What employees complain about before they quit

Most retention dashboards watch the mood score. They flag the employee who logs a "Not Great" or a "Terrible" and assume the negative response is the warning. It is a weak one. When we compared 4,532 employees who later exited against 3,185 who stayed, the gap in negative mood was real but small: exited employees logged a 14% higher mean negative rate (3.75% versus 3.29%). That is not a number you can run a retention program on.

The signal lives one click deeper. After a negative check-in, Happily asks an open follow-up question. We analyzed 7,828 of those text responses and classified them into themes. The finding is that the exit risk does not depend on whether an employee feels bad. It depends on what they feel bad about. A complaint about a manager and a complaint about a head cold are both "negative" — and they predict opposite outcomes.

How we measured it

This study has two halves. The quantitative half asks whether negative mood predicts exit at all, and how the negative rate moves over time. The qualitative half classifies the follow-up text written after negative check-ins, then measures the exit rate associated with each theme.

Mood is recorded on a five-point scale through the daily "How do you feel today?" check-in. A response of "Not Great" or "Terrible" counts as negative. Each negative answer can carry a free-text follow-up; we linked those texts to the same-day negative answer. Themes were built bottom-up from the data through iterative review, then applied by keyword classification. Each response was assigned one primary theme. The exit rate for a theme is the share of employees writing that theme who later left, compared against the 30.8% base exit rate in the text sample.

The text sample skews English: only 2,490 of the 7,828 responses (31.8%) were in English and could be classified. The qualitative findings apply most cleanly to English-speaking employees. The mood findings cover the full population.

Finding 1 — Negative mood barely predicts exit

Start with what does not work. Employees who later exited were slightly less happy than those who stayed, but the gap is thin. The mean negative rate was 3.75% for leavers against 3.29% for stayers, a 14% difference. The mean happiness score differed by 0.06 points on the five-point scale.

The more revealing pattern is where the gap sits. It is not at the bottom of the scale. Exited employees were 5.7 percentage points less likely to answer "Very Happy" and 4.1 points more likely to answer "Okay." The two genuinely negative responses, "Not Great" and "Terrible", together accounted for only 1.3 extra percentage points.

Attrition shows up as erosion, not collapse. Employees who eventually leave do not suddenly become unhappy. They drift from "Very Happy" to "Okay" while their negative-response count stays low. A dashboard tuned to catch "Terrible" answers will miss most of them.

The trajectory adds one more wrinkle. Among employees with data in all four quarters before exit, the negative rate climbed 22% from a year out to the 91-180 day window (2.77% to 3.37%), while retained employees held flat near 2.5%. Then, in the final 90 days, the leavers' negative rate dropped to 3.14%. This matches our earlier "Engaged Exit" finding: once the decision to leave is made, emotional investment falls and mood can briefly improve. The mood signal is at its weakest in the exact window when the exit is closest.

Finding 2 — The theme is the signal

Now the half that works. When an employee writes a follow-up after a negative check-in, the theme of that complaint sorts them into sharply different risk groups. We classified the English-language responses into 15 themes and measured the exit rate for each.

Three themes form a danger band. Complaints about a manager (62.9%), about a team or colleagues (61.9%), and about a lack of growth or direction (60.7%) all carry exit rates near double the base rate. These are the relational and structural complaints — they are about people and systems that do not change on their own.

Compensation breaks the intuition. Pay and financial complaints carry a 22.2% exit rate, well below the base rate. This lines up with long-running retention research: pay dissatisfaction on its own rarely drives an exit. It is pay combined with a bad manager or no growth path that does. An employee who complains only about money is often engaged enough to stay.

The safe complaints

The other side of the table is just as useful. Physical health (20.4%), personal and family issues (20.6%), and accidental button presses (19.5%) all carry below-baseline exit rates. An employee who logs "headache and stomach ache" or "personal issues — death of a friend" is experiencing life, not workplace dissatisfaction. Safe complaints are situational and temporary: the employee knows the source, it sits outside the organization's control, and they often expect it to pass. Danger complaints are relational and structural.

