Robots are increasingly replacing people as decision makers in lending, a trend that will transform North Texas finance for better or worse.
The impetus comes partly from Africa, where startup technology needs only 10 seconds to determine whether to loan as little as $2.50 to the working poor, with interest rates as low as 6 percent. To establish who is a good risk, lenders’ technology culls clues of creditworthiness from an applicant’s smartphone by running private information like texts and social media posts through high-powered analytics. Candidates give permission up front for this so-called data scraping. Closer to home, San Francisco-based Social Finance offers consumers two-minute loan applications that consider factors such as education and work history.
Technology advances from emerging players like Social Finance—SoFi for short—are helping spur lenders of all stripes to take a closer look at artificial intelligence. “We are reviewing and investing in opportunities to deploy the concept, but [we’re] in the early stages,” says Norm Bagwell, chairman and chief executive of Dallas-based Bank of Texas. “The more standard a transaction is, the easier it would be to use AI, in my view. This suggests higher volume, smaller transactions. Reviewing standard documents is an initial application.”
AI is a group of hardware- and software-based technologies that use cutting-edge procedures to solve problems.
All of the largest banks are using AI, as are many a tier below, according to Christine Mcqueen, senior vice president and director of bank operations at BOK Financial, the Oklahoma-based parent of Bank of Texas. BOK, where Bagwell is a regional banking executive, had about $5.3 billion in assets at mid-year in the Lone Star State. “AI is being leveraged in components of the mortgage space and other consumer-type lending functions,” Mcqueen says. “Some companies appear to be using AI in small business lending.”
Perhaps due to AI’s regulatory and legal uncertainties, many community banks in Dallas-Fort Worth haven’t yet taken the plunge.
“We currently don’t use AI for any of our loan decisions. It’s all human brain power,” says a spokesperson for Dallas-based Veritex Holdings, the $2.5 billion assets parent of Veritex Community Bank.
As more powerful computers crunch more numbers simultaneously, technology increasingly can understand humans’ language, make expert-level decisions, and learn without explicit programming to teach it. Because AI needs information to work best, the biggest banks are using it on their vast data hoards to build elaborate systems for deciding who to loan to and at what price, according to Scott MacDonald, the president and CEO of the Southwestern Graduate School of Banking at Southern Methodist University’s Cox School of Business.
As credit scores only measure a borrower’s willingness to and history of re-paying debt, the largest reporting bureaus are racing to develop models that assess earning potential, especially for businesses, MacDonald says. “With small businesses, there’s more judgment involved, as lenders often must estimate revenue,” he says. “AI is not able to replicate that now because there is not enough data. So loan officers may use ‘soft information,’ such as visiting an applicant’s store.” But even there, change is on the horizon.
More credit for small businesses
Small businesses are poised to get more credit at better prices thanks to recent AI advances that make loan decisions based both on information in traditional databases, and “unstructured” data like online customer reviews.
“Tech startups are in for a rude awakening when they figure out how heavily regulated the … industry is.”Jimmy Sawyers
“The cost efficiencies of AI in lending make it particularly attractive for small-dollar lending to consumers and small businesses,” says Lee Wetherington, director of strategic insight at Jack Henry & Associates, a Missouri-based financial technology firm. Those customer segments have traditionally been difficult for lenders to serve, he says. Though there aren’t national studies yet on how AI performs in lending, “there are anecdotal reports of these technologies cutting in half the costs of originating loans to small businesses.”
AI’s true value, though, is in allowing lenders to tweak credit they’ve extended to businesses based on real-time analysis of those companies’ financial states, Wetherington adds.
“This also enables lenders to target mid-term offers for loan extensions and renewals,” he says. But for small and mid-sized lenders to take advantage, they must first make existing systems talk to the AI and play nice together. Beyond that, lenders such as banks and credit unions must ensure any AI they use does not cause problems with government regulators, who haven’t blessed many advanced tools. “Many of these tech startups are in for a rude awakening when they figure out how heavily regulated the banking industry is,” says Jimmy Sawyers, chairman and co-founder of Tennessee-based Sawyers & Jacobs, a boutique technology and cyber-security consultancy.
A potential landmine is federal laws and regulations, starting with 1968’s Truth in Lending Act, which requires lenders to clearly disclose the terms of a loan. By clicking the “I agree” button on some loan applications, consumers give online lenders permission to read and analyze private information on their smartphones, Sawyers says. “There will be borrowers who are surprised at how much access they’ve given to lenders. Then, slowly, legislation will emerge to protect the consumer,” he says.
Beware of the Black Box
Another issue facing U.S. lenders in using AI is regulations relating to a 1974 law, the Equal Credit Opportunity Act, which prohibits loan discrimination based on demographics, according to Jerry Sanchez, Dallas-based senior counsel at Dykema Cox Smith.
In addition to barring explicit discrimination, the law also bars “disparate impact” in lending policies or practices, he says. Disparate impact is when ostensibly neutral policies have a negative impact on a group that the equal credit laws protect. Allowing AI to determine on its own who gets loans and what price they pay could unintentionally exclude some consumers from mainstream finance.
Another issue can be the technology’s ability to change its own rules and procedures based on what works, experts say. Humans can’t totally control AI in the way they can keep a handle on today’s rules-based systems for making decisions. Regulators would give more scrutiny if that results in AI, say, turning down loans to women, experts say.
An even thornier question is humans’ inability to understand decisions that AI makes. Existing tools can already incorporate tens of thousands of variables into approval determinations, a so-called “black box” whose reasoning for determinations is beyond what lenders can grasp.
This could become more intense in the next decade, when ultra-powerful computers come online that harness peculiar laws of “quantum mechanics,” which the Wall Street Journal calls the “physics of atoms and particles.” “Google claims to have a quantum computer that is 100 million times faster than today’s systems,” Sawyers says. “Borrowers who are impressed with decisions in hours will soon expect them in seconds.” But federal and state watchdogs won’t cotton to loan decisions that lenders can’t explain.
“Regulators require banks provide a clear reason for declining a loan or pricing it higher than others,” Sanchez says. “They will not permit lenders to use a ‘black box’ to make credit decisions in the U.S.” Tools have emerged from the likes of California’s ZestFinance to essentially reverse-engineer AI’s loan decisions so humans can get their arms around them. But regulators haven’t blessed them yet, Sanchez adds.
Banks have recently been shrinking headcounts in India as automation takes over jobs humans historically performed. “Loan officers will be expected to deliver even more business,” says BOK’s Mcqueen. “Those with the best-selling skills should be able to even further differentiate themselves and ultimately deliver more revenue.”
But, she adds, banks will be wise to keep humans involved in determining where to extend credit. “I cannot envision a healthy environment where all lending decisions are made by software,” she says. “Commercial lending has a component of art that will be impossible to replicate with technology, no matter how sophisticated.”
Jeff Bounds is a freelance business writer in Garland.