A Blog by Jonathan Low


Jan 5, 2019

How Financial Firms Are Using AI To Boost Lending While Managing Credit Risk

They want to make more loans, but not assume greater risk. AI is helping assess who is a good prospect - and who is not. JL

Becky Yerak reports in the Wall Street Journal:

Artificial intelligence takes increasing amounts of data and finds relationships between variables to help determine creditworthiness. AI is able to look at more signals. If you’re building an AI model you can have hundreds or thousands of such signals, including whether people have defaulted on rent payments or cell-phone bills. The types and amounts of data used in underwriting have quadrupled in the past 20 years and “that number (will) grow. Artificial intelligence and machine learning allow us to lend more to people who were invisible. “All of that creates the ability to better understand the customer.”
A few years ago, subprime auto lender Prestige Financial Services Inc. found itself rejecting 25% more borrowers than usual as delinquency rates climbed for borrowers with bad credit.
“It got to the point where we, as an executive team said, ‘We’ve got to find a way to increase our volume,’” Steve Warnick, Prestige senior vice president for risk management, said of approval rates that had fallen from a typical 40% to 30%.
Wrestling with how to boost business while maintaining underwriting standards, Prestige last year partnered with ZestFinance Inc., an artificial-intelligence software developer founded by Google’s former chief information officer, and started drilling down into some 2,700 borrower characteristics instead of the couple of dozen that the lender typically had on its risk-assessment scorecard.
Prestige, owned by the same family that controls NBA’s Utah Jazz, is among a growing number of lenders, including Synchrony Financial and Ford Motor Credit Co., that have looked into the future and believe it could include artificial intelligence as a tool to take increasing amounts of data and find relationships between variables to help determine creditworthiness.

Some Tips on How to Incorporate AI

  • Have a clear goal to generate results that can be used in the real world.
  • Focus on a few meaningful projects so you can get your machine-learning effort out of the lab, and avoid getting sidetracked by science projects.
  • Get as much data into the model as possible to maximize return on investment.
  • Have an internal champion who can move the project along, and get risk, legal, compliance and business teams engaged in the process.
  • For companies involved in lending and credit, make sure the machine-learning models are compliant with banking regulations and can be explained to regulators.
  • Source: Douglas Merrill, CEO of ZestFinance
Instead of looking simply at whether a potential borrower has ever filed for bankruptcy, for example, the machine-learning system helped Prestige consider such factors as when the bankruptcy happened, and analyze that data with other variables, including previous car-payment records and time spent living in his or her current residence.
The timing of a bankruptcy is important because individuals are restricted from repeatedly trying to wipe out their debts. So someone fresh out of bankruptcy for the first time might be a better credit risk than someone who filed, say, six years ago, said Douglas Merrill, ZestFinance CEO and founder.
Merrill joined Google in 2003 and spent five years there as CIO and engineering vice president. He lost his hearing as a young child after an auditory-nerve infection but eventually got his hearing back.
“The power of AI is being able to look at more signals,” he said. “If you’re building an AI model you can have hundreds or thousands,” of such signals, including whether people have defaulted on rent payments or cell-phone bills. Additional criteria can include the number of payments that are 90 days past due, the number of months since the most recent delinquency, and how many accounts are current.
Synchrony Financial, the credit-card business that General Electric spun off in 2015, recently said it was deploying new artificial intelligence that would ease access to credit. The company says the types and amounts of data it uses in underwriting have quadrupled in the past 20 years  and it expects “to see that number grow in the coming year.”
“The way credit has been done for the last 30 years or so is using credit bureaus and credit-bureau data and maybe some other things,” Synchrony Chief Executive Margaret Keane said in October at Money20/20, a payments and financial technology event.
“What artificial intelligence and machine learning allow us to do is to get much broader perspective on consumers, and we’re going to be able to lend more to people who were invisible” thanks to additional data shedding light on their creditworthiness, she said.
Synchrony long has worked with many small companies, including ZestFinance last year, to test artificial intelligence. The company ultimately decided to develop in-house a new system to make better use of the significant and rising amounts of data available on consumers, including purchasing history, bank data, utility information and social media habits.
Such tools could improve the ability to make credit available instantly, helping shoppers and retailers, said Carol Juel, Synchrony chief information officer.
“All of that creates the ability to better understand the customer,” Ms. Juel said. “The key is having the technology and a sophisticated platform that allows users to use that data in underwriting” with speed.
For nearly two decades, Synchrony was the sole issuer of Walmart Inc. credit cards, but, in July, the retailer said it was switching to Capital One Financial Corp. Walmart executives had grown irritated because, among other issues, they wanted Synchrony to approve more applicants. Walmart introduced Synchrony to ZestFinance.
Ford Credit said last year it teamed up with ZestFinance to study alternative data’s ability to help forecast defaults, and found that machine learning improved risk prediction and had the potential to broaden approvals, including to those with low credit scores or thin credit files.
Ford Credit spokeswoman Meredith Libbey says her company continues “working on models with machine learning.”
Backers of ZestFinance, which was founded in 2009 and has about 100 employees, have included Lightspeed Venture Partners, Upfront Ventures and PayPal co-founder Peter Thiel.
Prestige Financial is part of the Larry H. Miller Group of Cos., which also owns car dealerships and controls the Jazz.
In September Prestige said its lending volume doubled, to $55 million in August from a low of $25 million during the tightest credit conditions. The new loans perform as well as those made under the old system, said Mr. Warnick.
Mr. Warnick said the deployment was uneventful because potential problems were anticipated. He said Prestige shifted immediately into the  new system, and said companies that want to run a new lending model in tandem with an existing process might have challenges that his company didn’t experience.
While attributing some growth to expanding into new dealerships, he also credits the use of artificial intelligence. Approval rates peaked at 40.9% in June. The additional qualities that Prestige weighs through artificial intelligence “allow us to be better at predicting the result of the loan,” Mr. Warnick said.


Post a Comment