Banks Need More Robust Analytics to Thrive - and Even Survive According to a 2016 Juniper Research infographic, online transaction fraud is expected to hit $25.6 billion by 2020, meaning $4 of every $1,000 spent online will be fraudulent. The research firm also reports that retailers stand to lose $71 billion globally from fraudulent card-not-present transactions over the next five years. As settlement for ACH and wire payments accelerates, retail banks need to focus intently on catching fraud in real time—or risk losing business. One key tactic will be a greater reliance on machine learning and adaptive behavioral analytics. These technologies will not only help banks reduce real losses, but also cut the number of genuine transactions that get turned down—incidents known as “false positives.” Wire fraud alone has had devastating effects. By one measure, email account compromises amounted to losses of $361 million worldwide in 2016, up 47 percent from a year earlier, according to data compiled by the Internet Crime Complaint Center, a division of the FBI. Outstanding risk is even higher: exposed dollar losses—including incidents where funds were recovered—amounted to over four times as much. Fraudsters also hack email accounts to target ACH transfers, with real estate being a common target due to the large sums of money involved. In one U.S. case, (in the state of Georgia), criminals hacked the email account of a real estate agent representing the sellers of a house. While machine learning and adaptive behavioral analytics have been around for years, many banks haven’t embraced this technology due to the perceived cost of investment. Financial institutions also historically had the luxury of time: up to two days to stop suspected fraudulent transactions, the historic standard for ACH settlement. Yet now the faster world of ACH and wire payments is changing the dynamics. Institutions have started processing same-day digital ACH debit transactions. Same-day payments can settle in hours, increasing the chances that fraudsters can steal large amounts of cash. Meanwhile, settlement times for wire transfers now take as little as 10 minutes to complete. ACH and wire fraud might lag credit card fraud in terms of the number of hits, but they tend to be much costlier per incident. In one common instance, fraudsters first use phishing emails to obtain logins and passwords, then hack the email accounts themselves. These hackers will then pose as highlevel personnel, such as chief executives and chief financial officers, tricking accountholders into transferring funds to false recipients, often under a pretense of urgency and secrecy. The hackers obtained the contact information for an attorney that was representing the buyers of the home. Posing as the seller’s real estate agent, they sent an email to the attorney requesting funds be sent to a fraudulent account number. The attorney unwittingly obliged, thinking it was the real estate agent representing the seller, and $200,000 was gone. The victim is now suing his bank and the attorney. Transaction fraud management redefined As ACH and wire fraud escalates, it’s not hard to see why stronger real-time monitoring is needed. Fraudsters have become more organized and sophisticated, some of them former bankers working in real offices and some even working inside actual banks. They are constantly changing how they’re interacting with financial institutions to pose as customers or hack into email accounts. In this environment, banks need more than traditional static fraud monitoring systems, since they degrade quickly as fraudsters change tactics.
Adaptive behavioral analytics can be used to detect anomalies in real time to identify new types of transactional fraud when they occur, without time-consuming (and expensive) manual intervention. The technology also has become extremely cost-effective to implement. Adaptive behavioral analytics brings two key components to help financial institutions minimize fraud and reduce friction: 1. Individualization Rather than only applying global rules to a vast data set of thousands of customers, behavioral analytics focuses on building anonymous statistical profiles of individuals, using hundreds of inputs. This micro approach helps banks understand changes being made to the account, to contextualize what is going on, and detect anomalies across multiple, complex data collectors. For example, real estate attorneys have very different transaction patterns versus those of printing businesses, so it’s important that a platform is able to contextualize the types of transactions a bank would expect. 2. Machine-learning capability Models using adaptive behavioral analytics can teach themselves. They can “study” customer transactions and detect different patterns of fraud, constantly updating their assessment of what is fraudulent and what is not. By contrast, static models will degrade quickly. Adaptive models are “non-degrading systems,” obviating the need for a massive update each year, or continuous manual updates by staff. Machine-learning models can also constantly tweak how they weigh inputs to make an accept or reject decision, including how accountholders logged on, what their usage patterns are, the destination of the funds, and how long accounts have been opened. David Excell Co-Founder & CTO at Featurespace Dave is Co-Founder and CTO of Featurespace. He studied with Professor Bill Fitzgerald (Featurespace Co-Founder) at the University of Cambridge, where he developed the ARIC platform. Under Dave’s leadership, Featurespace has grown from a concept to a commercial success with many blue-chip customers. Dave has been awarded 11 prizes and scholarships for his academic and commercial achievements, including the 2011 ITC Enterprise Award for Young Entrepreneur. A practical approach to fraud management The fundamental takeaway to this changing landscape is adaptation and evolution. Fraudsters are not static, so risk models should not be either. They must be able to detect behavioral anomalies in real time to identify new types of transactional fraud when they occur, without manual intervention. This reduces false-positive alerts, allowing banks to accept more genuine transactions by better understanding the legitimate behavior of individual customers. It’s not a question of whether banks can fight fraud, it’s how efficient and cost effective they can do it in a world of shrinking payment windows for ACH and wire transfers. Machine learning and individualization will help banks adapt rapidly and avoid degradation of their transaction fraud monitoring. Early adopters of this practice will ultimately have a competitive advantage over their peers due to decreased fraud expenses, increased operational efficiency and most importantly, a better customer experience. About Featurespace Created the ARIC platform, a real-time, machine learning software system. With offices in the UK and US, Featurespace has deployed ARIC to financial services and gaming organizations that have services or products in more than 180 countries. Customers include Ally Bank, TSYS®, Worldpay PLC, Playtech, PaddyPowerBetfair PLC, Vocalink Zapp, CashFlows and William Hill. 13