Subscribe / Unsubscribe Enewsletters | Login | Register

Pencil Banner

Toyota uses analytics to keep delinquent customers in their cars

Thor Olavsrud | March 29, 2016
Toyota Financial Services' Collection Treatment Optimization (CTO) program helps its collections agents optimize which borrowers to call to help reduce delinquencies and keep customers in their vehicles.

"It's really an innovative area," VanTassel says. "Before the two teams spoke, there hadn't really been a concept that there was something that could be optimized there. There was no name for this before."

Collections meet statistical modeling

"We had not used mathematical optimization for collections," Bander adds. "But we knew that we had to change if we were going to deal effectively and fairly with over 100,000 people per day."

Using FICO Xpress Optimization Suite and FICO Model Builder, TFS combined statistical modeling, forecasting, predictive modeling and optimization into a single framework that allows it to rapidly simulate multiple scenarios and then deploy an optimal strategy into production. The implementation divides customers into micro-segments based on risk, ensuring that the collection treatments are delivered individually, one customer at a time.

"We're a very process-centric company," Bander says. "Modifying the processes was the first step. We did that long before we got the math done. We bet the math would be solved."

The result was a decision engine that helps collectors identify the right customers to target. TFS developed a Delinquency Movement Metric, borrowed from the credit card space, to help collectors see how effective their efforts were. The same metric is also used by the risk department to minimize losses.

"It relates the success of the collector partnering with the customer to the success of the company," Bander says. "It's a win-win. You maximize performance and make everybody happy."

At first, Bander says, the rules of the decision engine weren't very sophisticated. But they've grown over time. TFS now processes 4.2 million accounts nightly with over 400 attributes.

"We're storing a lot of data," he says. "Internally we look at over 70 attributes based on data that we buy from various vendors. We're crunching through 3.25 million rows a night to make decisions."

The process uses a method called champion-challenger testing in which a percentage of collectors use a decision engine based on the "champion" process — the process that has achieved the best results to date. Another percentage uses the 'challenger' process, a theoretically improved process. Once a quarter, TFS validates the results. If the challenger proves more efficient, it becomes the champion.

"We're currently doing a quarterly process," Bander says. "It's more time consuming to validate the process that comes out of the model and get it implemented on the floor than it is to run the process. My goal is to do it monthly."

Keeping delinquent customers driving

In its first year, the CTO program helped more than 6,000 customers stay in their cars and 50,000 customers avoid reaching a stage of delinquency that would affect their credit.

 

Previous Page  1  2  3  Next Page 

Sign up for CIO Asia eNewsletters.