SAN FRANCISCO – Credit scoring is becoming increasingly digitalized, with systems doing more of the work, but the auto-financing world doesn’t appear ready to turn things over to artificial intelligence and machine learning.
At least not yet.
“In the next three to five years, we’ll try to figure out what we have,” Mike Kane, Ally Financial Services’ vice president-consumer credit, says of the prospects of using big data to make credit decisions about and profiles of prospective borrowers.
The goals are the same as they were when credit scoring first emerged decades ago.
Among them: “Do I do this loan? What can I do about this credit history? How do I make the right decision about pricing to risk?” Kane says during a panel discussion, “On the Horizon for Credit Scoring and Decisioning,” at the American Financial Services Assn.’s annual vehicle-financing conference. It is held here in conjunction with the National Automobile Dealers Assn. annual convention and expo.
To empower more-advanced systems means setting them up correctly in the first place so they can handle, sort through and assess huge quantities of information in the big-data world, panelists say.
The finance industry is working on “moving it out of the lab and putting it into operation,” Kane says of work on future predictive analytics that use AI and machine learning.
But currently there’s something of a disconnect. “Data scientists are coming up with solutions when we’re not sure exactly what questions to ask,” Kane says. Those include: “How do we use it? Will the information flow?”
Other questions he raises: Do I have a real person? Is there a real employer? Is the (consumer’s) income realistically stated? Is there collateral?
“We’ll get better with data knowing these things,” Kane says.
He likens today’s industry talk of AI-powered loan-decision-making systems to society catching the autonomous-driving fever. “The fear is of not using humans to enhance value.”
Adds Kevin Moss, SoFi’s chief risk officer: “If you give someone a rocket ship, they might not fly it in the right way.”
Despite initiatives to use advanced technology in loan decisions, those remain primarily a human effort today.
“We find it is better to work with a person on the lending end, especially if we have a customer with credit that’s on the fringe,” Wes Lutz, 2018 NADA chairman and a Michigan dealer, tells Wards here.
Automated systems today rely on entered formulas and essentially give yes or no answers on loan applications. The predictive analytics of AI and machine learning would provide fuller customer profiles.
“Dealers want to get customers qualified and with loan terms that are desirable to the customers,” says panelist Ethan Dornheim, FICO’s vice president-scores and predictive analytics. “The challenge is how to roll out new sources that don’t create friction.”
That’s not easy, says Paul DeSauliniers, Experian’s head of risk scoring and alternative data.
“Put yourself in the customer’s shoes,” he says, referring to the potential of AI and machine learning profiling loan seekers.
Moss says there’s already a degree of friction when customers are asked to bring in pay stubs and otherwise prove their creditworthiness.
Having digital access to a potential borrower’s checking account helps lenders in their decision-making and provides more actionable data than traditional credit scoring, Moss says. “You can look at cash flow and do a better job because many of those outflows are not reported to credit bureaus.”