The Role of Machine Learning in Fraud Prevention

Emily VuittonFraud Prevention2 Comments

Using machine learning in fraud prevention

Fraud losses are a big problem: to the tune of an estimated $6.7 billion in 2016. Unsurprisingly, potential methods of prevention and mitigation flood the market. One such solution in machine learning might be the most promising solution merchant’s have on the horizon.

What is Machine Learning?

Simply put, machine learning uses artificial intelligence that allows software to pinpoint data patterns and trends in order to complete a task. The idea of machine learning is to improve the user’s experience, all with minimal intervention — think of your Netflix account suggesting a similar TV show to the one you’re currently watching.

So when it comes to preventing future fraud with your business, machine learning can be beneficial.


Earlier this week, Practical Ecommerce published the blog post 6 Ways Machine Learning Will Impact Ecommerce, written by Armando Roggio. The article details the potential areas of impact of machine learning on the ecommerce community specifically. In addition to elements like customer service chatbots and market-right pricing, Practical Ecommerce briefly introduces the idea of fraud detection machine learning and prevention.

Fraud Prevention Using Machine Learning

When used as a fraud prevention solution, machine learning seeks out customers that behave ‘unusually’ and provide suspicion scores, rules, or visual anomalies. The output from machine learning is only an indication of the likelihood of fraud. No statistical analysis can definitively determine whether or not a transaction is fraudulent.

Machine learning solutions are classified into two types of ‘learning’: supervised and unsupervised. Supervised machine learning involves taking a random sample and manually classifying the transaction as either fraudulent or non-fraudulent. The manually classified records are the used to build an algorithm to classify new records as fraud or non-fraud.

On the other hand, unsupervised machine learning does not use manually identified records as a starting point for the algorithm. PayPal has used unsupervised machine learning technology to study an individual’s purchase history, spotting patterns and implementing new rules to prevent repeat offenses. As a result, PayPal’s revenue fraud rate is 0.32 percent, far less than half the industry average of 1.47 percent.

Still, PayPal processes roughly $28B in transactions each month. With the revenue fraud rate of 0.32 percent, that means PayPal merchants surrender $89.6 million every month to fraud. LexisNexis data reveals to us that 79 percent of all fraud losses come from friendly fraud and chargeback fraud. Which means that PayPal merchants could recover $707,840 every month with automated chargeback responses. Furthermore, half of these chargebacks could trigger an early chargeback alert that would automatically reroute shipment, stop fulfillment, and/or refund the customer.

In addition to reducing fraudulent transactions, machine learning can drastically reduce costly false positives and manual review time. The IBM Detecting Fraud in Financial Transaction solution cuts down on these false alarms by analyzing the connection between transactions suspected to be fraud and actual fraud. The use of IBM’s solution at a leading bank resulted in a 40 percent decrease in false alarms from the e-payments system.

Other Benefits to Fraud Detection Machine Learning

What are the additional benefits to using machine learning with fraud detection? For starters, machine learning is very adaptive, which is especially handy since the ecommerce industry and corresponding technology is always evolving. Machine learning can learn new trends quickly and pinpoint new ways fraud is being committed online — all in real time. This gives organizations an upper hand in being more proactive in preventing fraud instead of reacting to fraudulent activity. From the data collected, businesses can create specific strategies and help protect their bottom line.

Furthermore, Practical Ecommerce mentions that “the key advantage is that a learning system will be almost unique to its ecommerce retailer.” So, again, machine learning can help businesses create industry specific fraud prevention tactics. And with ever improving fraud detection algorithms, this capability will only get better over time.

In addition, because machine learning can analyze thousands and thousands of data points at a time, it’s a more efficient way to fight against and prevent fraud than using analysis conducted by an employee or contractor.

Cost-Benefit Conundrum

Practical Ecommerce points out the cost-benefit challenge that small and medium size merchants face with fraud prevention solutions. “If your business experiences $1,000 in fraud losses each year and it would cost $3,000 per year to purchase fraud detection software,” Roggio states in the article, “It might make more financial sense to suffer the fraud losses and move on.”

This can be incredibly frustrating to small and mid-size merchants. It might not be financially logical to invest in a pricey fraud prevention solution, but does that mean you need to suffer through unchecked fraud losses? Our answer is firmly, “No way!”

The majority of fraud losses aren’t attributable to true fraud. Instead, between 70-80 percent these losses come as a result of friendly fraud and chargeback fraud. Merchants can recover those losses through automated dispute resolution at a fraction of the cost. Fraud prevention solutions are still necessary to stop large-scale attacks and losses, but to truly minimize your fraud losses you need to address post transaction fraud.

All merchants, small and large, can activate automated dispute resolution at a price that reflects their own transaction ecosystem. Advanced fraud prevention is expensive; fraud management shouldn’t be.

Need additional help with chargeback fees or dealing with chargebacks in general? From lowering the dispute ratio to saving you time and eliminating frustration, see how Chargeback’s automated dispute management platform can help your business today. Click here to learn more and stop potential chargebacks from happening in the first place.

Why Chargeback?

Chargeback offers you a variety of resources to help better manage payment fraud and disputes. In addition to our articles and reason code library, we also provide eBooks, guides, case studies, and events to help you learn the latest trends with disputes and chargebacks and how you can prevent them from happening in the first place.

Furthermore, from lowering the dispute ratio to saving you time and eliminating frustration, see how the Chargeback App’s automated dispute management platform can help your business today. We work hard to offer the best solution in order to help merchants lower their dispute ratio, improve their win rate, and save time that they could have lost from chargeback responses.

Click here to learn more and stop potential chargebacks from happening.

Managing Chargebacks In-House

2 Comments on “The Role of Machine Learning in Fraud Prevention”

  1. Pingback: 5 Ecommerce Trends We're Thankful For | Chargeback

  2. Pingback: Ecommerce Payment Fraud Outlook 2017-2020 | Chargeback

Leave a Reply

Your email address will not be published. Required fields are marked *