BCG Benchmark: Only 8% of Systemic Banks Have Comprehensive AI Agent Monitoring Frameworks
92% of the 26 benchmarked banks are piloting or allowing agentic AI, 73% allow agents in production - yet agents remain under 10% of AI portfolios at 88% of institutions.
Boston Consulting Group published a benchmark of 26 global systemic banks on 13 November showing agentic AI has crossed into production but governance has not kept pace. Some 92% of banks are piloting or permitting agentic AI, and 73% allow agents in production for selected use cases. Yet agents remain under 10% of the AI portfolio at 88% of institutions, and only 8% have comprehensive, standardized agent monitoring and control frameworks. About 54% have imposed agent-specific limits within their risk-appetite frameworks. Banks with mature risk management were 2.5 times more likely to be advanced AI adopters.
The bottleneck the data identifies is not model quality but operational oversight - audit trails, monitoring, and calibrated control of autonomous agents. For technology vendors serving banks, the implication is direct: proposals for AI in lending, document processing or customer operations will increasingly be judged on governance features first, and banks without dedicated AI-GRC teams will lean on external partners to supply that layer.