Fintechs Deploy AI for Real-Time Fraud
Banks are upgrading to AI-native platforms to combat fraud. LA-based Linker Finance is partnering with Sardine to integrate device, behavioral, and transaction signals for real-time risk management. Similarly, European bank Novobanco modernized its anti-money laundering and fraud defenses using Feedzai's AI platform.
The shift to AI is a direct response to the escalating sophistication of financial crime, with global fraud losses projected to reach trillions of dollars annually. Traditional rule-based systems are too rigid for modern threats like synthetic identity fraud and deepfake-powered social engineering, which now account for a significant portion of attacks. Sardine's technology focuses on behavioral biometrics, analyzing subtle user interactions like mouse movements, typing cadence, and how a user holds their phone. This data, combined with device intelligence that can detect mobile emulators and other fraud tools, creates a unique user profile to spot anomalies that transaction data alone would miss. Feedzai’s platform leverages network-wide intelligence, drawing insights from over $8 trillion in annual payment volume to identify complex fraud patterns. For one Tier 1 bank, this AI-native approach resulted in a 62% increase in fraud detection while simultaneously reducing false positives by 73%, a critical metric for minimizing disruption to legitimate customers. Architecturally, these systems operate in real-time, often making a decision in under 100 milliseconds. This is achieved through multi-agent AI systems where different models assess various risk vectors—like device, behavior, and network—in parallel, a necessity in the era of instant payments where manual review is impossible. For ML engineers, building these systems requires more than just model development; it demands production-grade data engineering and robust MLOps to combat model drift. The key skill lies in feature engineering, creating novel signals from raw data like transaction velocity and behavioral patterns to stay ahead of adaptive adversaries. A key trend is the move toward Explainable AI (XAI) to meet regulatory requirements and the use of federated learning, which allows models to learn from collective data across different banks without exposing sensitive customer information. In Los Angeles, companies like Upstart and Block (which owns Cash App) are actively hiring for machine learning roles focused on risk and fraud modeling, demonstrating the local demand for these specialized skills. Projects in this space often involve implementing graph neural networks to uncover fraud rings and unsupervised learning models to detect entirely new attack vectors.
