Machine Learning for Fraud Prevention

Machine Learning aids e-commerce to foil attempts at payment fraud, as they happen.

Long before the pandemic led to an avalanche of online shopping, e-commerce had become a way of life for many Americans, especially Millennials and Gen Zers. In fact, 60% of Millennials bought online in 2019, while 24% Gen Zers strongly prefer to purchase online and 13% through mobile. This has led to variety of online shopping choices, including e-shops, online banking, online insurance and other online services.

As Hil Davis, Co-founder of the online men’s retailer, said, “E-commerce and mobile commerce have dramatically changed the way brands reach customers, making it faster and easier for consumers to make purchases on the fly while avoiding the hassles of going to the store.”

This has led the global e-commerce market to predict a US$ 4.9 trillion growth by 2021, increasing 20% through 2022, to reach $5.8 trillion.

Even as e-commerce increased, so has online fraud, pushing retailers and banking institutions to seek new solutions to prevent, detect and eliminate fraud. A study showed Visa and Mastercard lost $750 million to credit card fraud. In fact, some of the commonest financial fraud include identity theft, phishing, which means collecting personal information through emails or illegitimate websites, and pharming, which is directing customers to fraudulent web sites.

It has also been found that only about 1/4th of customers understand how fraudsters operate. When fraudsters steal cardholder information on payment pages or through the dark web, they impersonate the cardholder and purchase items online.

This includes all components of the login process, from password entry to phone verification. The online seller, believing the purchase is valid, processes payment and ships goods to the fraudster. The cardholder notices the charges and contacts the bank.

The online seller is slapped with a chargeback plus fees. Studies show that only 20% of customers understand that it is eventually the retailers who pay for fraud.

The Association of Financial Professionals (AFP)  in its 2020 Payments Fraud & Control Survey, states that 81% of organizations were targets of payment fraud in 2019.

This is the second-highest percentage of reported fraud attacks/attempts since 2009. UK-based Juniper Research says payment fraud is already a billion-dollar business, and it is growing.

AFP President and CEO, Jim Kaitz, said, “Payments fraud is a persistent problem that is only getting worse despite repeated warnings and educational outreach.”

Jessica Lupovici, Managing Director, J.P. Morgan, said, “Businesses stand to suffer reputational risk, which can be severe, expensive and require significant clean-up efforts."

Aiming to halt this rising trend in payment fraud, almost 60% of retailers have ensured a fraud policy in place in their companies. 

The most plausible solution appeared to be machine learning (ML), which is an area of artificial intelligence (AI).  ML is the science of making computers learn and act like human beings, while autonomously improving their learning capabilities.

And, where ML most conspicuously shows its machine learning superiority, is in critically-needed speed. ML is akin to several teams of analysts running hundreds of thousands of queries, and comparing outcomes to choose the best result - in real-time, taking only milliseconds.

Anti-fraud systems generally employ two types of ML - unsupervised and supervised machine learning. Supervised learning focuses on training an algorithm to use labeled historical data.

With labeled datasets already in existence, the goal of training is to make the system predict these variables in future data. Unsupervised learning models process unlabeled data and classify them into clusters, detecting hidden relations between the variables. These two types can be either used separately or together, in creating more sophisticated anomaly detection algorithms.

Armed with this learning, the algorithms of ML engaging in risk analytics,  differentiate fraudulent transactions from legitimate operations much faster than any human analyst could, spotting even the stealthiest, and seemingly unrelated, patterns that evade human attention. What is more, these algorithms detect fraudulent transactions with significantly high accuracy.  

The faith in ML is largely due to its effectiveness, as 80% of fraud specialists endorse the technology as helping to reduce payment fraud and attempts at payment fraud. Furthermore, 63.6% of financial institutions which engage ML in their business transactions, also believe it helps prevent fraud.

Thus, the whole ML process is about the final product. When a ML model is provided with information about a new transaction, it generates a recommendation on whether it is an attempted fraud or not.

All customer transactions are sent to the model, which decides what transactions are approved, what are blocked because of attempted fraud, and what are marked for manual review.

By employing ML, a popular Latin American online travel agent, Almundo.com, reduced fraud, chargebacks and manual reviews by 70%.

Anti Fraud systems, therefore, are focused on detecting fraud in real-time, improving data credibility, analyzing user behavior and uncovering hidden correlations.

All of it makes fraud mitigation one of the most sought-after services of AI, as it is able to immediately ensure a return on investment by preventing fraud in real-time.

According to a new study from Juniper Research, network optimization and fraud mitigation solutions will be the most highly sought-after AI-based services over the next 4 years, and global corporate spending on AI solutions is expected to exceed US$ 15 billion by 2024. The estimated total global expenditure on AI solutions in 2020 is US$ 3 million.

Geoffrey Moore, American Management Consultant, said, “Without big data analytics, companies are blind and deaf, wandering out onto the web like deer on a freeway."