PRIYA SINGH

Issuer model

• Led end-to-end model development by performing thorough data exploration and feature analysis to create a multi-class XGBoost classifier for 12 issuer groups. Improved accuracy by 13% and ensured model transparency through Explainable AI techniques (SHAP, ALE, PDP) and proactive collaboration with governance stakeholders.

Model Governance:

• Spearheaded model governance for multi-class model. Implemented and documented continuous monitoring framework for model performance and annual explainability.
• Established rigorous tracking of input features and instituted proactive root cause analysis to quickly address deviations, ensuring model fidelity.
• Aligned governance practices with strategic business goals and fostered cross-functional collaboration, reinforcing data transparency and building stakeholder confidence in our predictive analytics.

APR intelligence:

• Reverse engineered a robust logic to compute APR from minimum due information available online for external credit cards and defined accuracy metric for this logic
• Partnered with the strategy team to leverage APR intelligence to get accurate interest revenue estimations for non-Amex cards and delivered insights recognized by senior leadership

BIN mapping & Fee model :

• Developed a robust multi-class fee model by first conducting comprehensive online research to map Bank Identification Numbers (BINs) to associated products. When faced with ambiguities, I implemented a fee equal balance analysis—leveraging the assumption that inactive cards exhibit distinct annual spikes—to ensure accurate fee determination.
• This fee prediction on Non-Amex cards utilised for strategic decision making, like which customers to target for premium products and which customers for lesser premium Amex cards

Wallet rank prediction model:

• Built a model to predict wallet rank of a newly acquired Amex card post 12 months of onboarding
• Analyzed model output for the use case of Line-assignment for customer cards and impact of incremental offers and incentives on likelihood of customer preferring Amex card

Customer Segmentation:

• Developed a predictive model using the k-Means clustering algorithm to group audiences into digital segments based on their historical channel activity data

Channel & Content affinity analysis:

• Analyzed segment-level data to uncover channel preferences and audience engagement trends, yielding actionable business insights. Deployed a topic modeling solution to pinpoint key customer content preferences and support strategic decision-making. Based on the results we have provided recommendations to the stakeholders which resulted in 30% more Conversion