AI Enters Bank Payment Operations
Finastra has unveiled a new AI-powered tool designed to accelerate bank payment operations. The move signals the continuing push to automate mission-critical banking functions, which could eventually lead to faster and more efficient post-trade settlement and reconciliation across the market.
Finastra's new "OperatorAssist" tool is designed to address payment exceptions, which are transactions that fail straight-through processing and require manual intervention. The AI-powered tool automates the analysis of these errors, suggests repairs, and guides operations staff through the resolution, aiming to cut manual investigation time by 20-30%. The system is built on a cloud-native architecture and is designed to be "ISO 20022-native." This is critical as the ISO 20022 messaging standard provides significantly richer and better-structured data, which in turn fuels more effective AI and machine learning applications for anomaly detection and repair automation. This move reflects a broader industry trend of embedding AI into core payment infrastructures to manage the growing complexity of modern payment systems. Competitors are also advancing in this area. JPMorgan Chase, for instance, has over 175 AI use cases in production within its investment bank, specifically using AI and machine learning to reduce friction and human intervention in its payments business, where transaction volumes have surged by over 50% in recent years. Goldman Sachs is deploying AI agents, specifically using Anthropic's Claude model, to handle exceptions in trade accounting and client onboarding. This approach uses AI to augment human analysts by managing repetitive data extraction and comparison, allowing them to focus on resolving complex exceptions rather than routine tasks. Goldman is also exploring agentic AI for trade surveillance to identify complex anomalies beyond simple rule-based detection. While AI tools like OperatorAssist address operational efficiency at the application layer, the push for modernization extends to the core infrastructure to reduce latency. For high-frequency trading and time-sensitive financial messaging, firms are adopting kernel bypass techniques, using libraries like OpenOnload or DPDK to allow applications to access network hardware directly. This approach avoids the standard kernel networking stack, eliminating latency-inducing overhead from context switches and system calls. At the hardware level, Field-Programmable Gate Arrays (FPGAs) are increasingly used to accelerate specific, computationally intensive tasks in the trade lifecycle. Unlike general-purpose CPUs, FPGAs can be programmed for specific functions like market data processing or order execution, achieving significantly lower latency—sometimes reducing processing time by a factor of ten. This makes them a key component in building ultra-low latency trading infrastructure.
