The client a niche financial firm offering easier lending and payment solutions for e-commerce transactions. It offers small and short term lending options to its consumers. The client wanted to devise a credit risk scorecard using machine learning to identify potential credit risk associated with each consumer and reduce credit fraud.

• To devise an ML-based credit risk scorecard to capture the complex buying behavior of different customers.
• To segment customers based on different buying patterns.
• To create customer-specific exposure levels based on their risk profile
• To derive cutoff levels for credit approval/decline decisions based on predicted consumer risk levels.

• We segmented consumers based on their purchase patterns (recurring and new) and engineered several features to capture their exposure levels and outstanding installments.
• Several merchant-specific variables were created to capture peculiar trends associated with certain merchants.
• Gradient Boosting was used to accurately predict the default likelihood of customers.
• SWAP Analysis was conducted to calculate lending approval rates based on varying default/loss thresholds.

The client implemented the model-based risk scorecard and adjusted the lending approval rates accordingly to increase the user-base as well as screen users according to their risk profile.