AI in Fraud Prevention: Identifying Suspicious Orders, Convincing Suspicious Merchants

AI in Fraud Prevention: Identifying Suspicious Orders, Convincing Suspicious Merchants

January 16, 2020

Fraud costs online merchants money. It’s as simple as that. In fact, a study from Juniper Research finds fraudulent transactions are expected to cost online retailers more than $71 billion in the next few years. And e-commerce fraud is growing. Experian finds it has grown by 60 percent since 2016.

“Online fraud affects the seller,” Gayathri Somanath, senior director of product management with Signifyd. “When fraud is committed with a credit card in a brick-and-mortar store, the merchant is protected by agreements with card issuers. But when fraud occurs online, in other words when the credit card isn’t present, the liability for the bad charge falls to the retailer.”

The immediate effect is obvious, says Somanath. The retailer is out the value of the product and additional fees and costs that are levied in the case of fraud. Inevitably, the merchant will face a chargeback filed by the rightful owner of the credit card, who understandably wants to be repaid for the fraudulent charge on his or her credit card.

“If a retailer faces too many chargebacks, things will get complicated with the credit card companies, which will increase the fees they charge the merchant, and with payment processors, which will raise fees, or possibly refuse to do business with the merchant,” says Somanath.

And, of course, there is the reputational penalty, as well, according to Somanath.

“Fraud rings are sophisticated and like any good enterprise, they follow industry trends. If a merchant becomes known as a profitable fraud target, it will experience additional fraud attacks until it is no longer an easy mark.”

But e-commerce retailers must process many—in some cases thousands—of orders daily. Without those transactions, they are out of business. So the very nature of their business opens them up to card-not-present fraud. And as Somanath notes, unlike brick and mortar purchases, CNP purchases are much harder to verify.

According to Louis Columbus, Principal, IQMS, part of Dassault Systèmes, in the face of more sophisticated fraudsters and fraud rings, traditional rules-based tools alone are becoming inadequate for verification in today’s fast-paced e-commerce market.

“Rules-based engines and simple predictive models could identify the majority of fraud attempts in the past, yet they aren’t keeping up with the scale and severity of fraud attempts today,” says Columbus in a recent post on Forbes. “Fraud attempts and breaches are more nuanced, with organized crime and state-sponsored groups using machine learning algorithms to find new ways to defraud digital businesses. Fraud-based attacks have a completely different pattern, sequence, and structure, which make them undetectable using rules-based logic and predictive models alone.”

 AI for the future of e-commerce fraud prevention

 Increasingly, online merchants are turning to Artificial Intelligence (AI) solutions, which learn from each transaction, and improve the accuracy of approving transactions.

“The idea is to ship every single order placed by a legitimate customer without shipping any orders placed by a fraudster,” says Somanath. “AI is the way the most progressive fraud protection strategies manage that. Unlike the more traditional rules-based systems, AI adapts and learns as fraud patterns and trends evolve, scaling the ability to continuously protect customer experience and deterring fraud.”

The use of AI for detection means the technology can recognize a suspicious order and block it or send to a human for review. But the ultimate goal is to be able to use the technology for speed and accuracy, which has “everything to do with customer experience,” says Somanath.

“Barriers, of course, slow commerce and cause frustration for shoppers. Maybe a customer has to log into a site before making a purchase. Maybe an order won’t be accepted if a customer is shipping it to an unfamiliar address. A fraud system using AI removes all those barriers.”

There is less chance of frustration on the part of the customer, but using AI tools also speeds up detection to the benefit of the seller as well, says Columbus.

“AI makes it possible to detect fraud attacks in real-time versus having to wait six or eight weeks until chargebacks start coming in,” he says. “When a digital business relies on structured learning and rules alone, new attacks are very difficult to catch. Chargebacks show up 6 to 8 weeks after the fraud has taken place, and digital businesses rush to update their rules engines. By balancing supervised and unsupervised learning, AI alleviates the need always to play catch-up to online fraud.”

Merchant concerns around AI

 Naturally, the newness of AI for fraud prevention—and sometimes the perceived cost associated with AI-based solutions—means merchants are wading in with some tentativeness.

“Many merchants have been bombarded with marketing messages about how AI and ML can prevent fraud, but analysts haven’t really seen the concrete examples that show how the technology is actually going to help them do their jobs better,” said Jeff Sakasegawa, a trust and safety architect with Sift.

Sakasegawa says Sift hears some confusion and questions around just how AI will work. And potential users want to know how the technology will improve metrics like blocked transactions, chargeback rate and reduction in manual review.

“There’s a lot of mental fatigue in the market regarding AI, or just varying messages on the technology which differs from company to company. This leads to incomplete or inaccurate perceptions which can be clarified in additional conversations,” he said.

The technology is often straight forward to implement but can take some time to reap benefits as AI tools learn to understand what to look for when it comes to fraudulent behavior.

“If a merchant is starting from square one, the data-gathering period is informed by the signals that merchant generates, which is essentially based on how many people are actively engaging with their business,” said Sakasegawa. “A high-value purchase on one merchant’s website may be an indicator of fraudulent activity, while for another merchant, be completely normal based on that merchant’s business and the goods or services they sell. Ultimately, AI and ML work best when paired with rich data sets that provide the proper context for assessing trust and risk.”

But while the longer-term picture of the effectiveness of AI is still unfolding, Sakasegawa said the benefits, in his view, are already clear.

“While the hype around AI and ML can be exhausting, these technologies can mitigate risk and losses and substantially increase revenue.”

Previous-Article-CNP    Next-Article-CNP

BOT Management

 

  • Share this Article:
Joan Goodchild