Cracking Loan Fraud’s ‘Blind Eye’ Problem

PointPredictive tracks loan fraud

Pretty much every consumer knows the pain of filling out applications for automotive loans, mortgages and other types of borrowing, of having to produce pay stubs or W-2s to prove that your income is high enough for the particular purchase. Moreover, pretty much every lender and financing provider knows the frustration of checking to make sure that information is correct — while hoping that a fraudster didn’t get by the defenses.

However, the best criminals are nothing if not determined and resourceful, and as two PointPredictive executives discussed with Karen Webster during a PYMNTS conversation on Thursday (Sept. 12), paycheck stubs are easy to forge, and fraudsters can make a good living from the theft of vehicles or via the theft of money advanced for home purchases. That said, digital technology and so-called Big Data promise to make the job of such criminals harder in the future, at least according to the story told by Tim Grace, CEO at PointPredictive and Frank McKenna, the company’s chief fraud strategist.

Blind Eye Problem

As Grace described it, a big part of the problem stems from what he called a blind eye. With lenders trying to authenticate so much documentation — and do so within a reasonable amount of time to not cause friction for would-be borrowers — it is pretty much inevitable that some fraudsters will make it past the gates. Indeed, according to Grace, about 1 in 12 applications for auto loans involve some form of income fraud — that is, the applicant exaggerating the amount of income so to gain access to a bigger loan, which in turn can result in a vehicle that can be shipped overseas and sold for a healthy profit, or a vehicle that can be difficult to track down by repo professionals, given that the fraudster also likely used a fake identification.

“For $5 in four minutes, you can create a paycheck stub to say you make anything,” McKenna told Webster. Indeed, a Google search turns up multiple sites offering such services. Sometimes the fraudsters are incredibly brazen, at least from an honest consumer — McKenna’s company validates loan applications, and he told about checking bank statements of applications and seeing clear charges for those fraudulent paycheck stub sites. “Just verifying documents is not working,” he said.

So how exactly does one protect against such fraud, and that blind eye problem? It’s all about the data, combined with machine learning, the two PointPredictive executives said.

“We have a consortium of data,” Grace said, adding that the company’s data comes from eight primary sources — including automotive and other lenders, and salary and other business databases. That collection of data, in fact, includes more than 35 million automotive-lending applications, which have such vital information as to employment history and income levels. “We also know the ZIP codes of borrowers, and can tie incomes to ZIP codes,” he added, providing one example of how the company’s fraud-detection model works.

Detecting Lies

The idea — using all that data, mathematical models and machine learning — is to gain a strong sense of when an applicant is lying about income, and to flag those applications while reducing documentation and verification for honest, would-be borrowers. More specifically, the technology is designed to identify when an applicant’s supposed income is 15 percent or more than what should be the case in that particular situation, according to the predictive modeling, Grace said. He told Webster that the model could clear 75 percent of applications with a 90 percent to 97 percent accuracy rate. That leads to reduced friction for all parties involved — lenders, sellers and consumers. That remaining 25 percent is what lenders have to take a closer look at, as fraud can happen within that segment.

Those rates could get better over time as more data is added to the fraud-detection system, McKenna said. “We can get more predictive on capturing fraud,” he said.

As well, as more data becomes available, Grace said the system takes into account the gig economy — that is, relatively irregular sources of income for would-be borrowers. “Twenty-five percent of our data in our consortium comes from self-employed borrowers,” he said.

There are differences in fraud among various industries — income fraud at it relates to mortgages, for instance, tends to be more subtle and sophisticated, given that mortgage lenders have 30 days to approve or reject loans, compared to about 15 seconds for automotive finance providers, Grace said. However, in general, he said, the broader tactics are pretty similar.

More data, of course, means better predictions. Also, better predictions can mean less paper and documentation for honest consumers — that is, much less friction when it comes to obtaining vital loans.