How AI is transforming fraud prevention in ecommerce

• Updated on: 2023-06-17
• Read original article here

Online payment fraud is continually on the rise. A recent study from Juniper Research found that cumulative merchant losses due to online payment fraud will exceed $343 billion globally by 2027.

Traditional fraud detection methods, often based on human-created rules that determined what would trigger a transaction decline, are giving way to more efficient, AI-based fraud detection. Rule-based fraud detection relies on policies that must prospectively predict impermissible customer behavior. This is cumbersome, inflexible and frequently inaccurate.

Fraud detection AI, on the other hand, is most often based on unsupervised learning models, wherein large data pools from multiple vendors and millions of transactions are analyzed by an algorithm. The algorithm isn’t taught what to look for ahead of time; rather the system finds patterns based on behavioral patterns in the data. AI adds flexibility to fraud prevention and can spot anomalies and suspicious behavior without using pre-established rules. AI can also provide decisions instantly.

In this way, third-party fraud detection technologies are also enabling more merchants to compete with massive marketplaces like Amazon and Alibaba. Fraud detection technologies aggregate data from thousands of merchants and millions of transactions, putting everyone on more even footing with giant marketplaces, both in terms of fraud detection and seamlessness of checkout experience.

AI-based fraud detection systems can adapt and make decisions that are increasingly nuanced as new behavior patterns emerge. For example, in the early days of the pandemic lockdown, people who had never purchased home improvement items or tools were suddenly making high-dollar purchases in those categories. eCommerce merchants had to adjust to avoid falsely declining purchases like these that would have appeared fraudulent prior to the pandemic. Fortunately, AI can adapt to changing market conditions like these in near real time.

Expedited shipping is another good example. This shipping method tends to be a red flag in fraud detection since it minimizes the amount of time a merchant has to cancel an order. But expedited shipping became much more common during the pandemic, and the practice has become increasingly safe over time. According to Riskified data, orders placed with expedited shipping increased 140% from January to December of 2020, while fraud levels decreased by 45% over the same period.

Suspicious payment activity can be especially hard to detect if it is perpetrated by historically legitimate customers. “Friendly fraud” is a common example, and merchants are increasingly relying on AI to tackle situations where a customer disputes a charge with their credit card company to avoid paying for something they’ve already purchased from a physical goods retailer.

In these instances, the customer will claim an item wasn’t received by filing an “item not received” chargeback with their bank or credit card company. Some fraudsters even engage in large-scale chargebacks, then sell items on the black market. This costs retailers millions of dollars each year and, if it occurred in a physical store, it would be classed as shoplifting.

There is also a rapidly growing customer trend in the form of policy abuse, which occurs when regular, paying customers break a retailer’s terms and conditions — usually with the motive of saving or making money. There are multiple types of policy abuse: One of the most common is connected to refunds and returns. For example, a customer may contact a retailer to falsely report a missing item, triggering a refund or duplicate to be sent. Similarly, a customer might post a return to the retailer using an empty box (while keeping the original product) or send back used or worn items which is commonly referred to as ‘wardrobing’.

Policy abuse is not the same as traditional fraud but it has similar consequences for the retailer in terms of its potential for financial loss — a fact that can sometimes go unnoticed by the retailers involved. In these situations, AI can spot sophisticated trends and patterns in the purchasing process to allow retailers to take action.

Additionally, “chargeback dispute services” use AI to gather data such as IP addresses, device fingerprinting and behavioral analytics, then cross-reference this across past orders in the merchant networks. If the customer claims an order was fraudulent and not placed by them, the system can verify that it was placed using the same IP address and device where the shopper has placed orders in the past. This helps merchants decide how to prioritize disputes and tackle policy abuse from the greatest offenders. These services also automate the dispute process for merchants to make it scalable and more efficient.

As fraud tactics become more sophisticated, so too are fraud detection methods, which will soon go beyond purchasing patterns to analyze biometric aspects of ecommerce, such as “voiceprint” or the angle at which a mobile phone is held. These advancements will become increasingly necessary to protect customer accounts from fraud.

T.R. Newcomb is VP of strategy at Riskified.