AI paves way for faster and better loan decisions

Lenders who adopt artificial intelligence could provide auto dealerships and their customers with immediate, automated decisions on a large proportion of loan applications.

The software also might demonstrate better judgment than human bankers and traditional models.

That has been the experience of a Southeastern U.S. credit union. After incorporating Zest AI's technology, it went from making instant decisions on 10 percent of auto loans to 75 percent of them.

"We're auto-decisioning so much more," Todd Peeples, All In Credit Union sales and lending senior vice president, said during an American Financial Services Association webinar in September. He said his credit union plans to expand this automation across all of its loan categories.

The All In-Zest partnership is an example of AI and auto-decisioning trends in financial services that could have positive effects for dealerships and their customers. Other lenders and tech companies are competing in the space as well. For example, AI provider Scienaptic has announced a slew of partnerships with credit unions to help automate various loan decisions, including auto loans. It says its tech can increase auto loan approvals by 20 to 40 percent and reach 60 to 80 percent faster decisions.

"For every CU we work with, auto loans make up over half of their portfolio, and in turn, we see many clients coming to us for AI/automation to specifically grow their auto books," Scienaptic partnerships Vice President Kathryn Bonesteel wrote in an email.

Bonesteel also said that Scienaptic's more than 60 clients include subprime auto dealers. She described that adoption as "usually more about having confidence in their decisions and their risk."

CU Direct Chief Technology Officer Darren Lorentz predicted more advanced AI credit models would grow. But he also pointed out a credit union could probably get to a 55 to 60 percent instant decision rate just by automating no-brainer denial and approval situations.

"That is a giant leap from where they are today," said Lorentz, who said credit unions often continue to manually approve and deny those easy calls.

This would free up underwriter time to examine the remaining loans, the ones that need "a little more attention," Lorentz said. Tack on AI to the rest of this process, and "that can really supercharge their lending program," he said.

Lorentz said some credit unions are 100 percent auto-decisioning, though that might not be the right move for every company in the segment. Peeples said All In opted to leave some less clear-cut decisions to staff, but "we could be auto-decisioning all of them."

Bonesteel was confident that 100 percent auto-decisioning was possible.

"It is just a matter of time before all lending is automated," she wrote. "We can auto decision up to 100 percent of loans while ensuring that quality of underwriting is significantly smarter than traditional decisioning systems."

Peeples said customer expectations had changed. Lenders used to render loan decisions in a couple of days. That shrank to a matter of hours, and now, borrowers want an answer "instantaneously."

For small and midsize lenders, "it's almost the place you have to go in order to stay competitive," he said.

"We are living in a world of instant gratification and there is little to no loyalty unless lenders can get decisions instantly," Bonesteel wrote. "This is where lending automation becomes so important."

In addition to gaining speed, Peeples said his credit union sought to improve consistency and availability by switching to auto-decisioning. Automating the process means a borrower applying online on a weekend doesn't have to wait for an answer, he said; it's "24/7 availability." It also helps the 27-branch credit union avoid situations where a borrower declined at one location gets approved at another.

Zest's auto-decisioning AI builds a model to predict future loans based off older ones. Zest AI CEO Mike de Vere told Automotive News his company's technology changes credit risk analysis from evaluating a couple dozen data points to hundreds. The hypotheses drawn by the computer can be tested by its ability to predict past loans.

Zest AI product head Nidhi Panday said at the webinar her company also provides software to monitor the loan portfolio after the fact, allowing a lender to verify that the model hasn't led the company astray.

Artificial intelligence has helped All In approve loans it would have incorrectly denied and deny loans it would have OK'd that later proved to be bad loans, Peeples said. He called it a rare example of increased returns with less risk. "Which doesn't happen very often," he said.

Panday said the largest benefit arises in "middle tiers," noting that it's easy for a lender to approve the highest tier of creditworthiness and deny the riskiest tier.

She said Zest's customers find a 70 percent increase in accuracy over traditional scoring in that middle bracket. She said this can be done with just credit bureau data, which is the case for most of Zest's models.

De Vere said national credit scores alone accurately predict behavior for superprime and prime borrowers. But AI indicates "most national credit scores are barely better than a coin toss," he said.

On average, Zest models produce a 15 percent increase in loan approvals while holding risk constant, according to de Vere. It also reduces charge-offs by 30 percent, he said.

Peeples said the credit union installed Zest AI's model in April. Early delinquency data found borrowers approved by the software to have been better bets than those approved by staff.

"So far, the performance has been very good," he said.

Peeples said the credit union also had to revise some of the variables it also checked after reaching a decision on creditworthiness. Something such as loan-to-value ratio remained a consideration even if the AI viewed the applicant as an acceptable risk, according to Peeples. But other factors such as debt-to-income ratio needed to be discarded, for the software already had taken this into account when scrutinizing the borrower, he said.

Machine learning based upon prior real-world human behavior can pose a "garbage in, garbage out" problem where AI inadvertently adopts the same biases and mistakes of the humans whose decisions train the model. For example, a bank whose lenders have consciously or unconsciously denied minority borrowers loans at a higher rate than white applicants with identical credit.

Panday said Zest avoids this issue through a technique it calls "adversarial de-biasing."

She likened it to two fighters in a ring. One looks for accuracy in underwriting decisions and has no idea of the applicant's ethnicity. The other is aware of it and watches for the chance the AI is being misled by past trends. For example, the second model might warn, "I see a high correlation between denials and people of color. Take this data back, modify your algorithm, and do better," Panday said.

This is done multiple times until a fair and accurate model arises, Panday said. She said one credit union has found a 26 percent increase in loan approvals for women and a 25 percent increase in approvals for races other than white.

She said Zest's model is fully compliant and explainable to regulators, and the Consumer Financial Protection Bureau and Office of the Comptroller of the Currency have both expressed praise for its documentation.

Bonesteel said Scienaptic's AI "has been carefully calibrated and tested to ensure that human biases are completely eliminated, and it adheres to fair lending norms in letter and spirit."

Other financial companies haven't made the leap on auto-decisioning. De Vere said most of the lenders his company approaches are in the "low single digits" in terms of immediate answers.

"We're nowhere near that 75 percent number," webinar moderator Chris Otey, chairman of South Bay Credit Union, said in reference to All In's proportion.

Bruce Upbin, Zest AI policy and communications vice president, wrote in an email that lenders with single-digit percentages of loans automated "are typically happy to get up to 30-35% auto-decisioned. And those who are already auto-decisioning 50-60% are happy to boost their levels more incrementally to 65%-75%."

Upbin estimated that indirect auto lenders tend to have higher auto-decisioning rates because of the competition and time pressure.

A 2020 Cornerstone Advisors survey of 69 financial institutions found even the majority of the lenders with more than $1 billion in assets auto-decision less than 40 percent of all loans. Among sub-$1 billion lenders, 37 percent didn't auto-decision anything.

But even lenders who do auto-decision arguably aren't truly doing it. Cornerstone found 30 percent of lenders review 75 to 100 percent of automated decisions.

According to Peeples, it can be a moment of faith. He said All In has been approving loans it wouldn't have previously considered. In some cases, staff will wonder, "You sure about this?"

He said All In's response is, "Use the model and go."