- What happened: Credit Union of Colorado used Scienaptic AI software to approve an additional $40 million in consumer loans, which would have been declined using traditional methods.
- How it works: The AI software uses alternative data sources like rent payments and bank account data, in addition to traditional credit bureau and FICO scores, to provide a more comprehensive view of an applicant's financial behavior. This helps approve applicants with limited or no credit history who would typically be denied.
- The benefits: Beyond loan decisions, the AI system also helps with setting appropriate credit limits, identifying early warning signs of potential defaults, and streamlining the lending process by making quick automated decisions for a significant portion of applications, allowing staff to focus on more complex cases and maintain scalability.
Credit Union of Colorado says AI-based credit decision software has helped it approve $40 million in consumer loans it would have declined using traditional methods, while at the same time eliminating charge-offs and cutting overall consumer loan losses nearly in half.
The credit union started using Scienaptic AI software in October 2022 to underwrite auto loans, credit cards, personal loans and lines of credit, according to Chad Wilcox, senior vice president of lending at Credit Union of Colorado.
Prior to that, the credit union used traditional loan software and time-honored criteria to make these loans: debt-to-income ratios, FICO score, payment-to-income ratio and loan-to-value ratio for vehicle loans.
"If it fell within all these certain parameters, you could get an automated approval," Wilcox told American Banker. "The primary driver was the information on the credit bureau report."
Scienaptic's model ingests alternative data such as rent payment data from LexisNexis, as well as bank account data from Plaid and the credit union's own core system and credit bureau data and FICO scores to generate a detailed, in-depth view of a loan applicant's ability and willingness to pay.
"The broad principle is to create a singular view of a member in a coherent and concise way, using Fair Credit Reporting Act-compliant data sources," Vinny Bhaskar, chief operating officer at Scienaptic, told American Banker. About 150 financial institutions of all sizes use the software, he said.
Scienaptic plugs the credit union's underwriting criteria into its model to analyze each loan application that way, and provides its own risk score on each loan.
"Ultimately, the goal is to say yes to more people in an automated, real time fashion," Wilcox said.
The $40 million in new loans comes from approving people who would normally be denied through the old system, Wilcox said.
"The biggest piece we knew we were missing was with individuals that had limited credit history or no credit because they're new to the market," Wilcox said. "Traditionally, lenders that are not in the subprime area typically would say no to those because there's not enough credit report information to say yes. So we knew we were missing an opportunity to say yes to more individuals, to help some of the credit challenged or uncredited individuals establish credit."
Credit Union of Colorado's adoption of AI-based lending is part of a growing trend of lenders turning to alternative credit metrics and AI when making lending decisions.
"An increasing number of banks are adopting artificial intelligence and machine learning in retail lending to expand the scorable population, lend to underserved customer segments, increase automated underwriting rates and increase loan approval rates while effectively managing credit risk," Craig Focardi, principal banking analyst at Celent, told American Banker.
A just-released survey he conducted of more than 100 U.S. consumer lenders found that 57% use AI in consumer loan origination today, 34% have such a system in development and 8% are evaluating one.
Where there are holdouts, this is because of the need for better data management, modeling expertise, lending system modifications and new policy guidelines, he said.
"Some banks lack the agility to update or change their loan origination processes to incorporate machine learning," Focardi said.
In addition to helping with loan decisions, the Scienaptic software also helps Credit Union of Colorado calculate appropriate credit limits, Wilcox said, so it doesn't offer too much money to a borrower who can't handle it.
The ability to use alternative data is helping the credit union make better decisions, he said.
For instance, the ability to see how well people have been paying cell phone bills and their rent may not show up on a traditional credit bureau report.
"We're now using better logic and better data to say yes and also ensure that those loans don't default at a later date," Wilcox said. "And then, as the relationship has morphed, now we're able to do bundled offers. So if you're applying for a credit card, it's automatically looking to see, hey, can you be pre qualified for an auto loan, or do you have an auto loan on your credit bureau that we can refinance, and we know we can save you money, we can present that offer to the borrower real time."
The Scienaptic system also monitors loan performance and cash flows for month-to-month changes and detects early warning signals of potential defaults, losses or hardships.
For instance, if a member has always made loan payments on the 13th of the month and suddenly starts paying on the 18th, 19th or the 20th, that could be a red flag. Or if the person's direct deposits stop, it's possible they're no longer putting their paycheck in the institution because they're planning to walk away from that debt.
"These trigger points don't always mean something's going wrong, but they allow us to do an outreach, because a lot of times, no one wants to call you when they're having financial struggles," Wilcox said. "But if we can do it in a way that's genuine and helpful, and to say, hey, it looks like something might be going on, talk to us, we're here to help."
The credit union, depending on the circumstances, could offer some relief by letting the borrower skip a payment or restructure their loan. "There's a lot of different things that we can do, and part of that really is just being able to have the conversation when we don't know what's going on," Wilcox said.
Humans could do this kind of evaluation, but it would take 15 to 20 minutes per account, whereas Scienaptic AI can do it in seconds, he said.
About 60% of consumer loan applications are automatically approved or declined by the Scienaptic software, Wilcox said. Generally these are on the "bookends" of the credit spectrum – people with a high credit score and a low Scienaptic risk score, or with a low credit score and a high Scienaptic risk score. Such applicants can always question the decision or ask for a second look, he said.
For potential borrowers in the middle of these extremes, humans look at the risk and make a decision. All decisions, human and automated, are reviewed on a monthly basis. A fair lending module does a quarterly review to make sure there's no inadvertent disparate treatment of protected groups.
Part of the reason the new system has increased approvals while lowering losses, Wilcox said, is that Credit Union of Colorado is using the Scienaptic risk score as an enhancement to traditional underwriting rather than as a replacement.
"It's an additional data point and data source to get us better, but it also allows us to make changes," Wilcox said. "If we see changes in performance or degradation, we can massage the tool and the rules. As we're feeding in our losses into their or delinquencies into the model, it's adjusting."
No one's been laid off as a result of using the new software, Wilcox said.
"We can scale and grow without adding headcount or requiring a bunch of overtime and extra work on the existing employees," he said.
And it helps make quick decisions.
"In today's world, if you ask for something, you want an immediate answer, you don't want to wait 30 minutes or six hours for somebody to tell you 'Yes' or 'No,'" Wilcox said.
