New report reveals AI trust gap as poor data management undermines reliability
Real-time data has become a non‑negotiable requirement for organisations deploying agentic artificial intelligence, but most enterprise data architectures are not ready to deliver it, according to a new report.
The report from Denodo, based on a global survey of 850 executives and business decision-makers, found that 66% of organisations said AI outputs must be based on real-time data to be considered trustworthy.
Yet much of the infrastructure underpinning their AI projects was built for historical analytics rather than live operational decision-making.
"Agentic AI is moving enterprise AI from answering to doing," the report states, with agents now expected to perceive conditions, decide what to do next, and then execute actions across core systems.
"This shift raises the bar for data management, as the quality and relevance of data used by agents directly impacts the trustworthiness and effectiveness of those agents."
According to the report, 47% of the respondents said AI results must rely on data that is up to date "in real time" in order to be trustworthy.
Another 19% require data from "within the last minute" and 20% from "within the last hour." Only a small minority were comfortable with data that is more than a day old or purely historical.
For organisations already running AI in production, this demand is even more acute.
"Organisations already running AI in production are significantly more likely to require real-time or near-real-time data access than those still planning initiatives," the report reads.
Data architectures not keeping up
However, traditional data warehouses and lakehouses remain largely geared towards reporting and analytics.
Denodo points out that many enterprise architectures were designed for historical analytics rather than real-time operational decision-making, introducing latency as data is copied, transformed, and loaded through multiple pipeline stages.
The stakes rise as AI agents are embedded in frontline processes such as customer service, compliance monitoring, and supply chain management.
"For AI systems embedded in operational workflows, situational awareness is not optional — it is a prerequisite for trustworthy autonomy," the report reads.
"Data architectures designed primarily for historical analytics may struggle to support AI systems that must sense and act on real-time conditions."
Denodo argues that meeting these expectations will require on‑demand access to distributed operational systems, performance techniques such as caching and pushdown, and governance controls that ensure real-time access does not bypass security or policy enforcement.
Dominic Sartorio, vice president of Product Marketing at Denodo, said AI's transition to autonomous actions changes the data requirement entirely.
"When an AI agent triggers a business outcome, there is zero room for stale or ungoverned data. To scale agentic AI with confidence, businesses must move beyond static data silos and adopt a foundation of live, governed, and contextually relevant information," Sartorio said in a statement.
