedIn profile activity spike | External tools (Lever, Gem) | Strong signal, ethically tricky |
| Engagement score drop >1 point | Pulse surveys | 2x in 90 days |
| Skip-level escalation about manager | Skip-level notes | Strong qualitative signal |
| Compensation compression vs market | Comp benchmarking | 1.5–2x |
Profile-watching and message-content mining destroy trust if leaked. Use behavioural signals at AGGREGATE team level for intervention; never confront an individual with surveillance data.
Exit interviews that produce signal
- 1DecisionWhen did you first start thinking about leaving? What was the trigger?
- 2PushWhat pushed you out — manager, work, comp, growth, culture, life event?
- 3PullWhat pulled you toward the new opportunity?
- 4StayabilityWhat could we have done differently to keep you? Be specific.
- 5ForwardIf a friend asked, would you recommend joining? Why / why not?
Not the manager. Not HRBP if the issue is HR. Use a neutral facilitator and aggregate themes monthly. Best practice: 30–90 day post-exit follow-up captures more candor than the exit week.
Stay interviews
Stay interviews are the prevention to exit interviews' autopsy. Conducted with high performers every 6–12 months, they identify and remove the friction that would otherwise show up in an exit.
- What keeps you here?
- What would tempt you to leave?
- What's the worst part of your week?
- Is your work using your strengths?
- When did you last feel proud of work you did here?
- What would you change about your role / team / company tomorrow?
Interventions that work
| Diagnosis | Intervention | Typical effect |
|---|---|---|
| Manager-driven attrition | Manager coaching + replacement if chronic | Sharp localized improvement |
| First-year attrition | Pre-boarding + 30-60-90 plans + buddy system | 20–40% reduction |
| Comp compression | Market refresh + targeted comp adjustments | Reduces top-quartile leavers materially |
| Growth stagnation | Career ladders + lateral moves + sponsorship | Reduces 2–3 year attrition |
| Burnout signals | Workload audit + manager 1:1 reset + leave policy | Quick wins possible |
| Culture drift | Re-anchor values + ritual reset + leadership behavior | Slow, deep, real |
Mini-cases
- Stripe published an internal 'why people leave' breakdown by reason and tenure; the act of publishing reduced attrition the following year.
- Buffer's transparent salary formula reduced compensation-driven leavers; remaining attrition was almost entirely growth-related.
- Microsoft used Workplace Analytics (Viva) to identify after-hours overload patterns predicting exit, and intervened at team level via manager coaching.
References
- Harvard Business Review — Predicting Who Will Quit — HBR
- Visier — Attrition Benchmarks — Visier
- MIT Sloan — Toxic Culture and the Great Resignation — Sull et al., 2022
- Gallup — Why Employees Stay or Quit — Gallup
- SHRM — Cost of Turnover Calculator — SHRM
Read next
All playbooksA short, defensible dashboard for leaders. Skip the vanity metrics and track what predicts hire quality and retention.
Most HR dashboards are unread. The reason is consistent: too many metrics, no decision context, no benchmarks, and no narrative. This is how to build a one-page people dashboard for the CEO and board that gets opened weekly — modelled on Andy Grove's indicator philosophy and modern people-analytics practice.
Annual engagement surveys are dying — slow, low signal and rarely acted on. Modern listening is a system: census surveys, pulse, lifecycle, always-on, and qualitative loops, designed end-to-end with the close-the-loop ritual that determines whether anyone fills it out next year.