Fighting Fraud with Data Science

Fighting Fraud with Data Science

“Predictions are hard—especially about the future.”


Yogi Berra

Beyond a doubt, eCommerce crime is big business. And CNP fraud accounts for 81 percent of it—equaling billions of dollars in losses annually, according to Javelin Research.

Today’s reality is that fraudster behavior is quickly adapting to the many ways issuers attempt to prevent it, while legitimate customer transactions can look like fraud. Conventional authentication methods are not enough to keep up with, let alone stay ahead of, these trends.

It takes advanced data science—predictive analytics, neural networks, and machine learning—to change the game in reducing fraud risk while maintaining a positive cardholder experience.

Big Data + Deep Expertise

With real-time analytics, every transaction, such as a log-in event or CNP purchase, is examined using contextual data. This analysis can make fine-grained decisions about any transaction’s implied or inherent risk.

For CNP fraud prevention, predictive analytics is nothing new. But effective analytics take an incredible amount of relevant, globally diverse risk and fraud data. We’re talking about—just as an example—hundreds of millions of devices associated with billions of global e-commerce payments. Plus, the known good or bad behavior associated with these devices and transactions—across both issuers and merchants.

Yet, as important as this data is, it’s easy to overlook that the expertise required to wrap data science around it is just as critical. As analytics become more prevalent in fraud prevention schemes, it’s easy to make mistakes and misapply machine learning algorithms. There are plenty of war stories where a model was being biased the wrong way.

This proficiency is critical in knowing how to accurately apply the techniques of data science to understand legitimate and fraudulent behavior in context of the individual cardholder in real time. And while we’re talking about real time, it’s important to understand that some data models are based only on confirmed historical fraud—essentially chasing after it instead of predicting fraud before it happens.

In half of all CNP fraud schemes, the second transaction occurs within 3.6 minutes of the first, and in 15 percent of these cases it’s less than one second. So when it comes to major fraud events, it’s all about quickly recognizing the first fraudulent transaction to avoid the second, the third, and so on.

This can happen only with sophisticated analytics, using neural networks and a system that continues to learn from all purchase transactions—as they happen.

Finding the Right Balance

In the end, when data science is accurately applied to payment fraud prevention, it allows issuers to find the right balance between risk mitigation and cardholder experience. The result is security that doesn’t get in the way of genuine online transactions, minimizes fraud, avoids false declines, and keeps cardholders happy.

Bottom line? More data means more powerful solutions built on predictive analytics. But building the advanced machine learning needed to optimize user experience and drive out fraud depends on how the data is leveraged.

And that’s where our world-class team of data scientists—with a combined hundreds of years of experience in payment fraud prevention—lead the way.

About the author

David Chiu is a product marketing principal at Broadcom, specializing in digital business transformation, security and ecommerce. He has spent the past 20 years consulting with and bringing innovation to Global 1000 brands at leading technology companies, ecommerce platforms, and digital agencies including Publicis and McCann. Today, David works closely with our Layer7 product team in Vancouver, Canada.