Predicting High-Performer Turnover: Signals and Data

Answering honestly the question “Can we know in advance, from data, when a high performer will leave?”, accurately predicting a specific individual’s actual departure is difficult, and understanding attrition risk through changes in several signals is more realistic. A number like “83% probability of leaving” sounds attractive, but human behavior isn’t that simple.

In this piece, I’ll lay out the difference between turnover intention and actual turnover, why the attrition of high performers costs more, what early signals we can observe, and the privacy issues that must be considered alongside all of this.

01. Turnover intention and actual turnover are different

Research commonly measures turnover intention, that is, “Do I intend to leave this company?” Across many studies, it has been repeatedly observed that the higher the burnout, the higher the turnover intention tends to be.

But there’s an important trap here.

Turnover intention ≠ actual turnover

That turnover intention is related to actual turnover, and that a person with high turnover intention will necessarily leave, are two different stories. Far more variables intervene in actual turnover.

• Compensation and promotion prospects
• Leadership and peer relationships
• Career growth opportunities
• Job offers from other companies
• Labor market conditions
• Family and personal events

Some people stay for economic reasons even when severely exhausted, and some leave for a better opportunity even without much burnout. So it is more accurate to see turnover intention as one signal for understanding attrition risk. The research basis for this relationship is covered in more detail in Burnout and Turnover Intention.

02. Why does the attrition of high performers cost more?

Not every departure creates the same cost. There are reasons the attrition of high performers is especially burdensome.

• Concentration of tacit knowledge — key members often carry undocumented customer context, decision-making history, and informal networks along with them.
• Leadership gap — even without a formal title, when someone who served as the team’s standard of judgment or the center of problem-solving leaves, the team’s pace can slow noticeably.
• Ripple effect — the departure of a trusted member can affect the engagement and morale of the people who remain.

These costs don’t leave an invoice the way recruiting fees do, but they actually arise in the form of project delays and rework. The overall cost structure that burnout creates is covered in more depth in The Cost of Burnout.

03. Early signals: resignation is an event, but what precedes it is a process

Even if a member resigned on a specific date, that person’s experience may have been changing much earlier.

Rising workload → fatigue that doesn’t recover → declining engagement → increased cynicism toward work → reduced voluntary participation → job searching → resignation

Of course, not every departure follows this order. The reasons and the process differ from person to person. But this perspective prompts a question. Why are we looking only at the final event, “resignation”?

Turnover rate is a lagging indicator. It is measured after something has already happened. By contrast, changes such as engagement, sense of recovery, and attitude toward work are closer to leading indicators that can appear before the outcome. That said, interpreting them as definitive precursors is dangerous. You have to view several signals together with the organizational context.

04. Why baseline change rather than an absolute score

Here SYMPLE’s core principle comes in. We look not at comparisons between people but at change relative to the individual’s personal baseline.

For example, consider two people.

	Usual	This week	Change
Minsoo	70	72	+2
Jiwoo	30	58	+28

Looking only at the absolute score, Minsoo is still higher. But looking at the amount of change, the change that needs attention right now may be appearing in Jiwoo.

The higher the performer, the more this perspective matters. When a person who usually delivers high performance quietly begins to move in a different direction from usual, a system that looks only at the absolute score easily misses it. So SYMPLE cares about the individual’s baseline, formed through repeated check-ins, and the longitudinal change from it over time.

Here, voice can be used as one supplementary signal. When a person’s state changes, changes may also appear in the way they speak. That said, you cannot judge resignation or burnout from any single voice feature. The perspective that voice is a supplementary signal rather than a single decision-maker is laid out in the Voice Biomarker Guide.

05. A problem that comes before prediction: trust and privacy

Even if it were technically possible to estimate an individual’s probability of leaving, a problem arises the moment you put it on an HR dashboard as “Kim OO · high likelihood of leaving.”

It’s hard for members to trust such a system. And without trust, the quality of the data drops too. The more exhausted a person is, the less honestly they respond to a system that feels like surveillance.

So SYMPLE tries to design predictive accuracy and privacy protection at the same time.

• For the individual — the self-understanding that “my state has recently been diverging from usual; this may be a point where recovery is needed.”
• For the organization — group-level signals that cannot identify any individual. For example, in the form of “signals related to workload have been steadily increasing in a certain team recently.”

To be honest, we don’t aim to guess who will leave. Rather, we think what matters more is how early we can detect that a change different from before is beginning in a person and an organization, and whether we can provide an intervention that actually helps at that moment. Why we research this problem can be found in About SYMPLE.

References

1. World Health Organization. Burn-out an “occupational phenomenon”: International Classification of Diseases. 2019.
2. Maslach C, Schaufeli WB, Leiter MP. Job Burnout. Annual Review of Psychology. 2001;52:397–422.
3. Swider BW, Zimmerman RD. Born to burnout: A meta-analytic path model of personality, job burnout, and work outcomes. Journal of Vocational Behavior. 2010;76(3):487–506.
4. Maslach C, Leiter MP. Understanding the burnout experience: recent research and its implications for psychiatry. World Psychiatry. 2016;15(2):103–111.