FIRM: CoreLogic

PRESIDENT AND CEO: Anand Nallathambi

PRODUCT: LoanSafe Fraud Manager

U.S. PATENT 7,587,348: Pattern recognition fraud detection technology IDs fraud, slices false positives

The wide nets often deployed to fight mortgage fraud unwittingly snare a lot of innocent bystanders, which not only dilutes the effectiveness of prevention, but can also add unnecessary time and expense for both lenders and legitimate borrowers.

CoreLogic is attacking the false positive problem with its new LoanSafe Fraud Manager. The subsidiary of First American is using patented pattern recognition technology and access to the nation's largest repository of mortgage data to examine a wide array of consumer activities and information to alert financial institutions of potential fraud, a mix of analytics and access to information that's won the business of Wells Fargo and almost a dozen other lenders in less than a year- lured by the technology's ability to reduce the false positive to fraud ratio from the industry average of almost 100 to one to about three to one.

"The way fraud has always been ferreted out has not been [by] looking for abnormal behavior or patterns of fraud, it's only been a true/false check," says Tim Grace, svp of fraud analytics for CoreLogic. "If you check a name on an application against a public record and if it's the same person, then [the identity] is OK'd. If not, then a flag goes up." Grace says that approach is a "one dimensional" view of fraud detection in which name spelling or typographical errors in an application can be ID'd as possible fraud, setting off a needless investigation. "It sets off a lot of false alarms," he says. "In pattern recognition, no single data element looking abnormal is going to trigger a red flag. It's a number of characteristics that are looked at together."

Grace says LoanSafe changes the game by analyzing the behaviors and full profile of a borrower against what's considered to be inside the range of "normal" for that borrower based on a historical database. A borrower with a certain job will have a certain income and will be able to afford a certain sized-loan. How closely a new borrower's profile and behavior matches the "norm" determines whether that borrower is examined more closely. Pattern recognition technology is combined with Core Logic's huge property, loan and mortgage fraud database of more than 85 million loan applications and more than 200 million servicing records (the vast majority of mortgage data in the U.S.) to produce a risk score from one (low risk) to 999 (highest risk), which is then matched to a lender' business rules for further fraud study. The product also offers more loan information categories and alerts grouped by likely fraud types.

"False positives are a really big problem in fraud detection," says Ellen Carney, a senior analyst for Forrester, who says it's counterproductive. "The real fraud can slip through the cracks."

CoreLogic, which did not reveal users beyond Wells Fargo, is doing battle for marketshare against fraud protection products such as Interthinx's FraudGUARD, which offers electronic loan review, pre-funding and post-funding loan audits, mortgage broker and third party reviews, training, education and other services; and DataVerify's DRIVE, which offers identity verification, property valuation checks, and protection from employment and income fraud. Grace says CoreLogic "listened closely to our consortium members, which include representatives from the top ten mortgage lenders" in developing the technology.

Carney says the technology should find demand. "Fraudsters are getting a lot smarter and organized, and most of these financial institutions have done a poor job of finding fraud to begin with," she says. "If you think fraud was crazy in the [boom] days, imagine what it's going to look like in the modification days."


Subscribe Now

Access to authoritative analysis and perspective and our data-driven report series.

14-Day Free Trial

No credit card required. Complete access to articles, breaking news and industry data.