Predicting people is useless; predicting conditions is actionable
Models that return a list of at-risk employees create discomfort and little action. Models that return combinations of conditions — shift, supervisor, commute, tenure in role, pay gap against the local market, learning curve — let you act on the cause instead of the symptom.
Variables that almost always explain
In labor-intensive operations, a handful of variables explains most of the variance.
- The first ninety days: most turnover concentrates there
- Direct supervisor quality, measured by turnover in their own team
- Pay gap against the local market, not the national average
- Schedule predictability and shift stability
- Commute time and site logistics
Governing the data and its use
A people model demands explicit rules: which variables are excluded for bias or legal risk, who can see the output, which decisions the model cannot make on its own. Without that frame, the project dies at the first compliance objection — deservedly.
From model to P&L
The final conversion is economic: recruiting cost, lost productivity curve, coverage overtime and, in operations, service availability. Expressed in those units, the model stops being an HR project and becomes a business decision.
Key takeaways
- The value is in predicting conditions, not people.
- The 90-day cliff concentrates most avoidable turnover.
- Without translation into cost, no people model survives the committee.
NS frameworks and reference sources
- MIT Sloan Management Review — People analytics practice
- McKinsey — The State of Organizations
- BCG — Predictive workforce analytics
This article develops proprietary NS Business Strategy frameworks, drawing on our project base and on public industry literature and studies cited above. Figures are reference ranges; each project is measured against the client's actual data.