The aliases were giving Listia’s six-person tech team fits. An online marketplace and auction site based in Mountain View, Calif., Listia entices new customers to shop by giving them 500 credits as soon as they sign up. Predictably, there are users who abuse the offer. “All you need is a bunch of e-mail addresses,” says co-founder Gee Chuang. “And you could sign up for a bunch of new accounts, with each account having [the] free credits.”
Enter Sift Science. Its cloud-based service tracks consumer e-commerce behavior patterns and assigns every user a “fraud score” that recognizes patterns of fraudulent behavior. “Most e-commerce businesses have only one engineer to defend themselves,” says Jason Tan, co-founder and CEO of the San Francisco-based company. “We said, ‘Why not build a third-party solution and make it available to the world?'”
Sift Science’s algorithms constantly adapt to fraudsters’ changing tactics, updating statistical models in real time–a process Tan calls “machine learning.” In less than two months since offering its program to the public in March of this year, Sift had logged more than 1 million fraud patterns and variables in its database.
Some early findings: The more numerals in a person’s e-mail address, the more likely they are engaged in fraudulent behavior, and Gmail users are more trustworthy than Hotmail users. Buyers who select priority shipping raise red flags, because “they want to liquidate what they ordered as soon as possible and turn it into cash,” Tan points out.
“Our goal,” he adds, “is to leave no stone unturned. We are building statistical models that have millions of variations, soon to be billions. We’re building a Sherlock Holmes for the internet.”
For Listia, “bad users” with suspicious scores, such as those using multiple e-mail addresses to get the first-time free credit, or people who post items on the marketplace that they don’t actually own, are highlighted by Sift and tracked by the Listia team, or banned outright.
Listia wouldn’t reveal what it pays for the service, but Sift’s basic rate structure charges two cents per transaction; every month, the first 10,000 transactions are free. Chuang says Sift has greatly streamlined operations for Listia’s tech support staff. Before, staffers spent much of their day trying to track fraudulent behavior themselves, playing a virtual, never-ending version of Whack-a-Mole. “They can now concentrate on more pressing issues,” he says. “In terms of the hours it has saved us, it’s been tremendous.”
A Second Opinion
The folks at Sift Science “are not the first people to come up with this, but they are the first making something useful of it,” notes James Cannady, professor of information assurance at the Graduate School of Computer and Information Sciences at Nova Southeastern University in Fort Lauderdale, Fla.
However, he says, the tool is not a panacea: “Machine learning is not magic. It gives a system the ability to infer probabilities of activity based on what it has seen in the past, but if there is a truly new way of perpetrating fraud, it’s not going to recognize it. It is certainly advanced, but companies need to be smart about updating their own systems to catch fraud.”