May 2025
May 2025
This project focuses on building a predictive machine learning system that identifies Relationship Officers (ROs) at ABSLI who are at elevated risk of attrition. It does so by analysing historical performance patterns, engagement trends, and employment histories. Early identification of ROs likely to leave allows the organization to implement targeted retention strategies, improve workforce stability, and minimize disruptions to sales operations.
Data Collection: Historical data on Relationship Officers (ROs) was gathered, including demographic information, tenure, sales performance, target achievement, incentives, and promotion history, to serve as inputs for the attrition model.
Feature Engineering: Relevant features were created, such as monthly performance trends, NOP (Number of Policies) sold, incentive consistency, promotion flags, target achievement percentages at various checkpoints, and vintage (tenure) classifications to effectively capture early warning signs of attrition.
Model Development: Multiple classification machine learning algorithms, including logistic regression, random forest, and XGBoost, were trained to estimate the probability of each RO leaving the organization.
Model Validation: Models were rigorously validated using in-time and out-of-time test sets to ensure predictive reliability and generalizability. Key metrics such as F1 score, precision, recall, accuracy, AUC, and third-decile capture rates were used for evaluation.
Deployment: A scoring pipeline was created to assign attrition risk scores to current ROs on a regular basis. High-risk individuals were flagged for proactive retention strategies, such as additional support, training, or engagement interventions.
Behavioural and External Factors: External influences such as job market conditions, personal health issues, or life events can significantly impact an RO’s decision to leave but are difficult to quantify and require sophisticated modelling or proxy variables.
Dynamic Workforce Profile: The evolving nature of the RO workforce—with frequent changes in roles, performance expectations, and market dynamics—necessitates ongoing model retraining and recalibration to maintain accuracy and relevance.
Engagement and Retention Strategy: Predicting attrition is only one part of the solution; designing and executing personalized retention strategies for high-risk ROs is crucial and requires alignment between business, HR, and leadership teams.
Improved RO Retention: Early identification of high-risk Relationship Officers enables targeted interventions, reducing voluntary attrition and retaining experienced sales talent.
Optimized HR Efforts: Human resource and managerial attention can be focused on at-risk ROs, improving the efficiency and impact of retention strategies.
Increased Sales Continuity: Retaining performing ROs ensures consistent sales output and minimizes disruptions in customer acquisition and service.
Stronger Employee Experience: Personalized support and timely engagement improve job satisfaction, fostering a more motivated and loyal workforce.
The RO Attrition Prediction Model, deployed since the start of FY25, supports workforce stability by enabling HR and business teams to proactively manage attrition risks. In JFM 2025, it captured an exceptional 66% of actual attrition within the top 3 deciles, allowing targeted interventions on high-risk segments and optimizing efforts and impact. Ongoing improvements aim to further enhance its effectiveness.The model predicts RO attrition for JFM 2025 with strong accuracy, identifying 530 out of 836 actual attritions within the top 3 deciles. These deciles rank ROs by predicted attrition risk, with the 1st decile indicating the highest likelihood.
