The Data Foundation in The Agentic AI Era

This whitepaper captures perspectives from data, analytics, and AI leaders across e-commerce, fintech, travel, healthtech, SaaS, and automotive, exploring how enterprises are strengthening data foundations, real-time infrastructure, and observability as they move toward agentic AI.

Agentic AI doesn't create new categories of data problems so much as remove the room to defer solving old ones. Across sectors as different as travel marketplaces and consumer lending, one observation recurred: the volume of data an organisation holds is rarely what stands between it and agentic AI at scale. What matters is whether the infrastructure underneath it, latency, knowledge layers, observability, and access controls, can keep pace with agents querying systems continuously, at machine speed. Three patterns emerged. The first is that speed comes at a price. "Real time" means something different to every organisation, and that ambiguity carries a cost: teams routinely build expensive real-time pipelines before confirming the business actually needed sub-second delivery, when a five- or fifteen-minute cadence would have worked. That same price shows up in duplication and pipeline cost, as organisations keep real-time and historical systems in sync, with only a fifth to a third of data genuinely earning real-time treatment; in the still-unresolved difficulty of updating mutable records, like a payment status, at real-time speed; in joins that have to span fast-moving event data and slower recommendation systems within a single journey; and in data silos that leave organisations untangling fragmented systems before agentic AI can even enter the picture. The second is that intelligence has to sit above the data, not just describe it. Metadata and semantic layers are largely solved, but the knowledge layer above them, the one that lets people and agents reason consistently across systems, is not, and it keeps evolving as governance and accountability demands shift. The third is that trust has to operate at machine speed. Observability now has to account for the agent, not just the infrastructure it runs on, pushing leaders toward LLM gateways and smaller models evaluating larger ones. And access control has not caught up with how agents actually operate: a role assigned once, at login, cannot govern an agent whose risk profile shifts mid-task. Taken together, these patterns point to one conclusion. The organisations furthest along with agentic AI are not the ones with the most data or the newest models, but the ones that had already begun confronting latency, governance, observability, and access, long before agents made those questions impossible to postpone.

Category: Whitepaper | Published: 2026-08-24