November 22, 2016By Luke Reynolds, Fraud Director, Featurespace Last year, CNN reported that over $3 billion was spent online on Cyber Monday in the U.S. alone. With online purchases predicted to hit another high again this year, banks, card issuers and merchants are faced with a dual challenge. Not only are they coping with additional website load during the spike in online purchases, but their fraud systems churn out vast volumes of alerts, due to the increase in new types of customers, different payment patterns and huge volumes of transactions over a condensed time period. Naturally, there’s an increased risk of fraud that comes with high volumes of transactional activity. Surprisingly, however, the bulk of operational costs come from managing false alerts which incorrectly block genuine customers. So how do banks, card issuers and their merchants manage customer acceptance while still identifying true fraud? Accepting bumper business without taking on more fraud risk Having worked for more than 20 years in risk and security in the banking sector, I’ve seen the impact on both customers and merchants when good customers are blocked in an effort to catch fraud. During the seasonal period from November through to the New Year, criminals will be taking advantage of increased online activity to find the weakest links and commit large-scale fraud. The number of fraud alerts produced for analysts to manage, however, is often also the fault of fraud systems being overly sensitive to different payment patterns as people react to seasonal offers. To cope with the increased volumes of alerts, many merchants and financial organizations tweak their fraud system rules. Dialing down rule thresholds is a risky position to take. On the other hand, dialing up the thresholds on unusual activity can have a negative knock-on impact by blocking genuine customers. More blocked customers means more unhappy customers—always bad for business. So how can financial institutions and retailers maximise customer acceptance without risking accepting more fraud? Truly understanding customer behavior: adaptive behavioral analytics What I’ve learnt from years battling fraud in the retail banking and cards industry, is that financial institutions need fraud systems that stay one step ahead by understanding every individual customer’s behavior in real-time—protecting them from fraud, while providing a frictionless experience. Instead of ratcheting up the thresholds on incumbent systems, the answer is to understand each customer’s behavior better. Identify what is normal and what is anomalous—or uncharacteristic—behavior, even within the profile of a genuine customer. Most approaches to behavioral analytics focus entirely on whether or not the person is who they say they are. A different approach is to enable merchants and card issuers to understand the behavior of individuals, and detect when a customer may be starting to change their behavior in subtle ways. So how can financial institutions or merchants spot truly suspicious transactions when a customer’s behavior is constantly changing as they respond to deals? The answer lies in the latest machine learning technology, which combines cutting-edge research from the fields of computer science and data science. Profiling context: Cyber Monday (and beyond) During last year’s Cyber Monday, mobile sales grew by more than 53 percent in the U.S.—over $838 million was transacted via mobile. A machine learning system, which understands this purchasing context, is able to take the bigger picture into account to make informed fraud decisions around an existing customer interacting on a new channel. Machine learning systems automate the process of viewing events in context, building a deep understanding of every single customer. By monitoring every event and transaction taking place in real-time and from multiple channels, fraud attacks stand out and genuine customers are easy to recognise. All this takes place from one automated fraud prevention system, in real time, within the authentication stream. This is crucial for fraud prevention. Up until now, we have largely relied on rules-based systems and human analysts to look at data to spot trends and anomalies. However, this is becoming an increasingly inefficient way to analyze the vast volumes of data that financial institutions and their merchants receive on a daily basis. Cyber Monday is just one example of a day when spending patterns change. The reality of online purchasing is that offers and deals are frequent, and they change the way that customers behave. 70% reduction in genuine transactions declined Machine learning systems look at complex data sets with a deep understanding of customer behavior and offer accurate predictions on behavior when making transactions online. The impact is significant: for one major card issuer, implementing a machine learning fraud system demonstrated a 70 percent reduction in genuine transactions declined (also knowns as “false positives”), which helped to maximise revenue and keep customers happy. It is an approach being adopted by the largest payment processor in the United States, TSYS. TSYS wanted to strengthen its position in faster payments using machine learning to provide clients with actionable insights in real-time. What does this mean for merchants, banks and card issuers? Regardless of the size of the business, protecting customers is key, which is why financial institutions are embracing advanced deep machine learning to upgrade their fraud management systems with adaptive behavioral analytics.
The benefits are clear for merchants and financial institutions – machine learning systems mean there is no need to turn off rules on the busiest days of the year and no need to turn away and upset good, loyal customers.
On days like Cyber Monday, it is now possible to identify fraud without blocking genuine customers. You can have the best of both worlds and the power of machine learning is making it possible.Luke Reynolds is the Fraud Director for Featurespace, a U.K.-based fraud-prevention technology provider that grew out of machine-learning research at Cambridge University. Reynolds has nearly two decades of experience in fraud and risk management for the banking sector, most recently with Callcredit Information Group and Lloyds Banking Group.