As digital interactions across financial services grow exponentially—giving more people than ever access to the global financial system and the cashless economy—financial crime has become a much more prevalent threat.
The annual impact of global fraud exceeds $1 trillion, according to LexisNexis, with every dollar lost to fraud costing $4.23 for U.S. financial services firms. In 2022, financial services businesses have seen a 79% increase in document fraud compared to the previous year. And given the economic environment in 2023, it’s safe to say financial fraud will only get worse.
In such a challenging environment, fintech startups and incumbents don’t just need to defend against today’s most sophisticated methods of fraud. They need to stay ahead of the curve by anticipating new methods of fraud and rooting out bad actors, while meeting anti-money laundering (AML) and know your customer (KYC) regulations and ensuring a superior customer experience.
Enter AI and deep learning models, which can handle a much larger and more complex set of data inputs to identify patterns and anomalies that humans alone would not be able to detect. This not only saves time and resources spent on investigating fraud, but also improves customer experience by reducing the number of legitimate transactions that are mistakenly flagged as fraudulent. By investing in deep learning models and partnering with the right platforms such as NVIDIA and AWS, financial institutions can achieve new levels of accuracy in identifying and preventing fraudulent activities at scale.
The growing cost of fraud
Financial services companies have a fiduciary and regulatory duty to protect customers, but the cost of fighting fraud and complying with the rules on financial crime can be high. FIs spent an estimated $213.9 billion on financial crime compliance in 2021—a number that’s double the amount they invested just two years earlier and keeps growing.
These increases are partly a consequence of the rapid digitization of the industry and the rise of a cashless society. Digital payments are expected to reach $9.5 trillion this year and then grow by 15% year over year for the next five years. The sector is also becoming more complex as a result of new business models, including buy now, pay later (BNPL), that open up new risk categories in the market and provide more opportunities for fraudsters. In the current economic climate, marked by a cost of living crisis, the sector is likely to see renewed acceleration of risk.
As a result, the fintech industry is under pressure to tackle fraud and proactively address the risk that comes with increased digitization. According to a recent survey by ComplyAdvantage, around 40% of FIs are prioritizing improvements to their fraud detection methods within their compliance programs, which suggests a lack of confidence in their current technology’s ability to effectively detect and prevent fraud. Fintechs and FIs need to find new solutions to manage their compliance and fraud prevention efforts, all while keeping pace with the breakneck speed of advancements in financial crime.
“It’s not just the dollar that’s lost in the actual fraudulent action. It creates a domino effect that results in costs for the financial system and us all,” explains Kevin Levitt, Head of Global Industry Business Development for the Financial Services industry at NVIDIA. “From regulatory penalties to damaged culture and morale, customer loyalty takes a hit too, creating a compounding negative effect on the perception of safety in the financial system.
Fighting fraud with deep learning
When FIs migrate into deep learning, they unlock significant growth in the types of anti-fraud protections they can quickly implement.
Traditional rule-based systems can struggle to identify new patterns of fraud, since in many cases, they identify fraudulent transactions in comparison to historical transactions. But deep learning doesn’t require a set of predefined rules and constantly learns on its own to better recognize anomalies. This way, it can detect patterns even if the fraud is just slightly different from what FIs have seen in the past. Given the ever-evolving methods of financial fraud, it’s crucial to adopt a proactive approach that identifies novel forms of fraud and prevent them from spreading.
AI increases success in identifying fraudulent practices and bad actors by increasing the types of data used in the fraud prediction models, including computer vision for ID document analysis, voice recognition for login, and natural language processing to monitor evolving actions of individuals and organizations. These models enable fintechs to detect fraud in real-time while they onboard new customers. HyperVerge, a leader in AI-based identity verification and business verification, leverages the AWS AI/ML platform powered by NVIDIA’s accelerated computing platform to deploy voice recognition alongside 50+ other deep learning models, including forgery checks, document classification, and information extraction.
“We are at the forefront of the battle against financial fraud and continuously developing new models to keep up with emerging frauds, such as deepfake generation,” says Vignesh Krishnakumar, the Chief Technology Officer at HyperVerge. “This approach enables us to ensure that all fraudulent attempts are thwarted, while minimizing the turnaround time for a seamless customer experience.”
