Agentic AI in banking is generating considerable interest among financial institutions, but a recent expert roundtable hosted by GlobalData concluded that very few have managed to move beyond experimentation into production at scale. The session, which brought together voices from Impetus Technologies and GlobalData’s own research team, pinpointed data readiness, fragmented enterprise context, governance gaps, safety risks and regulatory expectations as the chief obstacles standing between pilot projects and meaningful deployment.
The roundtable, titled ‘Agentic AI in Banking and Financial Services: Beyond the Hype’, 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.
Why the bar is higher in financial services
Khosla was direct about why the sector cannot simply transplant AI approaches from other industries. ‘An agent in banking is not just summarising a document,’ he said. ‘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.’
His point is well-grounded in the regulatory reality of UK and global banking. Autonomous systems that affect credit decisions or fraud flags sit inside a web of conduct obligations, model-risk rules and consumer-duty requirements that do not apply to, say, a retail recommendation engine. The moment an agent acts rather than advises, accountability questions multiply.
Khosla argued that meeting that bar requires two things: strong AI-ready data foundations, and AI agents grounded in proprietary business processes, operational realities, historical interactions and governance policies. ‘Our approach starts with the belief that agentic AI success depends on the quality of enterprise context available to agents,’ he said. ‘If that context is fragmented, stale or poorly governed, they will fail in production.’
Context engineering as the missing piece for agentic AI in banking
The session’s conclusion centred on what the participants called context engineering: building trusted semantic layers, ontologies and knowledge graphs that AI agents can reliably reason over. Without that infrastructure, agents operating in complex financial environments are prone to error, hallucination or decisions that cannot be explained to a regulator.
Impetus Technologies has moved to put that thesis into product form. Impetus has launched its Context Fabric™, described as a semantic intelligence layer for the agentic enterprise that connects ontology, knowledge graphs, business rules, policies, lineage and memory into governed enterprise context. The product is designed to address precisely the fragmentation problem Khosla described: institutions that have accumulated years of data in disconnected systems cannot simply point an agent at that data and expect reliable output.
Impetus is not alone in recognising that governance infrastructure needs to be built alongside AI capability rather than bolted on afterwards. Yahoo Finance reports that Rubrik has unveiled its Semantic AI Governance Engine, or SAGE, described as the data security industry’s first AI governance engine designed to secure and control autonomous agents in real time. The announcement reflects a broader push by technology vendors to build the control layer that regulated industries require before they will commit to production deployment.
On the services side, StockTitan reports that Genpact has launched what it calls Service-as-Agentic-Solutions, with AI agents trained using industry-specific semantic layers to enable what Genpact describes as personalised, precise and controlled business operations. The move suggests that professional services firms see the semantic-layer approach not merely as a technology choice but as a competitive differentiator when pitching to clients in regulated sectors.
Taken together, the product launches from Impetus, Rubrik and Genpact indicate that the vendor community is converging on context and governance as the battleground for enterprise agentic AI, echoing the diagnosis the GlobalData roundtable reached from the client side.
The roundtable’s practical recommendation for financial institutions was to prioritise high-impact use cases where return on investment is measurable and risk can be actively managed, rather than pursuing broad deployment before the underlying data infrastructure is ready. Walker, as the retail banking analyst at GlobalData, contributed the institutional perspective on how regulators and boards are currently viewing autonomous systems in customer-facing and back-office roles.
For institutions weighing where to begin, the session’s implicit message was clear: the question is not whether to invest in agentic AI, but whether the enterprise context those agents will depend on is fit for the purpose. Without that foundation, Khosla’s warning stands: they will fail in production.
