NVIDIA announced an AI Blueprint for financial fraud detection at the Money20/20 financial services conference, addressing projected worldwide credit card transaction fraud losses exceeding $403 billion over the next decade. The blueprint leverages accelerated data processing and advanced algorithms, to improve AI's ability to detect and prevent credit card transaction fraud, with up to an estimated 40% improvement in fraud detection accuracy compared to traditional methods.
The reference architecture provides financial institutions with tools to identify subtle patterns and anomalies in transaction data based on user behaviour, improving accuracy while reducing false positives. The blueprint includes reference code, deployment tools, and demonstrates how to build financial fraud detection workflows using the NVIDIA AI Enterprise software platform and NVIDIA accelerated computing.
Currently optimised for credit card transaction fraud, the blueprint could be adapted for new account fraud, account takeover, and money laundering use cases. The system enhances traditional XGBoost machine learning models with NVIDIA CUDA-X Data Science libraries, including graph neural networks (GNNs) to generate embeddings that reduce false positives by analysing relationships across transactions, users, and devices.
NVIDIA RAPIDS enables payment companies to accelerate data processing and transform raw data into powerful features at scale, addressing gaps in traditional data science pipelines that lack compute acceleration for massive data volumes. NVIDIA Dynamo-Triton, formerly NVIDIA Triton Inference Server, boosts real-time inferencing while optimising AI model throughput, latency, and utilisation.
The blueprint is available for customers on Amazon Web Services and Hewlett Packard Enterprise, with Dell Technologies availability coming soon. Service offerings are available through NVIDIA partners including Cloudera, EXL, Infosys, and SHI International.
Financial organisations already leveraging AI include American Express, which began using AI for fraud detection in 2010, monitoring all customer transactions globally in real-time, with fraud decisions generated in milliseconds. European digital bank bunq achieved nearly 100x faster model training speeds with NVIDIA accelerated computing for its AI-powered transaction-monitoring system. BNY became the first major bank to deploy an NVIDIA DGX SuperPOD with DGX H100 systems in March 2024 for fraud detection and other use cases.
The blueprint addresses increasing online and mobile fraud losses reported by large North American financial institutions, by providing accelerated compute solutions for real-time fraud detection. Graph neural networks enable detection of complex fraud networks involving linked accounts and devices, expanding beyond individual transaction analysis.
NVIDIA's fraud detection blueprint positions the company as a comprehensive AI infrastructure provider for financial services, competing with traditional fraud detection solutions through superior accuracy and processing speed. The 40% accuracy improvement and 100x training speed enhancements demonstrated by early adopters provide compelling ROI justification for enterprise adoption. Partnership integrations across major cloud providers and systems integrators enable rapid market penetration.