SAN FRANCISCO – Automated decision-making – which futurists see as a great leap forward – has taken a big step backwards in the vehicle-financing world.
Some lenders are deactivating computer systems that approve auto loans automatically, using predicative algorithms to analyze creditworthiness.
The reason for pulling the plug? Major losses by lenders that had relied on the systems. Turns out, too many of them approved loans to deadbeats and schemers.
“The biggest cost of automation is the losses,” Preston Davenport, executive vice president of BB&T Corp., says at a Consumer Bankers Assn. auto finance conference here. “We don’t use it a lot.”
In a new book “Super Crunchers,” Yale University professor Ian Ayres says automated decision-making is turning bank-loan officers into mere functionaries.
Not so fast, says Oscar Joaquim, senior vice president and credit manager for Citizens Auto Finance based in Warwick, RI.
“We’ve cut back a lot on automation in certain states,” he says. “We’re putting it in human hands more than before. That’s got to be done in today’s market.”
Lenders still rely on automated decision-making to reject loan requests. But many loan approval decisions – 99% for Joaquim’s firm – are back in manual mode.
“It means more work for lenders,” he says, with lenders still striving for “a quick turnaround” in handling loan applications sent by car dealerships on behalf of their customers.
“We will turn auto decision-making back on when we figure out some of the glitches,” Joaquim says.
“You still need the human element,” says Jeff Hodge, operating partner ofWorld, a dealership in Downey, CA. “Sometimes I’ve seen (automated decisions) that don’t make sense. And some of those were in our favor.”
The algorithms can fail to account for complexities in changing situations, he says, referring to many consumers suffering from recent economic hardships.
For instance, some borrowers saw their credit scores drop in connection with the nation’s mortgage debacle. Those people now can be met with resistance in getting auto financing, even though they’ve maintained good credit histories in paying off car loans.
The automated systems fail to recognize such differences, Hodge says. “With times getting tough, you need a human touch.”
Sometimes, it is bad information in, bad information out. Automated systems can’t detect if would-be car buyers are lying or fudging on credit applications.
“One of the biggest challenges is the accuracy of information found on loan applications in recent months,” Joaquim says. “We try to work with dealers on that. Granted, it’s the customer who puts down the information. But it’s done at the dealership.”
He cites applicants overstating income and participating in “straw deals” in which one person’s good credit is used to get a vehicle loan for someone else.
Efforts to validate loan-application information are why lenders are backing away from automated-decision systems, he says.
“Automation has two sides,” Davenport says. On one hand, it’s fast. On the other, it’s not good at detecting shady human behavior.
“A customer gets turned down a couple of times and then decides he better put down a different income,” he says. “And if dealers are stretching and trying to get deals through, you are going to see more fraud.”
Another culprit: outside firms that some dealerships hire to come in and move aging inventory during ballyhooed “special sales events.”
“Those firms come in and leave, and the dealership is left to clean up the mess,” Joaquim says. “Two years ago, you didn’t see so many of them. Now, more dealers are relying on them.”
Many dealership finance managers warn lenders to expect a rash of bad-loan applications when such outside operators are in town.
“Dealership finance people will call and say, ‘You’re going to get bad deals all weekend – don’t buy them,’” a CBA conference attendee says.