AI and big data are becoming better at identifying the 'tells' of fraudulent purchasing, but in a co-evolutionary process, the criminals are learning how the retailers' systems evaluate them in a never-ending game of cat-and-mouse. JL
Khadeeja Safdar reports in the Wall Street Journal:
When people shop online, their visits are tracked and shared with an outside firm that scores their behavior and decides whether to approve or deny purchases. Chip-card technology has reduced fraud in stores, but many retailers say they are now seeing the behavior shift online. A third of the 50 largest retailers in the U.S. say online fraud has risen 30% since they transitioned to the new chip technology. Evaluat(ions are) based on online browsing behavior, like transaction data and geolocation information.
When people shop online at retailers like Macy’s Inc. and Finish Line Inc., their visits are tracked and shared with an outside firm that scores their behavior and decides whether to approve or deny purchases.
Shoppers who exhibit behavior that such scoring firms often associate with fraud, such as buying without checking the return policy or paying for the fastest shipping option, have a higher chance of getting declined. Sometimes they never find out why their transactions are canceled or who made the call.
Behind these decisions is a group of firms hired by retailers to weed out fraudulent online customers. They determine which indicators correlate with fraud and then evaluate shoppers based on hundreds of data points. A Tel Aviv-based company called Riskified works with Macy’s and Finish Line. Its rivals include fellow startups like Signifyd and Forter, as well as players like CyberSource, which was acquired by Visa Inc. in 2010, and Accertify, which American Express Inc. bought the same year.
Online shopping has created new challenges for retailers trying to combat fraud, such as people using a stolen card or falsely reporting a transaction as fraudulent. Credit-card companies typically cover the cost of unauthorized purchases in stores. But with online transactions, retailers are usually on the hook when a customer reports a fraudulent charge. As card issuers can’t verify online purchases through chip-card technology and other measures, they take the money back from the retailer to reimburse the shopper, which means the store loses both the product and the cost of shipping.
Retailers are increasingly turning to third-party firms for help, but many don’t publicize which firms they use, why customer purchases are declined or the criteria for the decision.
Michael Green, 50 years old, found out about one of the firms by accident. He ordered headphones online for his son’s 18th birthday. Days later, when he hadn’t heard anything about the order, he contacted the headphone brand, Audeze, which told him the purchase had been canceled because a third-party firm had determined he was a fraud risk.
Mr. Green noticed that the status of his order said “Riskified Rejected.” When he emailed Riskified to ask why he had come up as a fraud risk, a customer service agent told him the company had no further information.Red Flags
Examples of indicators firms use to score shoppers as potential fraudsters, according to people familiar with scoring firms
- Not bothering to compare prices
- Choosing the fastest shipping option
- Not checking the store’s return policy
- Sorting by price and picking the most expensive items
- Using a proxy server--used to hide your location--when accessing the site
- Using an email address that doesn’t match the name on your Facebook profile
“There was no explanation, no appeal,” said Mr. Green, a financial professional in Austin, Texas. “Big data affects a lot of things, but you never hear of it affecting your ability to return something or buy something in the first place.”
Audeze CEO Sankar Thiagasamudram said it uses Riskified because, as a very small company, it can’t afford to police fraud on its own. “Unfortunately, we do not have any visibility into Riskified’s decision-making process,” he said.
After The Wall Street Journal contacted Riskified, CEO Eido Gal said Mr. Green’s order was incorrectly declined. “Riskified tends to be far more accurate and efficient than traditional fraud-prevention methods, but no solution is perfect, and we’re still improving,” he said.
Riskified evaluates shoppers based on online browsing behavior, along with other details like transaction data and geolocation information, and then issues approvals or denials at the time of purchase. A retailer can override a denial, but the retailer assumes responsibility if the purchase turns out to be fraudulent.
The threshold for issuing a decline varies by retailer and tends to be set lower when the products are more attractive to fraudsters. For example, Audeze’s headphones are more prone to fraud because they are compact and cost up to $3,995, making them easy to transport and lucrative for resellers.
Riskified said it works with more than 1,000 merchants. Some retailers use its services only to assess international online orders. The company said it provides its clients with tools to understand the reason for declines, but leaves it up to the retailer to decide what it wants to tell shoppers.‘There was no explanation, no appeal.’
A spokeswoman for Finish Line, a Riskified client, said that when the shoe retailer is contacted by customers whose purchases have been declined, it directs them to contact their banks because their information has often been compromised by some other means.
A spokeswoman for Macy’s said the company uses such services “for the limited purpose of fraud detection.”
“Like any company with a large online presence, Macy’s has a robust fraud detection system that combines our proprietary internal detection resources with multiple vendor solutions,” she said.
Chip-card technology has reduced fraud in stores, but many retailers say they are now seeing the behavior shift online. A third of the 50 largest retailers in the U.S. say online fraud has risen more than 30% since they transitioned to the new chip-related technology in 2015, according to a survey of members by the Retail Industry Leaders Association, a retail trade group.
Many retailers still use rule-based or manual systems, which decline purchases based on fixed criteria. Riskified said its use of machine learning and big data has reduced the number of erroneous declines.
“Our whole reason for being is to approve more transactions for the retailers,” said Riskified’s Mr. Gal. “Because we have so many resources and we have network effects across our merchants, we tend to have better performance than retailers managing this process internally.”
Retailers have been scoring shoppers to combat other types of fraud. The Journal previously reported that Retail Equation, an Irvine, Calif. service, has been generating risk scores on people to limit the amount of merchandise they can return.
0 comments:
Post a Comment