Rishav Ganguly

Acquisition risk scorecard (generic) validation for the microfinance portfolio

• Conducted an end-to-end validation of an acquisition scorecard using diverse data (credit, utility, telco, short-term credit lines), ensuring regulatory compliance and robustness.
• Validated model integrity via assessments of data sampling, segmentation, and feature selection, and tested out-of-time performance using KS, Gini, and rank order for logistic regression and XGBoost models.
• Verified regulatory adherence (adverse action codes) of the model and score calibration.
• The validation recommendations ensured closure of all medium and high severity observations associated with the model.

Synthetic identity fraud model (generic) validation

• Validated synthetic fraud model data preparation: credit history pull, feature engineering (demographics, account data), and early delinquency target variable creation.
• Validated the XGBoost model by developing alternate models with tweaked hyperparameters and various combinations of independent variables to ensure the best possible model.
• Assessed model stability with out-of-time sample validation and checked the interpretability using SHAP values.
• The validation recommendations ensured a fast and efficient model deployment for early-on revenue generation.

Strategic Asset Management & Leadership

• Led the design and implementation of the organization-wide 'System of Records (SOR)' asset inventory, a strategic UI for CXO-level reporting on product development, model maintenance, and asset lifecycle management, collaborating knowledge exchange across 8 global geographies.
• Defined key performance metrics and user interface aesthetics for the SOR, enabling data-driven decision-making and ensuring scalability for future business needs.

Personal Loan Cross Sell Acquisition Scorecard Development

• Created an acquisition scorecard for an Indian NBFC, optimizing personal loan approvals for existing consumer durable clients by predicting early delinquency.
• Conducted portfolio analysis (excluding COVID-19 impact), risk assessment (roll-rate, vintage), feature engineering (demographic, performance), segmentation, and variable selection (VIF, p-value).
• Developed a logistic regression scorecard, calibrated scores, and defined lending strategies.
• Received Wizard@Work award for contributions to the project.

Behavioral model validation for a leading US based credit bureau

• Assessed the data quality, choice of bad definition and its alignment with the objective of the model, independent variable selection and final model performance.
• Analyzed model performance using PSI, Characteristics analysis, and KS statistics, identified risks associated with independent attributes and segmentation schemes, and provided recommendations to clients.
• Prepared comprehensive documentation with insights from the model results.

End-to-end development of a collections risk profile for a leading middle-eastern bank to mitigate roll forward risks associated with multiple portfolios across the institution

• Performed EDA, feature engineering (delinquency, repayment, demographics), and segmented the portfolio.
• Built risk models using CHAID decision trees (low volume segments) and XGBoost.
• Implemented monthly model refreshes and collection channel tagging, resulting in case resolutions from 86% to 93%.
• Received Rising Star award for project contributions.