Machine Learning: Separating Fact from Fraud Fiction for Merchants

The data war

While fraud is committed by people, it is also committed using data. Stolen credentials, faked accounts, account takeover – in every case a fraudster uses certain data points or identity markers to successfully make a fraudulent purchase.

Not only do they use data, but they also leave a data trail… and stopping fraudsters is largely about spotting that trail.

Machine learning models are also data-powered. They are essentially designed to combat adversary models built by rogue data scientists who are often at the root of fraud. A kind of battle of the algorithms, if you will. When properly trained and supported, machine learning models can prove extremely valuable and effective in this battle – but a good supply of relevant historical data is critical to that training.

What machine learning models can do – and what they can’t

Merchants need to be clear about what machine learning models can and can’t do and what advantages they can deliver to fraud prevention. From there, it’s how to incorporate machine learning into a strategy that strikes the right balance between minimizing fraud and maximizing conversions.

The core capability of machine learning models is that they quickly and efficiently process and analyze vast amounts of data — turning it into intelligence that can build customer profiles, spot fraud signals and combat emerging fraud threats.

Machine learning models can…

• Act rapidly at the point of sale without the customer noticing any intervention
• Learn quickly from millions of historical transactions and remember behaviors that allow prediction far faster than any human could achieve
• Identify patterns and trends that are too complex to spot through other means
• Work without tiring, making decisions as quickly on the first transaction as on the millionth. Plus, they’re driven solely by historical data, not feelings or emotion.

But… to be effective, machine learning models must be:

• Trained on sufficient, relevant data that usually includes internal and external confirmed fraud intelligence
• Built, trained and optimized by data science experts, with input from experienced fraud analysts
• Continuously monitored for performance levels, and re-trained as new behaviors emerge

Machine learning as part of the fraud prevention toolkit

Professional fraudsters work as hard as we do to predict the industry’s next move and to circumvent the controls or predictive measures we use to defeat them. Every fraud prevention tool or technique attracts the close attention of fraudsters seeking to subvert it. A single tool or layer of fraud prevention is not enough to stop fraud – good fraud prevention requires a solution with multiple dimensions.

For this reason alone, machine learning is not a silver bullet for effective fraud detection and prevention. And, it’s also fair to say that machine learning models working alone may not always give the right answer. There are always nuances that cannot be taken into account by a machine learning model, and they do not offer the same flexibility as a sophisticated rules engine, for instance. During unusual periods of trading, where customer and fraudulent behavior may change rapidly, rules can be more easily flexed to ensure that genuine customers are not mistakenly blocked or fraudsters inadvertently let through.

It’s not just about combining rules and machine learning either. Consortium data, shared intelligence, human expertise, automated decisioning and alerts should all form part of the overall fraud prevention solution if merchants want to stay one step ahead of the fraudster.

Machine learning models are critical, but their role and requirements need to be properly understood. A combination of tools and techniques is the most successful way to minimize false positives while also increasing conversions.

Watch our on-demand webinar: Find a New Ally in the Fight Against Fraud with Machine Learning