Why Banks Struggle to Deploy Agentic AI Beyond Pilot Projects

GlobalData convened a panel of industry leaders to examine why agentic AI remains stuck in pilot mode at most financial services firms. The session featured Jeff Veis, Chief Marketing Officer at Impetus Technologies; Deepak Khosla, Chief Growth Officer and Head of AI Business at Impetus Technologies; and Stephen Walker, Retail Banking Analyst at GlobalData.

Panellists moved past buzzwords to identify concrete obstacles preventing production deployment. Among the challenges discussed: insufficient data readiness, lack of unified enterprise context, inadequate governance frameworks, safety concerns, and evolving regulatory expectations.

Banking applications raise the bar for autonomous systems, according to Khosla. Unlike simpler document summarisation tasks, agents in this sector influence decisions around credit, fraud detection, payments, customer service, regulatory reporting, and financial advice.

An agent in banking is not just summarizing a document. It could influence and impact credit, fraud, payments, customer treatment, reporting, or advice. The bar for production is therefore much higher in the banking and financial services sector.

He stressed that deploying agentic AI at scale requires robust AI-ready data infrastructure. Agents must be anchored in proprietary business logic, operational realities, customer interaction history, and compliance policies unique to each institution.

Khosla outlined Impetus Technologies' approach, which prioritises enterprise context quality over model sophistication. Fragmented, outdated, or ungoverned data undermines agent reliability regardless of underlying algorithms.

Our approach starts with the belief that agentic AI success depends on the quality of enterprise context available to agents. If that context is fragmented, stale or poorly governed, they will fail in production.

The firm recommends building semantic layers, ontologies, and knowledge graphs that provide agents with structured, trustworthy information to reason over. This foundation enables institutions to target use cases where return on investment is clear and risk is manageable.

Participants concluded that banks should concentrate on high-impact applications where ROI justifies the engineering effort. Success hinges on constructing trusted data architectures before scaling autonomous systems.

The panel advised institutions to engineer enterprise context first, then identify strategic use cases where agentic AI can deliver measurable value under controlled risk conditions. This measured approach contrasts with experimental deployments that lack production-grade data foundations.