One number to keep in mind when tuning alerts: 5.1% of all follow-up responses were accidental presses or from people who stated they were actually fine. That is a real false-positive floor in any mood-based system.

Finding 3 — The honesty paradox

If negative mood is a weak exit signal, the absence of negative mood is a stranger one. We binned 4,428 employees by their overall negative-response rate and measured the exit rate of each bin.

The lowest exit rate, 23.0%, belongs to employees who are occasionally negative — the 0-5% bin. Employees who are never negative exit at 31.3%, a higher rate. The always-positive employee is not the safest employee.

Two readings fit. The first is disengagement masquerading as positivity: an employee who never expresses negativity may not be genuinely happy, just emotionally checked out and tapping the positive button. The second is that healthy negativity is itself a retention signal. Employees who occasionally voice frustration but mostly feel good are invested enough to say when something is wrong and resilient enough to recover. That is the psychological-safety pattern — people who feel safe enough to be honest.

The bins also mark a hard threshold. Once negative responses pass 20% of the total, the exit rate jumps to 40.8%, nearly double the healthiest bin. When one in five daily check-ins is negative, the employee is in the danger zone.

Finding 4 — The 90-day manager surge

The single strongest signal in the study is not a level — it is a shift. Among exited employees, we tracked how the mix of complaint themes changed as their exit date approached.

Manager complaints rise 4.3× — from 4.0% of an exited employee's complaints a year out to 17.4% in the final 90 days. No other theme moves like this. Physical health complaints actually fall as exit approaches (7.5% to 5.4%), and workload stays flat. The shift is specific: it is the manager relationship, not the workload, that becomes the dominant grievance in the run-up to departure.

The arc plays out in the text. Early on, an exited employee wrote a two-word plea: "Guidance instead of criticism." Months later, the same person wrote a detailed account of being cornered with questions while colleagues were trusted, ending with "Learn to accept unfair treatment is normal." In the final 90 days, the writing turned to documenting blame meeting by meeting. The progression — from wanting help, to coping, to building a case — is the textbook exit trajectory. By the time the complaints become this detailed and self-referential, the decision has usually been made.

The management adage holds: people leave managers, not companies. The data puts a clock on it. The move from general unhappiness to manager-specific unhappiness takes roughly six months, and once manager complaints dominate the follow-up text, the employee is typically within 90 days of leaving.

What this means

The practical takeaway is to stop treating the mood score as the alert and start treating the follow-up text as the intelligence. The score tells you someone had a bad day. The text tells you whether they are leaving.

For managers, the rule reduces to two lines. If a team member's follow-ups have shifted toward how you treat them, that is the alarm, not their mood score. And "support instead of criticism" is a direct quote from people who left — when you give feedback, lead with coaching.

Limitations

• Correlation, not causation. Manager complaints may be a symptom of an exit decision already made rather than its cause. Employees who have decided to leave can become more critical of their manager in hindsight.
• Language coverage. Only 2,490 of 7,828 text responses (31.8%) were in English and classified. The remaining 68.2%, primarily Thai, were not. The qualitative findings apply mainly to English-speaking employees.
• Text classification is coarse. The keyword approach left 58.4% of English responses as "work tasks / unclear" — task descriptions or vague statements that did not name a source of distress. A stronger NLP method could reduce this.
• Self-selection in the follow-up. Not every employee who answers "Not Great" writes a follow-up. Those who do may be systematically more engaged, more frustrated, or more articulate than those who do not.
• The accidental-press problem. 5.1% of follow-up responses were accidental taps or from people who were actually fine, inflating the raw negative count.
• Survivorship in the retained cohort. The retained group includes only employees still active at analysis time. Some will exit later, which understates the true exit rate for certain themes.