Will PSD2 be AI's big break in fighting payments fraud?
In recent years, new machine learning algorithms and big data have reduced fraud losses to an extent — however, their impact has been relatively limited, in part because the industry has been reluctant to use them. But the use of such technology is soon likely to become far more widespread in the U.K., and across the EU.
Nearly half of all fraud incidents are made possible by a lack of advanced anti-fraud controls in the businesses targeted, according to a 2018 study conducted by the Association of Certified Fraud Examiners. This problem will likely be addressed by new PSD2 regulations, which come into effect in the second half of 2019 and require all transactions over £30 to have stronger authentication measures.
In particular, all payment providers will now be required to conduct real-time risk analysis on transactions to assess a range of factors including any abnormalities in behavior or spending, previous purchase patterns, and location of the customer and business.
“As this legislation is implemented, it will lead to the roll-out of more sophisticated solutions across organizations on both sides of e-commerce transactions,” said Dave Excell, founder and chief technology officer of Featurespace, an analytics company developing anti-fraud solutions for a range of companies including WorldPay.
These regulations are particularly hoping to reduce unauthorized fraud — where the account holder does not provide authorization for a payment and the transaction is carried out by a third party — which increased by 10% last year, according to the latest U.K. Finance data. As a result, numerous surveys have shown that the confidence of customers in e-commerce is steadily eroding, with Paysafe finding that 65% of online consumers now regard payments fraud as an inevitable part of shopping online.
“We already work with card issuers, helping them decide whether to facilitate a transaction request or not," Excell said. "And we work with retailers through their payments providers, trying to understand whether a transaction should be authorized, or whether it will result in a charge-back because it’s not the actual customer making that transaction. This technology should reduce fraud on both sides of that equation.”
One of the problems tech firms have always faced when using artificial intelligence to detect fraud is that fraud patterns are constantly changing as criminals find new ways to evade the system. However, many companies including Featurespace have begun to adopt a radical new approach which may well prove far more effective. Rather than using transaction data to try and model fraudulent behavior, they are instead designing intelligent systems which model ideal customer behavior, and then report anything which falls outside of this box.
“The challenge has always been trying to make sure we have a solid baseline to make a yes/no decision against,” Excell said. “Trying to model fraudulent behavior isn’t easy when fraud patterns are always evolving. But by taking the opposite approach, and modeling good customer behavior, behaviors which we want to allow, and have a frictionless experience, this gives us a baseline which we can trust and use as a premise. Anything that lands outside of that can be looked at more closely, and a decision made if it should be flagged up for further investigation or whether the transaction should be stopped if it seems suspicious enough.”
In addition, new tools which allow anti-fraud systems to rapidly access and process large amounts of historic transaction data for an individual cardholder are expected to drastically increase their accuracy.
“Instead of comparing new transactions against global patterns of customer behavior, our software will be able to build up an individual profile of a customer’s intentions and behavior,” Excell said. “When a new transaction comes through, it can quickly look at how well that fits with our expectations of their behavior. So for example, if someone goes to the Co-op, we will know whether they usually insert their card or make a contactless payment. And if all of a sudden, the customer appears to be using a different payment method at that brand of shop, the system will detect that this is unlikely to be the customer.”
Only time will tell whether this can make a genuine impact, but while some skeptics say fraud will remain an ever present issue, those in the tech world are optimistic that new approaches can reduce the scale of the problem.
“I think the landscape will change in quite a few ways in 2019,” Excell says. “As systems get upgraded and companies choose to deploy more of this technology then I would expect to see those fraud losses go down.”