Frenetically Fighting Fraud

E-commerce retailers are constantly at risk from fraudsters, especially those sophisticated enough to assume multiple false identities. To protect themselves, many e-tailers use fraud prevention tools to alert them to potential risks. To more fully evaluate those risks often requires long hours by investigators, who manually review customer orders that have been flagged as risky.

Online merchants reported an average fraudulent order rate of 1.1 percent in the United States and Canada last year, according to Cybersource’s 2009 Annual Online Fraud Report. The average number of online orders reviewed manually has fluctuated between 22 and 27 percent the last six years, according to the report.

The manual review rate is dependent, in part, on the robustness and flexibility of the retailer’s fraud prevention application.

Overstock.com implemented FraudNet for Ecommerce two years ago. The solution from Scottsdale, Ariz.-based 41st Parameter has cut the retailer’s fraud review costs in half while reducing the rate of fraud “significantly,” says Alan Johnson, Overstock’s director of payments and fraud prevention.

“I was impressed with what this tool could do,” he says. “The people at 41st Parameter are fraud people. They eat, live and breathe fraud.”

In addition to their expertise in fraud prevention, 41st Parameter had what Overstock considered “a very flexible system, and that was very important to us. We really like the ability to update rules in real time,” Johnson says. “We have been able to get our review rate down to what we wanted.”

FraudNet also has the unique ability to lift signals from electronic communication devices, which gives e-tailers information over and above what customers type into their online orders.

With approximately 400 algorithms or rules, FraudNet “gives us the ability to fine-tune what we need to drop into our risk queue,” Johnson says. “Before, we’d go through this long process for changing each rule … [now] a change is right at your fingertips.”

If a member of Overstock’s LP team sees a significant amount of fraud in any one particular area, “we can go in and increase the risk rules on the fly,” he says. “And we can do that for any of the algorithms. That’s what makes this tool so flexible.”

41st Parameter states that FraudNet collects more data than other fraud detection systems and allows them to write more rules that help them apply the data for different types of circumstances.

“It’s not only good to have the data, but you need to apply it in the right way,” says 41st Parameter founder Ori Eisen. Among these rules are the abilities to match the time zone of the billing address to that of the computer from which the order was placed, as well as match the language and locale of the device to that of the address provided by the person placing an order.

FraudNet also performs velocity checks to see if an unusual number of orders are being placed within a short time span from the same PC or other device. That service is available through PCPrint, a module that allows Internet retailers to identify every computer or communications device that connects with its website.

Detected by DataSpider
The information being collected includes, but is not limited to, the language in which the device is configured. If, for instance, someone places an order with an address in New Jersey but the PC is configured in Russian, that could indicate a potential fraud risk.

FraudNet also has link analysis capabilities via a module called DataSpider. This means, says Eisen, that once FraudNet has found a clear case of fraud, DataSpider can troll through the data to search for other orders that can be “linked to that fraud case.” This link could be a common e-mail address or phone number or the “fingerprint” of the browser used in a communications device.

DataSpider can be used to find fraud across multiple channels, not just orders submitted through the web (there is no additional cost to use any of the FraudNet modules across multiple channels).

FraudNet uses another application, called EZKeys, to root out incidences of what Eisen calls “lazy key strokes.” A right-handed person, trying to quickly type in a great deal of information, might be inclined to hit four horizontal keys – ASDF — located on the left side of a keyboard; a left-handed person might type in JKL:. As it turns out, says Eisen, “a lot of fraud orders” contain these combinations instead of a person’s name “because they are the easiest keys to type.”

FraudNet has helped Overstock.com link a number of orders that “we previously weren’t able to link, and that helped us catch and prevent further fraud from that ring,” Johnson says. And the ability to link cases has proven very effective not only against individual criminals but against gangs, significantly reducing the incidences of fraud perpetrated by organized retail criminals.

Prioritizing with InvestiGate
Another attribute of FraudNet is a user interface that includes InvestiGate, a risk assessment workflow process that lets investigators queue up and prioritize high-dollar risks when time is an issue – such as 20 minutes before UPS is due to pick up overnight orders.

FraudNet also allows retailers to assign cases based on their complexity; less complex cases to junior members of the fraud prevention team and more serious and/or complex cases to more senior members.

Overstock.com management places a high priority on being “non-obtrusive” in the fraud detection process, so “FraudNet is big win for us” in that context also, Johnson says. “We’re contacting fewer customers; we’re holding up fewer orders. And because we’ve cut our manual reviews in half, we’re able to allocate more of our staff’s time to enhancing the customer experience.”

FraudNet also helps e-tailers work with law enforcement to catch criminals committing fraud. One Internet retailer that prefers to remain anonymous “increased its arrest rate 50 percent within the first year of using FraudNet.” Eisen says.

The bottom line is that Overstock.com is “reviewing a lot fewer orders but our fraud rate is better than it was. That’s exactly what we were after,” Johnson says. “We saw an ROI in less than a year and we’re extremely happy with our very low [rate] of attempted fraud.”