How data and machine learning work together for added fraud protection

There’s no question that the pandemic may have changed the digital landscape forever, but those changes weren’t isolated to commerce. Instances of fraud rose¹ sharply and data breaches ran rampant, forcing everyone from small online mom-and-pop shops to Fortune 500 companies to rethink their approach to fraud protection.

“Unfortunately, with all of us living our lives online more than ever, scammers have exploited people’s concerns about health, wellbeing, jobs and finances to commit fraud,” says Tony Neate, CEO of Get Safe Online, the UK's national internet security awareness initiative. As a result, there’s no boundaries when it comes to online fraud. It doesn’t matter if you’re an individual, small business or global enterprise – everyone could be at risk.

But even with our online worlds expanding, many companies are under prepared to tackle fraud and secure their data. According to a report by the Ponemon Institute, sponsored by PayPal², 61 per cent of surveyed respondents say their organisations don’t have the right technologies to mitigate online financial fraud, while less than half of respondents say their organisations have the necessary in-house expertise to prevent and contain online fraud.

Machine learning on the frontlines of fighting fraud

Rules-based systems can be helpful in detecting scenarios that occur frequently, but these systems also deal in absolutes—which means they may not be able to handle the complexity associated with user behaviour. Fraudsters know this and create patterns to imitate it. As a subset of artificial intelligence, machine learning helps us create algorithms to process large datasets with multiple variables to find correlations at lightning-fast speeds. So not only can fraud protection happen in real time, but machine learning can detect patterns as they start to form.

Additionally, these models can be trained using thousands of transactions to help identify future bad behaviour and retrain it using the most up-to-date data available. While many companies worry about implementing a fraud protection solution that might create friction for customers, machine learning may be a viable solution that works with a company’s existing data and reduces operational costs thanks to its automation capabilities. 

While optimising the back end of your fraud prevention solution is important, minimising friction could be the key to success on the front end. It’s important to strike a balance between barring fraudulent activity and seamlessly optimising the customer experience. Get that balance wrong and “you could have a great customer that you're impacting with negative friction that will simply never come back,” says Arthi Rajan Makhija, SVP, Head of Global Fraud Risk, Digital Identity and Platform-as-a-Service at PayPal.

Where data comes into the equation

You could have the most sophisticated AI in the world, but it may not support your fraud protection efforts without the right data. The quality and type of data needed to feed machine learning models spans a wide range of information, including data points such as email address checks and identity scores, session analysis and data collected during enrolment. Different data sets can help detect different types of fraud.

For example, signup fraud is one of the fastest-growing types of fraud⁷, occurring when scammers use stolen identities or create synthetic ones to open a new financial account. This kind of fraud can be challenging to detect, as there’s no historical data to use to create a complete picture of the customer. However, a collection of third-party information helps machine learning algorithms spot scammers by picking up on things like differences between a user’s real and stated location. 

Login fraud—also known as Account Takeover (or ATO)⁸—is another fast-growing fraud type wherein a fraudster hijacks an existing customer account. This approach may grant them access to skim funds, place fraudulent orders or steal and sell personal information. Using data based on the fraudster’s device, email, IP, phone, transaction and behavioural user information, machine learning can determine a fraudster’s true location, and look for high-risk activity like high login attempts or strange transactions.

The right kind of data can also help detect and stop phishing scams, payment fraud, chargeback fraud and more.

Taking the first step into the future of fraud prevention

Ideally, a company would implement a strong fraud protection solution from the moment they interact with a customer, instead of at the point where money is exchanged, says Makhija. But as he points out, “You don't have to do it all yourself.”

Turning to partners who can supply this technology helps simplify the process, leaving companies to focus on their strengths. “Payment processors and fraud providers are in the best position to manage it,” says Makhija. “They have tons of historical information, behavioural information, they understand what types of users typically might transact with merchants of a certain size in a certain vertical and they can bring to bear all that intelligence in securing the ship.”

Fraudsters are seemingly getting increasingly sophisticated and creative. “This is a perpetual cat and mouse game,” says Makhija, “and it's important to make the right investments.”

The content of this article is provided for informational purposes only. You should always obtain independent business, tax, financial, and legal advice before making any business decision.