The class-action lawsuit against Navy Federal has financial institutions revisiting their fair lending practices and considering how AI fits in. For Amount’s director of decision science, Garrett Laird, the conversation is nuanced.
According to CNN, the close-to-29-point gap in Navy Federal’s approval rates was the widest among the top 50 mortgage lenders in 2022. It persisted when accounting for applicant income, debt-to-income ratio, property value, and down payment percentage.
The homeownership disparity is larger than it was in 1960. A white family is 55% more likely to own a home than a Hispanic family and 70% more likely than a Black one. Thanks to the power of home ownership, the average median wealth for a white family is $285,000, $61,600 for a Hispanic family and $44,900 for a Black one.
The questions raced through Laird’s mind when he heard the Navy Federal news. What was their underwriting process like? Was it more subjective or systematic? Did a systematic policy lead to disparities? What controls were in place to prevent this? Whatever the reason, it’s a significant failure, given the stark differences in publicized approval rates.
“At Amount, we pride ourselves on automating decisions, making them faster and easier,” Laird said. One of the benefits of that is greater control. You can remove human bias to some extent. There can still be bias in datasets, policies, and algorithms, but it can be empirically measured. You don’t have to worry about the human aspect as much there.”
“I want to know what those controls and processes look like at Navy Federal and how they got to where they are. Hopefully, some of that becomes public record.”
Most financial institutions still rely on the human element to some degree. Laird said they like to maintain a manual option, say, for a longstanding client. The danger comes if such exceptions are used for some groups more than others.
AI and fintech are massive opportunities for credit unions
Automated decision-making is becoming more prevalent. That’s one reason why Laird’s excited about Amount’s partnership with Velera (f.k.a. PSCU/Co-op Solutions). Amount supports Origination Solutions, Velera’s new digital lending suite.
Credit unions and smaller community banks have different mandates than the biggest institutions. They exist to serve the needs of their members. They are good at relationship-building but need help with automated scoring. Add that automation while maintaining the personal touch, and there’s massive opportunity.
Credit unions offer a much wider range of services than fintechs, which leverage the most modern technologies. When the two are combined, credit union members can benefit from a range of tailored opportunities.
Also read:
Separating the different types of AI
Laird cautions leaders to be careful about using AI. Yes, there are clear cases for more efficient document processing and data extraction. However, the fervor over large language models is misplaced because they lack explainability.
“There’s absolutely interest in AI for lending, but different flavors of AI,” Laird said. Logistic regression has existed for decades but still works and adds much value.
“We’ve seen that with other clients that use those models. For a fintech lender, I know they’re using more advanced algorithms for underwriting decisions. They’ve gotten comfortable with their ability to explain those outcomes and are pressing forward in that regard. So there’s still movement in that space.”
Laird sees credit unions and community banks as more interested in traditional machine-learning techniques that are explainable.
Amount’s Linear acquisition to soon bear fruit
In 2022, Amount acquired an SMB loan and account origination platform, Linear, for $175 million. The deal creates a company with strengths in consumer and commercial services.
Laird said the last 18 months have been spent combining the two companies. Once complete, customers will benefit from a broader range of products that can be offered when they need them. Perhaps a commercial loan applicant needs a small business checking account, or a consumer borrower would benefit from a credit card or refinancing.
“Having all that together in one place is exciting,” Laird admitted.
More clients seek proactive recommendations. Laird said that fosters the elusive stickiness that every institution strives for.
Open banking’s benefits
The onset of open banking is significant, as Laird sees it as helping to serve customers better in their time of need. It also forces staid institutions to be more proactive, especially with their data, because if they don’t, someone else will.
“I think it’s going to be a really good thing for consumers,” Laird said.
Why the SEC’s Upstart subpoena is important
Laird said the outcome of the SEC’s subpoena into Upstart’s models and loans will have a crucial impact on fintech’s use of AI and the role of explainability. If Upstart gets the green light, all is well. Should complications arise, many companies will be revisiting their plans.
That would be unfortunate, as many current assessment methods perpetuate disparities. As soon as a FICO score drops to 660 or lower, customers’ offers quickly get worse.
“There’s no difference between a 661 and a 659 besides being labelled near prime,” Laird said. “The typical banking industry is taking a step back and refusing to interact with that customer base. They’re not giving them loans, or if they’re giving them credit cards, they’re not the same credit cards that a prime customer is getting.
“It becomes a self-fulfilling prophecy that anyone below that threshold is more likely to default. That enabled fintechs to exist, gobbling up market share banks were unwilling to touch, but it also has bad outcomes for customers.”
