Big data takes aim at COVID-19 chargeback fraud
With the globe pivoting to e-commerce during the ongoing coronavirus pandemic, an array of new opportunities have arisen for fraudsters, and merchants have found themselves being increasingly hit by different forms of chargeback fraud.
This new wave of fraud challenges older fraud detection models, because many consumers are drastically changing their shopping habits. Previously, a brand-new e-commerce account making a large purchase may have been suspicious; in 2020, it's a perfectly normal behavior from people new to shopping online.
By collecting more data on shoppers, merchants can gain visibility into traits other than those they had previously relied on to determine each transaction's legitimacy.
Chargeback fraud occurs in three forms: card-not-present (CNP) fraud, where a purchase is made with stolen credit card information; friendly fraud, where someone places an order but then falsely claims it was made without their knowledge or approval; and account takeover (ATO) fraud, where criminals exploit stored payment methods or hacked credit card details to access a legitimate customer’s account. In all forms of chargeback fraud, merchants typically have to pay out a chargeback, plus a fee to the bank in question.
“This creates a real drain on business revenue,” said Elad Cohen, vice president of data science at Riskified, a Tel Aviv-based fraud solutions startup. “Due to the fear of fraud, many merchants become overly cautious in approving orders, sometimes potentially turning away legitimate customers. That hidden cost is even higher than the costs of the chargebacks themselves.”
Over the past year, data indicates that both friendly fraud and ATO fraud have become particularly prevalent in the U.K. According to UK Finance’s Fraud the Facts 2020 report, ATO fraud has cost the U.K. economy £20.3 million, a 13% rise on the previous year; while a recent survey conducted by anti-fraud company Signifyd found that 35.5% of U.K. shoppers admitted to committing friendly fraud during the pandemic.
Cohen said that the rise of ATO fraud is a particular problem because it had the additional effect of breaking the customer’s trust in that merchant. A recent U.K. survey from Riskified found that 51% of customers would stop buying from a retailer if their account was compromised.
With the pandemic exacerbating different forms of chargeback fraud, merchants who have had to rapidly switch to e-commerce, and failed to implement strong authentication processes on the physical collection stage, have proved particularly vulnerable.
“Not properly verifying who is collecting the product or validating how they performed the transaction is a source of fraud,” said Jason Lane-Sellers, director of fraud and identity at global anti-fraud solutions provider LexisNexis Risk Solutions.
However, big data may be able to assist. One of the problems merchants have is that consumers now expect near-instantaneous order processing, which makes it far harder to screen for fraud without making the checkout process unduly cumbersome.
“Merchants may have just minutes or seconds to review orders, and they have to do so without adding friction to the customer’s checkout,” said Cohen. “We’ve seen substantial drop-off of good customers when forced to use 3DS. This generally means that merchants cannot be overly conservative or they risk increasing their false declines and losing hard-won customers.”
In recent years, anti-fraud companies have developed more sophisticated machine learning models which can take transaction details, and instantly compare those data points with billions of prior transactions to look for anomalies. “A well-crafted machine learning solution will maximize the merchant’s profit, carefully balancing chargebacks with false declines,” said Cohen.
However, during the pandemic some of these models have come unstuck due to unusual patterns of customer behavior becoming more common.
“A normal machine-learning model may have attributed a high risk to someone accessing a website with a new device for the first time,” said Lane-Sellers. “But during the pandemic, more genuine people have been doing this. This means your solution could incorrectly associate risk to these consumers until they’ve made a number of transactions, causing mistreatment of the consumer.”
Lane-Sellers says the way forward for anti-fraud providers in tackling chargeback fraud is to incorporate more data on behavioural biometrics — for example, traits such as voice, keystrokes when typing, and navigational patterns — into the next generation of anti-fraud solutions. “It’s important to collate more information around the digital persona performing the transaction to validate their identity,” he said.
Cohen expects that over the coming years, a flywheel effect will arise, where the most effective fraud prevention solutions will gradually attract more and more merchants, garnering more data in the process, and by doing so, becoming ever more accurate.
“Over time, the most accurate solution may gain enough visibility into new fraud patterns to quickly protect the entire network once fraudsters try changing their methods of operation,” he said. “As the accuracy improves, the benefit from this network will continue to grow, pushing more merchants to join.”