/special_report.gif

 


[Collections & Credit Risk—Table of Contents] August 2002
SPECIAL REPORT:


Fighting Fraud At the Front-End


With criminals constantly searching for new ways to defraud credit grantors, vendors are rolling out a new arsenal of anti-fraud weapons designed to thwart them.

By Peter Lucas


Click to View Table


Mention fraud to lenders, merchants, or telecommunications companies, and they are likely to cringe. That’s because, no matter how thorough a business may be, a criminal is always coming up with new techniques to commit fraud.

From auto loans and credit cards to healthcare and mortgages, no industry is immune if there is a chance criminals can walk away with money or merchandise they can sell. It is a daunting problem that collectively costs lenders, insurance providers, telecommunications companies, and merchants billions of dollars annually. Losses due to fraudulent applications alone total about $35 billion annually, according to fraud prevention experts.

Telecommunications represents another hotbed of fraud. Global telecommunications fraud costs about $14 billion a year, according to the World Trade Organization. Healthcare providers lose roughly $100 million annually to fraudulent claims, the Health Care Financing Administration reports. About 12% of all mortgage applications contain fraudulent data, according to the Federal Bureau of Investigation.

Some fraud experts argue that fraud estimates are conservative, since firms are loath to reveal actual losses to avoid giving the impression they have poor risk management policies. Furthermore, competition among lenders and telecommunications companies for new accounts has made so-called easy credit applications, which require little documentation from applicants, commonplace. As a result, many lenders perform fraud checks on the back-end of the application process, which can take days to complete. “That’s pretty much the equivalent of shutting the barn door once the horse is out,” declares Ron Litt, vice president and chief information officer for Houston-based Allied Home Mortgage Co. “Good quality control starts at origination.”

No question that fraud is a painful thorn in the side of lenders and businesses. But there are hopeful signs. Software vendors are creating tools capable of speedily detecting questionable consumer data on the front-end of the application process.

Deploying fraud detection systems on the front-end can significantly reduce fraud. Royal Bank of Canada, for example, has reduced fraud losses in its card portfolio nine basis points of volume from 21 basis points, since deploying a fraud detection system in late 2000, according to Tom Spillane, vice president of marketing for Hazlett, N.J.-based Retail Decisions Inc. Other fraud experts claim anecdotal evidence shows such systems can reduce fraud losses by as much as 50% in the first year.

Fraud detection programs’ effectiveness lies in their ability to draw upon large databases of information to spot common fraud patterns. Some databases contain information from different industries. “The patterns of fraud don’t vary much by industry,” says Lyn Porter, vice president for Orange, Calif.-based Experian Fraud Solutions. “It does not matter whether you are an auto lender or telecom provider, criminals use many of the same techniques to perpetrate fraud.”

For example, criminals may list a stolen Social Security number, phony address, or phone number. They may present false W-2 statements or have friends pose as employers to verify employment. “We’ve seen appraisers and even employees involved in the schemes,” Litt adds. “It’s not uncommon for fraudsters to get inside help.”

To help reduce losses from fraud, Allied Home Mortgage recently rolled out an automated fraud detection system that allows loan officers to review files one at a time or in batches. The system checks to see if an applicant has a criminal record. It also validates phone numbers, Social Security numbers, birth dates, addresses, business locations, and employment, as well as other criteria.

The system also performs a property-flip evaluation that compares the appraised value of the home being financed to other houses in the immediate neighborhood, if the appraisal is available at the time of application. Property flips are important because criminals often purchase a low-priced property and then apply to refinance it within 90 days, taking cash out of the new mortgage. To do that, fraudsters typically work with appraisers to inflate the value of the property. Florida and the Northeast are hotbeds of mortgage fraud, according to Litt. “Once they refinance they can go to another lender and another and so on,” he adds. “Once they are done working an area, they move on. It’s not a hi-tech crime, you just need an appraiser that is dishonest.”

