Turnover analytics: the 4 metrics that actually predict attrition

Most turnover reporting is an autopsy. Monthly attrition rate, regrettable vs non-regrettable, exit interview themes — all useful, all backward-looking. By the time the number moves, the people are gone. The four metrics below are different: they're leading indicators. They tell you who's likely to leave in the next 90 days, while you still have time to act.

Why most turnover dashboards fail leaders

Headline attrition is a lagging number averaged across the whole company. It hides the manager whose team is bleeding, the comp band that's drifted under market, and the tenure cliff at 18 months that nobody noticed. Aggregates feel responsible. They're not actionable.

The goal isn't a prettier chart. It's a shortlist your CHRO can hand to a manager on Monday morning.

Metric 1: Comp ratio drift by tenure band

Comp ratio is current salary divided by the midpoint of the band. Drift is the change in that ratio over the last 12 to 24 months. The pattern that predicts attrition: employees whose comp ratio has fallen 5+ points while their tenure has grown — they're getting relatively cheaper to the company every cycle, and the market notices before they do.

Cut it by tenure band (0–1 year, 1–3, 3–5, 5+) and by function. Engineering at 2–4 years tenure with negative drift is the single highest-risk segment in most companies we see.

Metric 2: Manager span and tenure mismatch

Two signals stack here: span of control above 8 direct reports, and a manager whose own tenure is under 12 months. Either alone is manageable. Together, they predict above-average attrition on that team within 6 months — the manager doesn't have the bandwidth or the context to retain people through normal friction.

The fix isn't always reorg. Sometimes it's a senior IC promoted to lead, sometimes it's splitting the team, sometimes it's pairing the new manager with a peer coach. But you have to see the signal first.

Metric 3: Promotion velocity gap

For every level in your career framework, calculate the median time-in-level for people who got promoted vs people still sitting at that level. When the still-sitting median exceeds the promoted median by 50%+, you have a stuck cohort. Stuck cohorts leave — usually within 9 months of crossing that threshold.

This is especially sharp at L3 to L4 in engineering and at Senior to Manager in any function. If your top performers can see the ceiling, they'll find it somewhere else.

Metric 4: Engagement-to-action gap

Engagement scores by themselves are noisy. The predictive signal is the gap between what employees flagged in the last survey and what visibly changed. Teams where 3+ specific issues were raised and 0 were addressed within 90 days show attrition 2–3x the company average in the following two quarters.

Track this at the manager level. The number you want is "issues raised vs issues closed," not the abstract engagement score.

How to operationalize this

• ›Run all four metrics monthly, not quarterly. Quarterly is too slow to act on.
• ›Score each employee on a simple flag-count (0–4). Anyone at 2+ flags goes on a watch list reviewed by HRBPs and the manager.
• ›Pair the watch list with a defined action menu: stay conversation, comp review, scope change, promotion case, mentor match. Pick one per person.
• ›Measure the outcome at 90 days. Did the flagged employee stay? Did the action close the gap? Iterate.

What this is not

It's not a prediction model that replaces judgment. It's a triage tool. Some of the people on your watch list will leave anyway — for life reasons, for once-in-a-career offers, for fit. That's fine. The point is to stop losing the ones you could have kept with a 15-minute conversation six months earlier.

Where Tallo fits

Comp ratio drift, span and tenure, promotion velocity, and engagement-to-action all live in the same record in Tallo — so the watch list builds itself instead of getting reconstructed in a spreadsheet every month. Your HRBPs spend the time on the conversation, not on the pivot table.

If you're rebuilding your people analytics stack and want a second opinion on what's worth measuring vs what just looks good on a dashboard, we're happy to walk through it.