What 159 banking leaders say about putting AI agents to work in lending, service, and compliance.

Scaling Agentic AI Safely Across Regulated Workflows

Banks are ahead of the broader market on agentic AI, with adoption at 91% in finance and accounting, 89% in customer service, and 83% in risk, compliance, and security. Adoption is now moving closer to the customer in guided processes like lending and service.

As agents move into regulated work like loan applications, every step must be auditable, traceable, and reproducible. Within five years, 46% of banking respondents expect agents to coordinate work across multiple systems and teams, but only 4% expect workflows to run without people. As autonomy expands, keeping consequential decisions accountable and attributable remains the central challenge.

We partnered with KPMG and Banking Dive to survey 858 executives, including 159 banking leaders, on how they scale agentic workflows safely.

Key Takeaways from the Report:

• Keep Decisions Attributable: Record what every agent does, and set explicit points where a person decides.
• Secure the Path to Action: Connect agents to core and legacy systems through a controlled layer, with their own identity and permissions.
• Keep Your Choice of AI: Use multiple models and vendors under one one set of guardrails.
• Prove the Value First: Measure agent-supported work against your baseline before you put agents to work anywhere new.