BNY Builds Performance Review System for Its ~140 AI Digital Employees

BNY's AI Workforce Gets Formal Performance Reviews

Bank of New York Mellon (BNY) currently runs approximately 140 AI digital employees. Each agent has its own login credentials, an employee ID in the corporate directory, and a supervisor who assigns tasks and reviews its work. According to a Microsoft case study published in May, the bank's journey from concept to a deployable agent requires a 16-step review committee, compliance gates, and model risk assessment processes. BNY processes roughly $2.5 trillion in transactions daily, where a single misrouted payment could freeze a billion-dollar transfer.

These digital employees are performance-tracked through real-time dashboards, with metrics covering throughput, cycle time, missed transactions, and approval outcomes. Take the payment investigation agent as an example: when reviewers repeatedly send back its draft responses, the team coaches and retrains it much like a new hire. According to Rachel Lewis, head of AI enablement in BNY's operations division, who spoke to Microsoft, the earlier rejections stemmed more from reviewers' old habits than from errors in the answers themselves. The agent replaced a previously glitch-prone script bot whose maintenance costs were so high that "humans were effectively doing two jobs."

The payment agent currently reads payee addresses in cross-border payments, confirms country information, and drafts responses, which human managers review before sending. According to PYMNTS, BNY's open payment investigations have dropped by nearly 80%. Meanwhile, broker regulators are pushing to bring agent decisions into audit trails.

Source: https://www.pymnts.com/news/artificial-intelligence/2026/banks-put-ai-agents-through-performance-reviews/