Fraud is a challenging problem for ecommerce merchants of every stripe. In fact, for the last two decades, ecommerce fraud has grown without pause. It’s critical, therefore, for online retailers to have a fraud management program that’s rigorous and efficient. And many are now coming to the realization that their current way of managing fraud—through rule-based programming—is failing to protect their business.
The problem with the old school fraud management
The traditional approach to fraud management is based on using binary statements, or rules, to determine if a transaction is authentic or fraudulent. This rigidity leads to inaccuracies in fraud detection, causing either false positives or false negatives. An overly conservative risk policy, where the rules are highly sensitive to locking out potential fraud, leads to a high number of false positives. Essentially, benign transactions are wrongly marked as fraudulent and are consequently declined. Not only does this result in missed revenue opportunities, it also frustrates customers and lowers retention. Furthermore, what’s critical to note is that ecommerce businesses decline transactions due to false positives more than any other industry.
Striking a delicate balance
How much damage do false positives really cause? Research indicates that approximately 15% of all CNP transactions are false positives, resulting in USD118 billion of lost revenue a year. Even worse, more than a third of customers never return to a merchant that has wrongly declined their payment.
But perhaps what hurts the most is the fact that false positives alienate more genuine customers than actual fraud does. In other words, the fear of fraud costs more than fraud itself. On the other hand, false negatives—where truly fraudulent transactions manage to slip through the cracks—results in revenue loss. Ecommerce merchants that sell digital-only goods are at highest risk, since they can’t take afford to spend a lot of time on weeding out fraudulent transactions individually. And once sold, digital goods can’t be clawed back.
A need for sustainability and scale
In a rule-based fraud management program, suspicious transactions—where the verdicts are ambiguous—are directed to a human for manual review. While manual reviews aren’t limited to such programs, they do occur far more frequently as a result of the traditional rule-based approach than when more advanced fraud technologies are applied.
Manual review is a bottleneck in both the customer journey and the merchant’s workflow, says Aldrin Mangalabal, Fraud Manager at Payvision & Acapture. It not only slows down and adds friction to the entire purchasing process, it also drives up expenses in labor and leads to a higher amount of inaccuracies in wrongly-flagged transactions.
Yet, ecommerce retailers dedicate substantially more resources to manual reviews than merchants in other industries do. On average, more than 50% of ecommerce transactions are sent for manual review. In 2016, UK ecommerce businesses reported that manual review was their biggest challenge in fraud management. As many as 46% of respondents said that they were spending too much time manually reviewing orders and that their top three priorities all related to performing fewer manual reviews.
Machine learning: the future of ecommerce fraud protection
Today’s fraud experts regard machine learning (ML) as the most powerful fraud-fighting technology available. As ecommerce merchants look to elevate their defense against fraud attacks, leveraging ML can deliver results not previously possible with traditional methods. Essentially, a ML solution is based on self-learning models that are able to determine whether or not a transaction is fraudulent. Instead of the traditional rules-based system, a ML model analyses vast streams of data and assesses multiple signals to calculate a transaction’s risk score.
A ML model can also learn about behavior patterns, taking into account the actual outcomes and feedback provided by analysts and using that data to continuously fine-tune its predictive abilities. Simply put, machine learning eliminates the use of pre-programmed rules in favor of a much more flexible and autonomous approach – one that augments human decisions with improved precision.
For instance, ML is channel-agnostic, which means that merchants who want to grow their omnichannel operations can do so without having to create new rules or add more resources to accommodate transactions from different devices. And as their customer base grows, ML does the heavy lifting for the merchant’s fraud team, allowing them to scale their fraud protection abilities effortlessly, and with limitless flexibility. Another benefit of using ML is that companies can process enormous amounts of data without human bias or error, while being assured that the algorithm will refine and refresh itself for higher accuracy. For instance, after using Sift Science’s ML solution, OpenTable’s digital gift cards program reported a 200% improvement in fraud detection accuracy.
Want to learn more about how machine learning can fight fraud and optimize conversion? Then you’ll be interested in the free report from Acapture and Sift Science, Staying a Step Ahead of the Fraud Footprint, an essential overview of payments fraud that every ecommerce operator should know.