Use Case: How One Subprime Lender Used AI to Reduce Losses
Risk models eventually become less predictive or relevant due to evolving market conditions. New and improved versions routinely replace the older models, but sometimes there are game-changing model innovations that can potentially redefine credit risk decisioning. This is one of those cases.
A deep subprime lender, who has long been an Equifax customer, wanted to replace a current core model that it was phasing out. The lender understood the difficulty in trying to accurately differentiate risk when working with a deep subprime audience. Therefore, it wanted to leverage artificial intelligence. Traditional modeling techniques can simplify risk criteria to the point that it’s not helpful in the deep subprime arena. Therefore, the lender asked Equifax to build a new risk model with Equifax data and attributes.
To capture more “goods” and fewer “bads” relating to risk decisions, Equifax leveraged its new AI-enabled NeuroDecision® Technology modeling technique. It uses complex non-linear attribute interactions for deeper learning of consumer behavior. However, it still provides the required explainability through reason codes for the consumer. These codes are helpful for finding differentiators among homogenous populations like super subprime.
Equifax also used a mix of data to help the lender better segment risk levels among its deep subprime customer base, including:
- Customer data provided by the lender was used to automatically determine if a “new” applicant was an existing or past customer.
- Trended data from Equifax’s consumer credit database was used to reveal an applicant’s financial trajectory by showing their financial behaviors over 24 months.
- Alternative data from the utilities, communication and pay TV industries was used to provide expanded insight into how people pay their household bills.
- Peak attributes offered a premium set of tri-bureau enabled consumer credit attributes to help support the lender’s highly specialized needs. Interactive attributes use AI-enabled technologies to create “super attributes,” which fuse multiple attributes together to create one more powerful and predictive attribute.
Fresh new data sources leveraged within an advanced, machine learning model helped the lender approve 92,000 more accounts without increasing losses. They also delivered:
- $13.7 million in annual loss savings
- A 92 percent lift in KS¹
- A 7 percent lift in the retailer’s scoreable rate. This is a notable improvement when dealing with deep subprime consumers. That’s because lenders can’t score many of them due to lack of credit. The more customers the model can score, the more customers it can confidently approve and serve.
Ready to Innovate Your Credit Decisions?
Read our white paper, Innovate Your Credit Decisions, for more use cases. And if you’re ready to innovate, we provide three innovative ways to help expand credit access.
1 KS: The Kolmogorov-Smirnov (KS) statistic is the measure of the maximum distance (greatest separation) between the cumulative % of high spend (goods) and the cumulative % of low spend (bads) across all score ranges. It represents the model’s ability to differentiate “goods” from “bads” in the sample. Higher values indicate better overall separation and a stronger model.
Recommended For You
Since the last U.S. recession in 2008, financial risk management has seen significant changes. Lending requirements are tighter, verification procedures […]
This question sounds absurd. But this is one of the most important questions lenders must ask themselves when making consumer […]
Calibrating premium to risk is crucial to your business – and the business of daily life for your customers. What […]
The autumn air is crisp, and the delicious smells from the nearby Halal Guys cart are tempting my senses. It’s […]