LAS VEGAS—To get some understanding of just what artificial intelligence can do for loan underwriting and expanding lending into underserved markets in the future, it can help to look back—in this case, running an AI model with CU data from prior to the Great Recession of 15 years ago, according to one person.
Mike de Vere, CEO of ZestAI, said his company did just that, obtaining data on the now old-fashioned tapes from one credit union, that it then imported into its model. The details are below.
While many are wary of AI, de Vere observed several times that it is “not the bogey man” and, in fact, holds great promise.
Prior to offering his insights to an audience at the California and Nevada CU League’s REACH Conference, de Vere offered some quick definitions:
- AI is an umbrella term he said refers to “getting a machine to do something a human can do.”
- Machine learning is part of AI and refers to a machine learning from past experiences.
- Deep learning, he said, involves the kinds of breakthroughs that are now are driving the AI boom and represent the computation of multi-layer neural networks.
Slow to Adapt
Financial services has been slow to adopt AI, he stated, and a poll of his audience found about half saying they had AI initiatives under way, even as he showed a chart showing nearly every organization say they are exploring AI.
Noting AI is not easy to deploy, the largest credit union deploying its automated underwriting using AI in the U.S. is the $20.5-billion Golden 1 Credit Union in California, according to de Vere, while the smallest CU using AI is the $144.8-million Valley Isle CU in Hawaii.
“The mission for us is to make it more accessible for the long tail in credit unions,” deVere said.
What Members Need
According to de Vere, the majority of CU members, and by default, all Americans, need two things:
- Access to affordable credit
- Flexibility
“If I talk to a hundred credit unions, as I have, they are saying they have tightened credit to only serve the A tier,” said de Vere. “We need to do better than that. We can do better than that with the right tool.”
What One Survey Found
de Vere pointed to a Harris Poll from August of this year that it commissioned and which found 80% indicated they were fearful of a recession in the future.
“They need your help right now,” said de Vere. “A lot of people are just scared. Interest rates are up. House prices are up. Gas prices are up. Can we get a break?”
de Vere said what members are looking for, “It comes down to flexibility,” said de Vere. “Why does it matter? Our members are looking for us to help. Eighty-two percent of members said they believe credit unions care for families.”
One “surprising” finding in the survey, according to de Vere, is that 2,000 primary banking customers, when they asked who’s better equipped to help you in difficult financial times, 86% said credit unions.
“Even they understand you are better equipped to help members,” according to de Vere. “With technology, you can do that.”
Looking Back—And Forward
Using what de Vere called the “ZestAI Time Machine,” he said the company accessed physical tapes of all credit data from 2006 from one credit union, and then used its AI underwriting model to review that data and explore how different decisions might have been made.
“The AI model could more accurately predict delinquencies and defaults,” said de Vere. “If you are using an industry (credit) score, you are flipping a coin in the air. It’s rational to shrink your credit box rather than flip a coin on B, C, D, credit paper.”
But it’s not best for members or credit unions, he said.
He said the review of the 2006 data show AI would have reduced credit card delinquencies by $144 billion, would have saved the economy $50 billion, and would have saved $14 billion in charge-offs.
How does looking to the past help now? According to de Vere, by making loans smarter, more inclusive, and efficient.
Get Smart
Getting smart, said de Vere, means making better credit decisions.
“With more data points and better math, you can lend confidently and provide more access,” he said. “Instead of boiling a member down to a simple picture, you have a more complete view of the member.”
Alternative data can help and it does, he said, but a credit union has to be thoughtful about when it uses it and when it doesn’t.
“Let’s say you have a felony on your record. It might be super-predictive of loan performance, but it’s super discriminatory,” he said.
The credit data bought from the credit reporting agencies “has the most signal,” de Vere stated, but the next biggest signal comes from core banking data. He called the convergence of CRA data and core banking data the “holy grail.”
He added that the AI-based modeling is 400% more accurate in predicting credit in middle credit tiers. “You can increase approvals by 25% by not adding risk,” he said.
Get Inclusive
de Vere noted he didn’t have to tell credit unions that data has inherent bias.
“The average score for a male and female is only one point off. Yet the average female has 15% less debt,” he stated. “When you unpack that, it’s due to the income disparity. The debt to income ratio is off, but it doesn’t make them any less creditworthy.”
The Pillars
In the case of his own company, he said its foundation includes “three pillars”:
- What percent of your data scientists are diverse? “It’s kind of important that the entire room isn’t white males in evaluating an AI model. Sixty-eight percent of our data scientists and engineers are diverse.”
- Recognizing there are big differences between various markets and the data must be representative of those markets.
- Being purposeful about how to build the model. “You have to evaluate the signals in the model to see if any are a proxy for gender and race, and if they are you have to tune them down and de-bias a model so that all members get a fair shot. With models optimized for both fairness and accuracy, you can ensure you are being consistent and equitable, and help the underserved.”
Get Efficient
de Vere said with instant decisioning, due to accurate scores and optimized credit policy, a credit union can do more with less and “delight members.”
A CU can also deliver responses in less than a second, allowing them to “compete with the SoFis that have gobbled up 50% of the consumer loan market.”
de Vere added, “If we’re careful about how we build the model, we can make a difference in society.”
Just Because It’s Autumn Doesn’t Mean Falling for Schemes to Overcharge You for CU Industry News. Instead, Try This Crisp Offer
The biggest, best and freshest news reporting in credit unions remains free! Each morning CUToday.info delivers its daily Fresh Today news update offering the latest headlines and breaking news right to your email, with the easy-to-read headlines format allowing you to click on the stories that interest you most in order to learn more. So stop paying those bank-fee-like subscription prices from other so-called “news” publications!
If you haven’t yet signed up for the new email solution on which CUToday.info has partnered with ResponseGenius, you can do so here. Signing up requires less than one minute of your time—and it’s free!
Please note that after signing up you may need to go to your Spam/Junk folder and mark the morning headlines email as safe. CUToday.info does not provide its list of readers and emails to outside parties, and we will not be contacting you to sell you an extended warranty or sending you any links so you may cash in on an inheritance you didn’t know was coming.
And did we mention it’s free?