How Visa Is Making Payments Safer and Smarter with AI - Ep. 256

What It Covers

Sarah Laszlo, Senior Director of Visa's machine learning platform, explains how Visa leverages AI for fraud prevention, personalized cardholder experiences, and agentic commerce while managing petabyte-scale data in proprietary data centers.

Key Questions Answered

• •Virtual GPU optimization: Visa isolates memory and compute into separate virtual GPU instances on single cards, giving users one-tenth of a GPU instead of full allocation, dramatically improving cluster utilization while maintaining user experience for enterprises with limited cloud access.
• •Code modernization with GenAI: One engineer used GPT-4 to convert 50 legacy jobs from an unsupported programming language to Python in a single quarter, saving Visa $5 million by automating code translation that no internal staff could perform manually.
• •Ray Everywhere strategy: Visa adopted AnyScale's Ray ecosystem for the entire AI pipeline from data conditioning through model training to serving, creating a unified factory approach that accelerates model refresh cycles critical for staying ahead of evolving fraud tactics.
• •Privacy-preserving personalization: Visa creates high-quality consumer embeddings from trillions of transaction observations without exposing raw cardholder data, developing abstract representations that enable better product recommendations than competitors while maintaining strict privacy standards and regulatory compliance.