Early warning indicators: identifying highly risky customers
A credit card company observed a worrying trend of high-risk customers who were suddenly defaulting on their payments within a month, and then fleeing the country after maxing their limit. The credit card company was based in a city hub for tourism and international business, so they needed a way to spot these risky customers who could default and flee early to intervene.
However, the company’s current early warning models were not attuned to rapidly evolving market trends. The risky behavior was observed across customers, but without attributes to serve as early warning indicators, the data could not be used effectively to identify which accounts were high-risk.
It was important to identify high-risk accounts early and accurately. False positives could create operational challenges for the company and have adverse impacts on the customer experience.
Neal Analytics built a custom, predictive model to analyze the credit card company’s recent customer performance data and identify which accounts were “high risk.”
The model was a decision forest binary classifier that used machine learning to identify patterns and spot indicators of high-risk accounts. This model looked at customer performance data to spot trends based on certain attributes such as purchasing behavior and demographic information.
To identify high-risk accounts, Neal Analytics built the model on attributes such as demographics, customer transactions, early pre-authorized transactions, behavioral patterns, account and payment history.
Neal Analytics also created attributes with advanced feature engineering. These attributes were based on business insights and data trends observed by the company’s subject matter experts, which helped the model identify more indicators of risky behavior.
By leveraging machine models and customer data, the credit card company could proactively identify high-risk accounts before they defaulted and left the country. This helped the company reduce losses and improve their overall efficiency and accuracy in identifying risky transactions and signs of a sudden default.
The model’s risk prediction score helped the company prioritize account and improve their collection efforts by reducing the number of false positives, which freed more resources to target high-risk accounts and perform strategic interventions.