Stockertim Dictionary · Human resources · Technology
Attrition prediction
Also called turnover prediction, flight-risk scoring, attrition model
In short
Using a statistical or machine learning model to estimate how likely individual employees, or groups of employees, are to leave within a set period.
- Advanced
- Technology
- Checked
In detail
Attrition prediction looks for patterns in past leavers, such as time since last promotion, pay relative to the market, manager changes, commute length or engagement survey results, and uses them to score current staff. At group level it supports workforce planning: if one team is likely to lose a fifth of its people next year, recruitment and training can start early.
Scoring individuals is far more sensitive. A 'flight-risk' label can change how a manager treats someone, for instance by holding back training or a stretch project, which can make leaving more likely. Inputs can also act as proxies for protected characteristics: absence patterns may reflect disability or caring responsibilities. Predicting behaviour from personal data is profiling, and the employee is entitled to know about it.
Accuracy needs honest measurement. Because most people do not leave in a given year, a model that predicts nobody will leave can look accurate. Precision and recall, checked by group, say far more. Many organisations limit individual scores to a small set of trained users, use them only to prompt supportive conversations and never feed them into pay, promotion or dismissal decisions.
Worked example
A made-up bank, Northgate, has 1,000 staff and 100 leave in a year. Its model flags 80 people, of whom 40 leave.
- Precision: 40 / 80 = 0.5
- Recall: 40 / 100 = 0.4
Half the flags are right and the model finds 40% of leavers: useful for planning, too weak to label individuals.
In practice
People analytics teams usually publish attrition forecasts by team, role and location, and treat individual scores as restricted data with a documented purpose, a data protection impact assessment and regular bias checks.
Watch out
- High overall accuracy can hide a model that misses most leavers.
- A score used to deny opportunities can create the very departure it predicts.
Learn it properly
Official sources
- Guidance on AI and data protection · Information Commission (UK)
- What do we need to know about accuracy and statistical accuracy? · Information Commission (UK)
Facts last checked on 3 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.