Going Deep: Automated Deep Behavioral Networks Emerge As Card And Payment Fraud Defense

For financial institutions (FIs) and banks, battling card and payments fraud demands real-time monitoring – adjusting risk scores as consumers’ behaviors change. In an interview with Karen Webster, Dave Excell, founder and president of Featurespace, said that artificial intelligence (AI)-driven risk modeling can adapt to shifts in spending – and can help FIs spot and prevent fraud and criminal activity.

The conversation came as the company announced the launch of the next generation of its modeling tool, Automated Deep Behavioral Networks. Featurespace has said the analytics capabilities help banks protect customers from scams, account takeovers, and card and payments fraud.

At a high level, the Automated Deep Behavioral Networks significantly reduce the manual feature engineering work traditionally done by data scientists. As Excell told Webster, the technology enables features to be self-discovered and continually refined, which for financial services firms reduces a complex human task in the process of model building.

In terms of the efficiencies gleaned from the more robust individual profiles, a comparison against the fraud detection rate of Featurespace’s first-generation market-leading models revealed an uplift of 38 percent.

Setting Up for Success

Excell said that even with powerful models, it’s critical to have an enterprise-wide view of the customer to help the FI quickly enact its arsenals of weaponry against payments fraud. “If you look at the complexity of integrating” advanced technologies, he said, where legacy systems have taken shape over decades, it has become challenging for FIs to understand data flows, and how to use that data to gain actionable insight from that information.

Time is of the essence, Excell stated, as fraudsters are becoming more organized and are enlisting machine learning technologies in their own efforts. Now, for FIs, it’s a case of fighting fire with (advanced tech) fire. Start at the beginning, then – and be mindful of that old phrase: garbage in, garbage out. And, of course, connecting the dots is critical.

“If you can’t match a confirmed fraud case to the original transaction request and understand how it was assessed by the fraud strategy, it’s challenging to measure current effectiveness or have the confidence to make changes,” Excell explained.

Featurespace is focused on helping customers understand the necessary data requirements, including utilization of the international financial industry message standard ISO 20022. A deep behavioral network automatically discovers the optimal data elements, and relationships between those data elements, to determine the risk for any given transaction, said Excell. “Importantly, Automated Deep Behavioral Networks is able to encapsulate the importance of when transactions take place, not just the order in which they occur – a first for deep learning algorithms.”

“It’s making the machines smarter, and it enables them to continue adapting those features over time as customer behavior shifts,” he said. “And then as the fraud shifts, it can start to discover what is the right thing to ‘learn’ about you to determine whether your transactions are genuine or not.” The advanced tech-driven modeling approach eliminates the constraints tied to human expertise.

Such a flexible construct is able to adapt to, and score, the behavioral changes that are seen in the midst of a digital shift, as with the pandemic, for example. That flexibility is of particular value when FIs operate in different geographies (or facilitate high-value transactions), where consumer behavior and payment activity can vary widely or subtly depending on who is transacting, and when and where.

In terms of analytics, the Automated Deep Behavioral Networks extracts what Excell termed a “new, time-based dimension of optimization” to determine whether to allow or disallow a transaction. With improved risk scoring, the networks don’t have to allow more fraud to improve the customer experience, as false positives are reduced. The automated, continuous learning inherent in the analytics also makes anti-fraud efforts more cost-efficient, said Excell, as new data can quickly be introduced, tested and adopted.

As real-time payments gain ground, emerging technologies will create new use cases for how that technology is deployed and used by both consumers and businesses, and Excell said that concerns about fraud shouldn’t slow the pace of innovation. The conventional wisdom may be that faster payments equals faster fraud; one hallmark of real-time payments is that those transactions are irrevocable.

But as Excell told Webster, advanced learning models can actually help introduce an appropriate level of friction into real-time payments – a “cooling-off” period (with the consumer’s confirmation that a transaction is indeed legit) that can further reduce fraud. FIs must get on board with faster payments, he said, or risk being left behind.

As Excell explained, Automated Deep Behavioral Networks enable banks “to go ‘faster’ [with faster payments] knowing that they do have controls in place to be able to do that.”