Indeed, most fraud is perpetrated using less-than-sophisticated techniques. Mail order/Internet fraud, in which consumers purchase an item to be delivered to an address without a credit card present, typically is committed using a variation of a single address. “This technique is used to trick negative lists of bad addresses that are manually read by merchants,” explains Jeff King, director of product management for Mountain View, Calif.-based CyberSource Corp. “The address may vary, but the merchandise usually gets delivered to the same address each time because the post office or delivery service knows where it is supposed to go.”

CyberSource has developed fraud detection applications for retailers in a card-not-present environment. CyberSource Advanced Fraud Screen uses neural networks, rules-based applications, and a fraud detection model developed by Visa U.S.A. “Fraud is a constantly moving target and no one technology is going to be that effective a fraud detection tool,” says King. “You really need applications that view data in different ways to detect a wide variety of fraud.”

Building effective fraud detection applications, however, depends on the quality of data used to build the scoring model. Until recently, lack of fraud data was a hindrance. In addition to concern that publicizing fraud might reflect poorly on risk management policies, lenders preferred not to release fraud data for fear of being held in violation of privacy laws. So fraud detection models were built primarily using data supplied by the end-user. As a result, it was difficult for the models to flag a questionable address or Social Security number the first time it was used in a specific industry.

“The best anti-fraud solutions come when information is shared across industries, because it provides a fuller picture of fraud patterns as criminals move from one industry to another,” says Experian’s Porter. “It has taken some time, but major lenders in the U.S. are getting on board with data sharing.” Experian helped pave the way for data sharing by establishing reporting guidelines that conform to privacy laws and FCRA regulations. The data is compiled in the National Fraud Database, which was created about five years ago and is overseen by a steering committee. Participants include American Express Co., Toyota Financial Services, Sprint PCS, Dell Financial Services, and Bank One Corp./First USA.

In addition to the National Fraud Database, Experian offers a suite of fraud detection products that validate information on a credit application or authenticate an individual’s identity by asking questions based on information contained in the individual’s credit report but unlikely to be obtained by stealing a wallet.

Bringing together fraud data from a breadth of industries promises to make a huge difference in the quality of neural-based fraud detection models. “When you use questionable data to train a model, you get garbage going out of the model,” says Retail Decisions’ Spillane. “With lenders now labeling fraud data more accurately, rather than as bad debt, it’s a lot easier to train a neural model. Proper labeling of data is essential to the effectiveness of neural networks.”

Retail Decisions offers neural-based and rules-based fraud detection applications for merchants in the card-not-present market, an area criminals have begun to target due to the popularity of e-commerce. Card-not-present fraud totaled about $700 million for online merchants in 2001, Spillane says. For some online merchants, fraud losses can total as much as 5% of sales volume.

When questionable information is spotted, Retail Decisions’ fraud detection applications – including Prism, acquired from Providence, R.I.-based Nestor Inc. in 2001 – recommend an action, such as denying a transaction or requesting the customer use another card or a check. The system may suggest the merchant call the customer while they are online to verify the purchase.

Neural networks are also proving effective at detecting fraud in the telecommunications industry, where between 1% and 5% of revenues for cell phone calls are lost to fraud. San Diego-based HNC Software Inc. has developed neural-based models to scan cell phone applications on the front-end for suspect data, such as the Social Security number of a deceased individual.

Once a phone is activated, HNC tracks call patterns, such as duration of the call, geographic destination called, call frequency, and so on. “Someone may fraudulently apply for a phone and then sell that service on the street on a per-call basis, pocket the money, and never pay the monthly bills,” says Michael S. Chiapetta, executive vice president of analytical products for HNC. “Competition is so fierce among telcos that they may be reluctant to do anything that may reduce their ability to sign new customers that are a good risk, so you need back-end screening to cut off service when fraud is suspected.” HNC is also developing a cross-industry database of fraud information from telecommunications and financial services companies.

“There are a lot of advanced tools available for fighting fraud,” says Les Reidle, a senior vice president with Speer & Associates, an Atlanta-based consulting firm specializing in payments systems. “And with banks starting to aggressively mine their investment in data warehousing, these tools will get better with time.”

For lenders and credit grantors looking to reduce fraud losses, that day can’t come soon enough.


Current Issue
Return (Back to TOC)


[Subscribe Now!]