Unexpected turnover is expensive: recruiting, training, know-how walking out the door. In banking, where roles require specific skills, the impact is even greater.
From HR data to prediction
By combining compensation data, salary reviews and HR interview outcomes, a churn prediction algorithm estimates the probability of resignation over the next six months. Data is transformed and modeled on Microsoft Fabric and PySpark; Power BI makes the results readable for HR.
From prediction to action
The value is not the number itself but the actions it enables: targeted salary reviews, conversations, growth paths, internal moves – and calmer HR planning for replacements and training.
The case study: Churn Analysis
Background
Unexpected staff turnover drives up costs: recruiting, training and loss of know-how. In banking, where roles often require specific skills and in-depth training, these costs can be particularly significant. To mitigate them it is essential to put effective retention strategies in place: professional development programs, competitive compensation policies and initiatives to improve the workplace climate.
Solution
The solution is based on a predictive churn algorithm designed to estimate the probability that an employee will leave the company within the next six months. Using data from the HR suite – including compensation data, salary reviews and HR interviews – the system makes it possible to anticipate departures and take preventive action, improving HR planning and strategy.
Objectives
- Predict resignations over the next 6 months
- Analyze structural causes of turnover (pay, morale, career growth)
- Implement corrective actions (salary reviews, HR interviews, internal moves)
- Support HR planning (replacements, training)
- Improve employee wellbeing and financial return
How it works
- Data acquisition: compensation data, salary reviews, HR interviews
- Churn prediction algorithm: probability of resignation over the next 6 months
- Data transformation and modeling
- Dashboards and reports in Power BI



