Predicting staff turnover before it happens for a Canadian retirement home group

The Challenge

The company was losing nursing and care staff faster than it could replace them. HR wasn't finding out someone was likely to leave until they handed in their notice — by which point the window to act had already closed. With qualified care workers in short supply across the sector, every departure meant a slow, expensive hire. The goal was simple: know earlier.

Our Approach

The analysis started with the workforce data the company already had — 34 attributes covering tenure, role, performance ratings, shift patterns, distance from home, and various engagement signals. A few patterns came through clearly: single employees left at higher rates, nurses were disproportionately represented in departures, and attrition spiked at the two-year mark.

Multiple models were tested to find the best predictor of individual attrition risk. Given the imbalanced dataset — most employees stay — AUC was the right performance measure. Logistic regression came out on top after tuning, with a score of 0.86.

The model was deployed on the company's internal server. HR managers now open a dashboard each week that ranks every active employee by attrition probability, shows what's driving the score, and flags where to step in.

Key Deliverables

• Exploratory analysis across 34 employee and behavioural attributes
• Multi-model evaluation using AUC as the primary performance metric
• Trained attrition prediction model with an AUC of 0.86
• Internal deployment with an employee risk-ranking dashboard for HR
• Automated flags for high-risk employees within existing HR workflows

“We can now see who might leave before they decide to. That has changed how we think about retention.”

— VP Human Resources, Canadian Retirement Home Group

Results

Engagement

Facing a similar challenge?

Tell us about your data problem. We'll help you find the right path forward.