The engaged exit: people who leave don't go quiet

Most retention monitoring rests on one assumption: people who are about to quit start pulling away before they go. They answer fewer surveys, skip more check-ins, and the drop in their signal is the early warning. Catch the silence, the thinking goes, and you catch the risk.

We set out to validate that assumption directly. Using a year of daily happiness check-in data from 5,713 employees, we pre-registered three hypotheses about how "silence" — gaps in check-in responses — predicts exit. We compared 2,099 people who left the company against 3,614 who stayed, matched on tenure and time window.

All three hypotheses were refuted or only partially supported. Departing employees did not go quiet. They responded more in their final 90 days than earlier in their tenure. The group with the smallest check-in gaps — the most engaged — had a higher exit rate than the group with larger gaps. The real signal was not behavioral at all. It was sentiment: people who left were consistently less happy from the start, and that gap never widened. It was simply always there.

How we measured it

The Happily daily check-in asks one question — "How do you feel today?" — answered on a 5-point scale. We treated two things as measurable: whether a person responded on a given day (engagement), and what they answered when they did (happiness). On the answer scale, a lower score is happier; the very-happy response is the lowest value, and the two least-happy responses are grouped as "unhappy."

Each hypothesis compared a cohort that exited the company in the past 365 days against a control cohort that stayed. Both cohorts required at least 180 days of tenure, so a six-month baseline existed for everyone. Test companies and known non-production accounts were excluded.

The three hypotheses were, in order: that exited employees miss more check-ins in their final 90 days (the gap signal); that a threshold of consecutive missed days predicts exit (the threshold effect); and that happiness declines before responses go silent (the mood-then-silence pattern). We report each below against what the data actually showed.

Finding 1 — Departing employees engage more, not less

The gap-signal hypothesis predicted that response rates fall as exit approaches. For exited employees we compared two windows: their final 90 days, and the 90 days before that. We also compared both against retained employees' most recent 90 days.

The result is the opposite of the hypothesis. Exited employees responded 11.5 percentage points more in their final 90 days than in the prior period — and slightly more than retained employees did over the same recent window. There is no fade. The standard deviation also tightened, from 26.0% in the prior period to 16.4% in the final period, so departing employees did not just engage more, they engaged more consistently as exit approached.

Finding 2 — The most engaged group exits the most

The threshold-effect hypothesis predicted that a critical run of consecutive missed days — 7 or more — marks people who are about to leave. We computed the longest check-in gap for every employee with at least 180 days of tenure and grouped them by that gap, then compared exit rates across groups.

Employees whose longest gap stayed under 7 days — the most consistently engaged group — exited at 41.6%, roughly four times the 10.2% rate of the 7-to-13-day group. The 14-to-29-day and 30-plus-day categories did not reach the minimum sample size of 30 and are not reported. The direction is clear within the segments that did: smaller gaps go with higher exit, not lower.

This does not mean engagement causes attrition. The plain reading is that the employees who are weighing a decision keep participating — they remain present in the process — while infrequent responders are a more settled, lower-turnover group. Either way, a threshold of missed days is not a usable warning signal here.

Finding 3 — Lower happiness, but no warning dip

The mood-then-silence hypothesis predicted a two-stage pattern: happiness declines first, then responses go silent. We measured average happiness in four windows before exit — 91–180, 61–90, 31–60, and 0–30 days — and compared exited employees against retained controls over the equivalent recent windows.

Happiness was the one place the hypothesis found support. Exited employees were clearly less happy: their average answer sat at 1.04–1.06 against 0.87–0.91 for retained employees, on a scale where lower is happier. They logged fewer very-happy responses and more unhappy ones. But the predicted decline never appeared.

The happiness gap is real and it is sizable — exited employees gave roughly 15% fewer very-happy responses than retained employees. What matters for monitoring is the shape: the gap is constant across all four windows. It does not open up in the final 30 days. It was the same gap at 91–180 days before exit as it was in the last month. There is a slight uptick in unhappy responses approaching exit (4.0% versus 3.5%), but it is small and does not amount to a progressive slide.

Response volume does drop sharply approaching exit — total responses fell about 70% across the windows, from 31,769 down to 9,511. But this is cohort arithmetic, not disengagement: as the window narrows toward each person's exit date, fewer employees are still inside it. The people who remain in-window keep responding at the same rate.

What this means

The three findings point the same direction. The thing that separates people who leave from people who stay is not a change in behavior. It is a steady, low-grade unhappiness that coexists with full engagement. The employee at risk is not the one who went quiet. It is the one who keeps answering and keeps saying they are not happy.

That has direct consequences for how a retention signal should be built.

A practical version of the rule: stop monitoring for the silence and start monitoring for the response gap — the difference between how engaged someone is and how happy they are. An employee in the top engagement band whose happiness sits below their team's baseline is the profile this study would flag. Engagement on its own is a measure of compliance, not contentment. Neither metric is useful alone. The two together are.

Limitations

• Association, not causation. Lower happiness is correlated with exit; this study does not establish that it drives the decision, nor rule out a common upstream cause.
• Reasons for exit are not separated. Voluntary resignations, involuntary terminations, and end-of-contract departures are pooled under a single exit date. The patterns may differ across those groups.
• Truncated gap categories. The 14–29 and 30-plus day gap segments fell below the n ≥ 30 minimum and could not be reported, so the threshold analysis covers only the smaller-gap range.
• Tenure floor. The 180-day minimum excludes early-tenure attrition entirely. Findings describe employees who stayed at least six months, not new-hire churn.
• Happiness is self-reported. The check-in measures stated daily feeling, which is sensitive to mood, recency, and how comfortable a person feels answering honestly.
• No downstream test of the response gap. The engagement-minus-happiness signal is proposed from these findings; it has not yet been validated prospectively as a predictor.