Payment fraud continues to plague the financial services industry. According to the American Bankers Association, fraud against bank deposit accounts totaled $25.1 billion in 2018. In 2022, eight U.S. Senators sent letters to the CEOs of seven of the largest U.S. banks concerning fraud at one real-time payment firm. With real-time payments growing globally by 41% in 2020, there is an obvious need to modernize fraud prevention as criminals try to exploit the system.
To help combat payment fraud, companies are investing in technology that leverages hybrid cloud architectures and AI / ML. In a hybrid cloud, compute workloads can be spread across on-premise data centers, private clouds, public clouds and even edge locations depending on requirements such as data sovereignty, latency, capacity, cost and more. Advances in AI / ML, allow machines to be trained to recognize patterns across billions or trillions of data points. These relationships are then incorporated into “models” which are built into real-time payment workflows.
One hybrid architectural pattern is for high privacy payments infrastructure to remain on-premise with the public cloud being used for model training. By using the public cloud, firms can parallelize training across a vast number of nodes, only pay for time used and have access to hardware acceleration such as GPUs. To protect privacy or improve data quality, firms can generate synthetic data which is transferred to the cloud and used for training. Trained models are then imported into a firm’s runtime environment where they execute on-premise with local access to privacy data.
For global financial institutions, data sovereignty requirements might dictate another architectural pattern that keeps payment and fraud data in the originating country. With federated learning, a single foundation model is created centrally and distributed to remote sites. These sites then train the model on their local, private data before sending their model, without privacy data, back to the central site. The models are then aggregated into a new global model that can then be sent to the remote sites for more iterative rounds of training. Once the model is fully trained, models run locally without ever having to move privacy data outside a regulatory jurisdiction.
While architectures will vary based on needs, financial institutions will all agree that running these workloads at scale requires a modern platform that leverages the hybrid cloud, improves operational efficiencies, reduces operational risks and helps improve the security posture. With a platform such as Red Hat OpenShift, firms can successfully build, modernize and deploy applications with a consistent experience both on-premise and in the cloud. As business needs evolve, workloads can then be shifted between on-premise servers or those running at Amazon AWS, IBM FS Cloud, Microsoft Azure or Google Cloud. To learn more, visit Red Hat
– Aric Rosenbaum, Chief Technologist, Red Hat
Aric Rosenbaum serves as the Chief Technologist on Red Hat’s Global FSI team, where he helps clients meet their strategic priorities through the use of open source technology. Prior to joining Red Hat, he led large, digital transformation projects at Goldman Sachs’ Investment Management Division and was co-founder/CTO of several FinTechs in equity and FX trading.
 American Bankers Association: 2019 Deposit Account Fraud Summary