Decoding flight risk: predictive attrition analytics and retention modeling
An employee flight risk model is a statistical model that estimates the likelihood of voluntary departure from HR data such as tenure, pay position, time since promotion and manager changes; attrition analytics is the wider practice of measuring and forecasting turnover.
"Flight risk" is the estimated likelihood that an employee will leave voluntarily within a set period. Predictive attrition models produce those estimates from HR data, and many HR platforms now offer them. Used well, they help an organization find where retention effort matters most. Used carelessly, they mislabel people and erode trust. This guide explains how they work and how to use them responsibly.

Start by measuring turnover correctly
Before predicting turnover, measure it consistently. A common definition is leavers in a period divided by average headcount in that period. Separate voluntary departures (resignations) from involuntary ones, since only the first is what flight risk models try to predict. The U.S. Bureau of Labor Statistics uses the same distinction in its JOLTS survey: quits are voluntary separations, reported separately from layoffs and discharges.
Break the rate down by department, manager, role and tenure. First-year turnover is often a different problem, usually about hiring and onboarding, from turnover among experienced staff.
How predictive attrition models work
Most models are classification models (logistic regression, decision trees or similar) trained on past employees: who left and who stayed, with what they looked like beforehand. Typical inputs include tenure, time since last promotion or pay change, pay position within range, manager changes, team turnover, commute or location, performance ratings and survey responses where available.
The output is a probability for each person or group. Two cautions apply. First, the model learns from the past, so a change in the labor market or the organization weakens it. Second, a probability is not a prediction about a person: a group with a 20% risk will mostly stay.
What research points to
Turnover research in I-O psychology has long pointed to a fairly stable set of drivers: job satisfaction and organizational commitment, the relationship with the immediate manager, perceived fairness of pay and promotion, growth opportunities, workload, and the availability of alternatives in the labor market. Gallup's research found that about half of voluntary leavers said their manager or organization could have done something to prevent them from leaving.
That is the useful lesson for a model: its value lies less in the score than in pointing to the controllable drivers, such as stalled progression, a manager with repeated departures, or pay below range, where action is possible.
Leading indicators that signal a retention problem
- Rising first-year turnover, or resignations clustering under particular managers.
- Long gaps since last promotion or pay change for strong performers.
- Pay below the role's range, or below market for hard-to-fill roles.
- Falling engagement or survey participation in a team.
- Rising quits in your industry, which BLS JOLTS data shows month by month.
How to identify flight risk employees without a model: an employee flight risk assessment matrix is a simple version, rating roles or groups on two axes, the likelihood of leaving (from the indicators above) and the impact if they left (criticality, time to fill). Groups high on both get attention first. Rate groups, not individuals, for the fairness reasons below.
Attrition forecasting applies the same data to the future: expected leavers by role over the next year, which feeds pipeline forecasting.
Using flight risk fairly
- Act on groups and drivers, not labels on individuals. Fix the stalled promotion path rather than flagging the people on it.
- Never use a risk score to deny opportunity. Withholding development or promotion from someone because a model thinks they might leave is unfair and becomes self-fulfilling.
- Exclude protected characteristics and check for proxies. Inputs that stand in for age, sex, race or disability, and outcomes that differ by group, need review.
- Be transparent with employees about what data is used and why, and follow the privacy laws that apply where you operate.
- Validate the model on recent data before relying on it, and recheck regularly.
From model to retention plan
Before building a model at all, check whether simpler analysis already answers the question. Turnover broken down by manager, tenure, pay position and time since last promotion often shows the same hot spots a model would, with no black box to explain. Exit interviews and stay interviews add the reasons behind the numbers, though exit interview answers tend to be guarded, so look for patterns rather than single comments.
A good retention plan targets the drivers the data surfaces: stay conversations by managers, clear progression paths, pay reviews for people below range, and support for managers whose teams lose people repeatedly. Measure the effect against the baseline turnover rate; measuring the ROI of people analytics explains how.
What BlissPoint offers, and what it does not
Organization Viewer does not produce individual flight risk scores; its attrition analytics work on groups. Its turnover dashboard does the measurement that comes first: the 12-month turnover rate, resignations against people let go, turnover by department and by manager, and first-year leavers, with small teams grouped so no individual is exposed. The pay range review lists people paid below their role's range, one of the controllable drivers above, and the span and layers health check flags managers with very wide spans. An I-O psychology consultant can review the patterns with you in the optional quarterly review, $750 a quarter, and the Employee Engagement Survey Program measures the attitudes behind turnover.
Questions
What is a flight risk score?
An estimated probability that an employee will leave voluntarily within a set period, produced by a model trained on past departures.
How accurate are attrition models?
Accuracy varies with data quality, the time horizon and how stable the organization and labor market are. Validate any model on recent data before relying on it, and treat scores as group-level signals.
Is it legal to predict which employees will leave?
Rules depend on where you operate, especially privacy and anti-discrimination law. Exclude protected characteristics, check for proxies and adverse impact, be transparent and get legal advice for your jurisdictions.
What data do attrition models usually use?
Common inputs are tenure, time since last promotion or pay change, pay position within range, manager changes, team turnover and, where available, survey results. Protected characteristics should be excluded and proxies for them checked.
Should managers see individual flight risk scores?
It is generally safer to share group-level patterns and drivers. Individual labels can bias decisions about the people flagged and are easily misread as certainties.
Does Organization Viewer predict who will leave?
No. It measures turnover by department, manager and tenure, and surfaces related signals such as pay below range, but it does not score individuals.