There’s a new wave of possibility in online fraud prevention. It’s overturning established best practices, making organizations reassess what were once accepted costs of doing business, and removing the delays that were once an inevitable component of stopping fraud. Online fraud prevention can finally be real-time, consumer-centric, and friction-free.
How is all of this possible? The short answer relies on three crucial things: BIG DATA, machine learning, and human expertise. And, while none of it would be possible without advances in technology, it’s important to bear in mind that approach and mindset are equally essential to using the technology effectively.
In other words, it’s not only the technology you have but what you do with it that matters. And it does matter, because the benefit is significant: Full automation in an industry that has long relied on manual reviews that delay, confuse, and frustrate customers, leading to loss of sales as well as high fraud losses and fraud-prevention costs. New-generation fraud prevention can increase sales and improve customer experience, as well as stop fraud.
What Is “Big Data”?
“Big data” is a term that is notoriously hard to pin down. Back in 2014, Forbes actually published an article entitled “12 Big Data Definitions: What’s Yours?” and the situation hasn’t become any simpler since then.
One thing is clear, though. The term relates to the fact that the data available in recent years has increased, and continues to increase, exponentially. Industry analyst Doug Laney’s famous ‘three Vs’ still help to explain what makes big data so big: volume, meaning the sheer amount of data available; velocity, meaning the unprecedented speed at which data now flows in and the need to deal with it quickly; and variety, meaning the diverse forms and formats in which the data is available.
All of this data is potentially valuable when you want to prevent fraud. The more you know about a customer, the more accurate you can be about whether a transaction is genuine or not. But dealing with it is clearly a challenge. For years, fraud prevention relied on manual reviews, meaning human fraud analysts reviewing transactions and deciding whether to accept orders. With big data, it’s obvious that manual reviewing won’t scale.
Automation is essential so that companies can use big data to its fullest potential. That means an end to intrusive requests for further information that interrupt the checkout flow, an end to delays in receiving confirmation (during which many buyers seek their product elsewhere), and an end to fraud prevention causing delays in fulfillment.
Automation also enables real-time fraud decisions. If 40% of people abandon a Web site that takes more than three seconds to load, as some sources have reported, then retailers can’t afford anything that adds delay on their site.
But traditional fraud prevention has delay built in, in one form or another. Rule engines and fraud scores, which have been the norm for years, contribute to the problem, since retailers must continually adjust their parameters for risk, decide how to use the scores in practice, and review transactions with a “maybe” score.
Now, retailers that approach big data as an opportunity rather than a problem can receive instant approve/decline decisions for every single transaction. Some vendors even back those decisions with a chargeback guarantee.
In its 2015 report “Stop Billions in Fraud Losses with Machine Learning,” Forrester Research found that “legacy fraud-management mechanisms fail in today’s economy.” They are less good at catching fraud because they don’t adapt fast to changing circumstances or new fraudster techniques. And they cause high false positives, cases when legitimate customers are wrongly rejected as fraudulent.
Instead, Forrester advocates a switch to machine learning, which leverages big data to become more and more accurate over time, adapts and improves automatically, and enables accuracy rather than risk aversion. The result is that false positives, and the lost sales and lost customers they represent, almost disappear, and fraud is blocked far more effectively.
The immensity of big data, which can feel overwhelming for humans handling transaction reviews, is actually an advantage for machines. With more data, they become more and more accurate. That’s because they use past data to predict what will happen with present cases, finding the relevant patterns and attributes.
Valuable though machine learning is, it won’t reach its full potential without guidance from human experts. While machines are great at finding patterns in a wealth of data, humans can find new techniques or trends in a very small sample, something that they can teach the machine to look for.
Moreover, though machines can learn, they can’t think. They don’t have the creative capacity that is an essential element of human research. They can tell that something is so, but not why, which limits their ability to make further related deductions.
It’s important to remember that behind every transaction is a person. Experienced analysts will be able to tell if the person making a particular transaction is really the person they’re pretending to be. Once they’re tested and confirmed, the elements that make up this insight can be taught to a machine, making it ever more accurate.
An Expert System
The right combination of big data, machine learning, and human expertise makes what we at Forter call an “expert system,” a system that leverages the potential of big data and the power of machine learning, guided by the understanding, experience, and skill of highly trained analysts.
Approaching fraud prevention with the right mindset and the right tools makes all the difference. Retailers can transform fraud prevention so that it is far better at stopping fraud, and starts enabling sales instead of blocking them.