DETROIT — In an effort to make more accurate underwriting decisions, Ford Motor Credit Co. is ramping up its credit risk model to include additional data that gives the lender a better view of a car buyer’s likelihood to make payments, the company said today.
Ford Credit and ZestFinance, a credit-decisioning technology platform, conducted a study this year in which each took the same sample of Ford Credit accounts from several years ago. ZestFinance used its methodologies and machine learnings to create risk models and place each customer on the credit spectrum — labeling them superprime, prime, nonprime or subprime — while Ford Credit used its traditional model to assess customers. The study then compared those credit ratings to the borrowers’ actual payment history. Ford Credit declined to give the number of accounts studied, but said it was a “statistically reliable sample size.”
The study found that ZestFinance’s approach was more accurate in predicting the consumers’ actual payment performance than Ford Credit’s traditional model.
“That’s important for all people in the finance space — knowing at that particular point in time where that consumer falls on the risk spectrum and whether it falls within the risk appetite or not,” Jim Moynes, vice president of risk management for Ford Credit, told Automotive News.
Two years away
The plan for implementing additional data and machine learning is still under development and will not take effect for at least two years, Moynes said. Ford Credit may develop a tool internally or rely on a third party, he said.
None of these changes toward machine learning affect Ford Credit’s risk appetite, he said, but “this modeling helps us get a better insight as to where that person falls at that particular time.”
The new model will enhance, but not replace Ford Credit’s existing credit-risk model, Moynes said. The most important factor will still be assessing loan and lease applicants’ payment records, he said.
Today, most lenders use a summary from a credit bureau to analyze consumers. With the enhanced model, Ford Credit will look at the individual elements within a consumer’s credit bureau profile that are not analyzed today. For example, Moynes said, consumers must include a phone number on their credit applications. Through the study with ZestFinance, Ford Credit learned that “if the person uses the same cell phone number over and over and over, that helps indicate a level of stability. Stability is usually a very positive indicator for someone to continue to pay on any obligation they have.”
If the consumer has a different phone number every time they apply for credit, that could indicate a lower level of stability, he said.
“We are going from a modeling technique which doesn’t look at as many variables as machine learning does,” he said. Machine learning “allows us to look at the existing data we have in a more granular way. These small insights help better place somebody on that spectrum of credit and then helps us assess that risk better.”
Machine learning tools are capable of adjusting over time as variables or patterns evolve, Ford Credit and ZestFinance’s statement said.
A more encompassing model may increase Ford Credit’s loan and lease portfolio, Moynes said, but not necessarily.
“We think ultimately understanding that risk better will help us reduce credit losses,” he said. “We believe that these deeper understandings may help some consumers that we may not have been able to purchase under our existing risk appetite. These new insights may help us realize that this person does really fall into the appetite of risk that we’ve always been comfortable with.”
Additional data allows the lender to assess thin-credit-file customers, but it also gives more insight into customers Ford Credit may have historically approved. Applicants can move up or down the credit spectrum with additional data, Moynes said.
“Theoretically, someone could move down as a result of these deeper insights,” he said.
This is an evolutionary change, not a revolutionary change, Moynes added. “This is not a change to our risk tolerance, our risk appetite, our approach to risk. We feel that it’s incredibly healthy as a risk organization [to] really understand at a higher level of detail where the consumer falls.”