Financial firms need governance before agentic AI can scale

Financial institutions are running into an operational bottleneck as core workflows depend on fragmented, unstructured data, nuanced human judgment, and strict regulatory controls. Governance gaps can raise risk in document-heavy processes such as AML and KYC, where heterogeneous data complicates traceability and error management. The CFA Institute reported that 90% of enterprise data is unstructured, while research cited by ProSight Financial Association put non-compliance costs at an average of $14.82 million annually, or 2.71 times the cost of maintaining compliance infrastructure.

Conversations with Reindeer co-founder and co-CEO Yoav Naveh and JPMorganChase executive Ajay Swamy emphasized that agentic systems require redesigned workflows before autonomous execution. Institutions need to digitize processes as they operate today, map every system and data source, identify judgment checkpoints, and treat exceptions as a first-order design requirement rather than an afterthought.

Governance must also be embedded at the decision level. Swamy argued that financial institutions must be able to trace every decision through data inputs, transformations, policy interpretation, and human review. Naveh stressed that mature agents should recognize uncertainty, ask specific questions of subject-matter experts, and capture feedback so exceptions improve future performance.

Sustainable deployment depends on a long-term operating model, not isolated prototypes. Agent ownership should cover lifecycle management, versioning, testing, approvals, retirement, build-versus-buy decisions, and controls that prevent agent drift as policies, regulations, and workflows change.