‘Tis the Season to Hunt Fraudsters (with Big Data)
Everybody loves Christmas–especially the fraudsters who cloak their criminal schemes in holly, tinsel, and the rush for last-minute gifts. For the folks at ecommerce fulfillment giant eBay Enterprise, keeping the fraudsters away has involved a judicious application of big data tech.
eBay Enterprise provides omnichannel fulfillment services for hundreds of brand-name merchants, ranging from Ace Hardware to Zales. The King of Prussia, Pennsylvania-based company, which recently separated from eBay Inc., does everything from hosting ecommerce websites on behalf of customers to operating call centers. With more than 7,500 employees, it shipped more than 174 million items in 2014 from more than two dozen distribution centers, giving it revenues of $1.24 billion for the year.
As the merchant of record for its clients, eBay Enterprise absorbs all fraud-related losses incurred by its clients. That gives it a powerful incentive to spot fraudulent orders before they can be fulfilled. In 2014, the company prevented $55-million worth of fraudulent transactions, and that number is expected to rise this year.
As we get closer to Christmas, the fraudsters use the last-minute buying spree as cover for their dastardly deeds. Tony Ippolito, the strategic risk and technology manager for eBay Enterprise, recently gave Datanami the low-down on how the company uses big data tech to spot the bad transactions, and how that job gets tougher as calendar counts down to December 25.
Fraud Clamp Down
“We like to say data is gold to us. The more data we have, the better,” Ippolito says. “A lot of it isn’t meant to capture fraud. It’s meant to whittle down the good orders that are coming into the system. The fewer good orders you have to sift through, the more fraud you have in the ones left over.”
Most of the fraud that eBay Enterprise uncovers is perpetrated by large and sophisticated criminal organizations that are using stolen credit card information to buy high-value goods—such as video games, electronic gadgets, and designer clothes–that can easily be re-sold on the black market.
“The basic fraud patterns involve long-distance and high-dollar orders, and expedited shipping,” Ippolito says. “The higher the dollar amount and the quicker they want to receive the goods, the riskier the order or the transaction becomes. It’s calling out that you need to be sure you have tight controls on those particular conditions, which can vary depending on the industry, but that’s the overall pattern.”
Spotting those phony transactions requires the company to quickly cross-check a range of fields—such as names, email addresses, billing addresses, and shipping addresses—against an Oracle database containing 1.3 billion entries. “We also collect as much information as we can about the product, the kind of item, and the amount of the order,” Ippolito says. “We do device fingerprinting, we collect IP address, and then we do geolocation lookups” based on time zones and countries.
All told, eBay Enterprise runs its transactions against a decision-management system equipped with about 600 rules, and subjects the transactions to one of 20-plus models that use machine learning algorithms to match incoming transactions against various known fraud patterns. “It’s a lot of data collection and aggregation, and seeing trends and applying that across the board to make sure we’re not missing anything,” he says.
Data scientists at eBay Enterprise use the information to get a better picture of how fraudsters operate, and where they might be headed. Like the game “Whack-a-Mole,” the strategy is designed to keep the dirty little scoundrels moving.
“We’re working on a precept of demoralizing them to the point where it’s so hard for them to take money from our sites or our partners’ websites that they move into other areas,” Ippolito says. “They’re very much like programmers. They’re inherently lazy and choose the path of least resistance. If we impose a certain amount of friction to them, they’re going to move where it’s easier to defraud people.”
Keeping one step ahead of the fraudsters requires eBay Enterprise to pay close attention to the data, and tweaking the fraud rules on a regular basis. For instance, if the company implements a rule that requires it to “queue,” or manually review, every transaction over $50, then the fraudsters will move their target to orders under $40. If eBay Enterprise moves the manual review threshold to $40, then the fraudsters may move their shopping targets to $35.
“When you close off a certain area, you have to be aware of what the next logical step for them is,” Ippolito says. “If you shut down overnight shipping, then they’ll move into third-day shipment. It’s a lot more nuanced than that, but that’s the general idea.”
It’s fair to say that eBay Enterprise is an expert in fraud prevention. The company recently shared some of its fraud findings with its first-ever Holiday Fraud Index. Among the findings: be wary of gift cards, which are very popular during the holidays but are notoriously hard to track. Scammers are aware of their advantage, which is why gift-card related fraud peaks on New Year’s Day—the same day that retailers see a rush of returns and requests for refunds placed on gift cards.
While eBay Enterprises enjoys a relatively lower level of fraud than the industry as a whole, the company realizes that it can’t sit still when it comes to fraud. To that end, the company is moving forward to adopt the next-generation of big data technologies, such as Apache Kafka and neural networks.
“We’re putting our foot forward into emerging technology and making sure we’re at the cutting edge,” he continues. “We’re implementing a process where we’re continuously evolving, because if we don’t, the fraudsters will, and then our industry leading rates won’t be industry leading, so we have to make sure we’re always ahead of the curve.”