Ford Credit to Use Machine Learning in Approval Process

Ford Motor Credit Co. has decided to change its approval underwriting process to incorporate machine learning to look beyond credit scores, the company said today.

The change stems from a study conducted by fintech startup ZestFinance on Ford Credit customers that measured the effectiveness of machine learning to better predict risk. But Ford maintains that any revisions to its credit policies will not increase its risk appetite, as it more accurately predicts risk against a broader group.

Though Ford Credit currently has a charge-off rate below the industry average, loan losses did rise 30% year over year in the second quarter to $82 million.

The platform applies algorithms to alternative data using a process called machine learning, which is a type of artificial intelligence that allows the technology to learn on its own without manual input, Douglas Merrill, founder and chief executive of ZestFinance, told Auto Finance News in March. And because the platform is capable of “learning” over time, it can — for example — propose changes to variables as patterns evolve or emerge, or recognize and incorporate macroeconomic changes into its assessments, Ford said in a statement.

The captive previously has been using a logistic regression tool, but will now be implementing machine learning for all applicants, Jim Moynes, vice president of risk management at Ford Credit, told AFN, adding that while the company has decided to make the change it is still figuring out how. “We know for sure it’s going to take us more than two years to do in any meaningful way,” he said.

Factors such as whether applicants supplied the same cell phone number on previous loan applications and whether they have occupational licenses are additional data points that machine learning can assess in loan applications, Moynes said, adding that FordCredit has always had access to this data has not used it with logistic regression tools. Additionally, the extra factors will not supplant credit scores or dramatically affect those who have prime credit, but will instead provide a better risk profile for thin-file borrowers, he said.

Part of the motivation for employing a more holistic approval process for thin-file borrowers is to court young customers who are new to financing. According to the Consumer Financial Protection Bureau, one in 10 American adults have no credit record, making them difficult and often impossible to underwrite using traditional methods, Ford Credit said in a statement. “This includes millions of millennials who are also part of the fastest-growing segment of new car buyers.”

Ford wants to engage in the consumer’s first vehicle purchase in order to build brand loyalty, and Zest’s platform aims to make lending to those millennials less risky, Zest’s Merrill previously said. “[The captive’s] view is that if your first car is their brand, your last car will be their brand,” he said. “So they are pretty focused on ‘How do I make sure the first interaction is a good one?’”

ZestFinance is confident its algorithmic platform could reduce losses and allow lenders to more safely lend into the Fico score range of the low 600s, he added. The platform also allows lenders to better monitor disparate impact and change the model to quickly limit unintentional discrimination of protected classes.