BUSINESS SITUATION
- Our client, a major credit card issuer in the US, was facing issues with payment holds, and consequently, sales declines were increasing without a corresponding reduction in fraud.
- A complex set of strategies, a legacy to the company, was not up to the mark to manage fraud.
SGA APPROACH
- Analytics Engagement
- Preliminary segmentation used to build key segments of customers based on spend patterns
- Features for analysis created from multiple data sources:
- ML techniques like XGBoost and LightGBM were used to identify top predictors of transaction fraud
- Simulation analysis used to optimize cut-offs for predicted positives using data on the sales decline rate
ENGAGEMENT
SGA’s engagement comprised two data scientists and one data engineer working in an Agile framework.
BENEFITS & OUTCOME
- A complex set of 50+ segments of strategies replaced by clearly delineated sets
- Easy to comprehend, resulting in quicker buy-in from stakeholders
- Less complexity helped with faster deployment
KEY TAKEAWAYS
- Build and implement robust ML supervised and unsupervised models and strategies to improve fraud capture
- Enhance customer experience by reducing false positives
- Reduce sales declines from good customers affected by fraud rules
- Build detailed tracking mechanisms for the implemented strategies
