Experts Provide Insights on Alternative Data, AI and Machine Learning

Adopting alternative data can provide many benefits to fintechs. During our monthly Market Pulse webinars, our audience submits questions to our expert panel. For our August Market Pulse webinar, our panel discussed the importance of alternative data in the fintech space. Experts included David Sojka, Risk Advisor at Equifax, and Brett Manning, Product Manager for Personal Loan Machine Learning for Origination at Upstart. Below are their answers on alternative data, AI, machine learning, and more.

Q: What are the best alternative data sources based on product? For example, what to use for a card, mortgage, or auto?

David Sojka: The correct answer, which is the answer no one wants to hear, is it depends. I say that because alternative means something outside of the norm that you're in. We talked about Fintech in this webinar. Traditional lenders, specialty lenders, and fintech companies all have different criteria for evaluating creditworthiness. Factors such as loss rates, approval rates, and booking rates have different meanings for those companies in terms of what they need to achieve. So lenders must determine which data signals are missing from their current strategies and explore alternative data sources to fill those gaps. Techniques like Chi-square tests and decision trees can be used to identify predictive factors for delinquency and approval rates. Think about what’s meaningful in credit decisioning - things like utilization, openings, delinquencies, and the overall profile of the consumer.

That’s why I say, it depends. There is no one-size-fits-all answer to the question of which alternative data is best. Lenders should work with their data vendors to identify additional sources that can provide lift in their decisioning models.

Q: How are you able to adapt machine learning/AI to remove systemic inequity for marginalized groups?

Brett Manning: Advanced underwriting systems are reducing the racial inequities that plague traditional lending systems. Lenders that assess only traditional and backward-looking variables lock historical biases into their decisions. Forward-looking underwriting models, however, collect and use nontraditional underwriting data to assess each applicant’s future potential. Machine learning models learn from repayment events in a virtuous cycle. When borrowers who would have been declined by a traditional lender take out a loan from an Upstart lending partner, their repayment events train our machine learning models for future applicants and help to drive more approvals for future applicants that would have been declined by a traditional lender.

This process repeats itself. Over time the model becomes increasingly fair in its outcomes. With Upstart AI, lenders can approve more borrowers at lower rates across races, ages, and genders. Our research showed that, in 2023, the Upstart model approved 116% more Black applicants at 36% lower APRs compared to the traditional model. For Hispanic applicants, the numbers were 123% and 37%, respectively. You can find more information in our latest Access to Credit Report here.

Q: Brett: What areas has Upstart had the biggest impact on lending? Lending to consumers that may have been declined? How have delinquencies been improved?

Brett Manning: We’re focused on improving access to credit for everyone. Our research shows that in 2023, in comparison to a more traditional underwriting model, the Upstart personal loan model approves 101% more applicants and results in APRs 38% lower.

For decades, traditional lenders used basic “scorecard” methods that combine credit scores and a handful of other variables to determine who is approved for a loan and at what interest rate. Unfortunately, these tools are limited in their ability to quantify risk. Artificial intelligence allows us to vastly expand the information used to inform a credit decision, thus delivering more accurate, efficient, and inclusive lending for all loan applicants.

 

*The opinions, estimates, and forecasts presented herein are for general information use only. This material is based upon information that we consider to be reliable, but we do not represent that it is accurate or complete. No person should consider distribution of this material as making any representation or warranty with respect to such material and should not rely upon it as such. Equifax does not assume any liability for any loss that may result from the reliance by any person upon any such information or opinions. Such information and opinions are subject to change without notice. The opinions, estimates, forecasts, and other views published herein represent the views of the presenters as of the date indicated and do not necessarily represent the views of Equifax or its management.