The speed of detection plays a key role in fighting fraud. Veriff, which partners with AWS and NVIDIA to build its state-of-the-art identity verification solution, uses deep learning models to quickly analyze more than 10,000 variations of government-issued IDs from over 190 countries. Trained using Amazon SageMaker or an AWS EC2 instance with NVIDIA GPUs, Veriff’s algorithms can update their models to address new fraud techniques in as little as a few hours.
“One of the critical models in our pipeline — and the first model ever trained by Veriff — is designed to detect document specimens,” says Jaanus Kivistik, the Chief Technology Officer of Veriff. “It was trained on hundreds of thousands of images and is currently on its 200th iteration, with periodic retraining involving both AWS and NVIDIA.”
Raising anti-money laundering alarms
Global regulatory requirements around anti-money laundering are the key drivers for improved identity verification. However, these AML regulations mean that banks and regulators must constantly monitor new data that could signal fraudulent activity — even among those organizations that were previously deemed good actors.
Yet again, AI-powered solutions can help FIs meet these important standards without limiting the customer experience. For example, if an investigative report discovered that a particular organization was involved in money laundering on behalf of criminal groups, AI analysis could quickly digest and structure this new information, apply it to existing AML rules, and shut down access to the FI’s network for entities engaged in nefarious or suspicious activities.
The financial system came under a lot of pressure after the sanctions following the escalation of the Russo-Ukrainian War in 2022. Many fintechs who weren’t using sophisticated techniques were really struggling to meet their obligations, as the list of people exposed to the sanctions changed rapidly. Against this backdrop is a regulatory landscape that is taking a tougher stance on financial crimes — particularly sanctions enforcement.
In this context, ComplyAdvantage, a NVIDIA and AWS partner that specializes in fraud and AML risk detection, uses AI to identify people and organizations targeted by government action. As a result, financial institutions using ComplyAdvantage can quickly pinpoint the correct individuals that are the subject of sanctions and update their screening services accordingly.
“Financial services companies that support sanctions enforcement are now able to update newly named individuals or organizations almost instantly, eliminating a time lag that, only decades ago, was easily exploited,” explains Vatsa Narasimha, CEO of ComplyAdvantage.
While it’s important to root out bad actors, it’s also crucial for fintechs to make accessing the financial system easier for consumers through streamlined processes and transparent customer service. Compliance to AML regulations shouldn’t come at the expense of customer experience.
“When you use deep learning techniques for fraud detection, you actually reduce your false positives, which means you’re flagging fewer good actors as potential fraudsters,” says Levitt. “This improves the customer experience for both customers, and for merchants looking to see revenue in their books as quickly as possible.”
Shaping the future of payments
Financial fraud is only going to get more sophisticated over time, predicts Sam Edge, Global Head of Fintech for Startups at AWS, and rules-based models are going to be insufficient to cope in this kind of evolving environment.
Deep learning will play a crucial role to address those challenges, particularly through the use of Generative Adversarial Networks (GANs). GANs create realistic synthetic data that can augment existing datasets, which can then be used to train models on potential future risks the banks haven’t seen before.
“If banks can predict the future to a certain extent and leverage AI to build synthetic data off the back of that to train models, they can reduce the cost of fraud in the long run,” says Edge.
Another emerging trend in fraud prevention involves companies acting in concert together to fight fraud. Federated learning, a way to train AI models while keeping the underlying data secure, enables banks to share data across organizations and geographies without revealing any sensitive information. Ultimately, this makes it easier for FIs to pool their resources and improve the industry’s fraud detection models.
Such a technique carries great benefits for the security of the financial system, as fraud is often committed not against a single bank or financial institution, but by leveraging multiple institutions to launder money and perpetrate fraud. With federated learning, cross-industry and cross-geography consortiums have a greater incentive to collaborate, enabling the financial system to understand relationships between data sets that were previously undiscoverable and combat new methods of fraud as they emerge.
Download the State of AI in Financial Services report to learn more about why deep learning is the future of fraud detection.