Two-thirds of business leaders do not fully trust their data. That single figure, drawn from research by the Institute for Data and Enterprise AI, may explain more about the slow pace of enterprise AI adoption than any conversation about model capability.
The scale of experimentation is real. According to the announcement, nearly two-thirds of companies were running AI agent pilots in late 2025, and 88% were using AI in at least one business function — up from 78% the prior year, per McKinsey‘s annual AI report. Yet only one in ten companies actually managed to scale those agents beyond the pilot stage.
The gap between experimenting and scaling, according to SAP president and chief product officer Irfan Khan, is not a model problem. It is a data problem. “The only prediction anybody can reliably make is that we don’t know what’s going to happen in the years, months — or even weeks — ahead with AI,” he says. “To be able to get quick wins right now, you need to adopt an AI mindset and ground your AI models with reliable data.”
Context, not volume
The conventional assumption has been that structured data is high-value and unstructured data is not. Khan argues that framing no longer holds. What determines data quality for AI agents is business context, not format. Supply-chain operations, financial planning, even high-volume telemetry and IoT feeds — all carry potential value, but only when delivered alongside the context that makes them interpretable.
“Anything that is business contextual will, by definition, give you greater value and greater levels of reliability of the business outcome,” Khan says. “It’s not as simple as saying high-value data is structured data and low-value data is where you have lots of repetition — both can have huge value in the right hands, and that’s what’s different about AI.”
The failure mode, then, is not a shortage of data. It is a shortage of grounding. When data lacks shared definitions, semantic consistency, and operational context, it accumulates what the report calls “trust debt” — a compounding liability that stalls AI readiness across the organization.
Sprawl as the structural obstacle
The architecture problem has a clear origin point. Over the past decade, separating compute from storage unlocked cloud-scale flexibility — Khan calls it “probably the biggest innovation that occurred in data management.” But that same move distributed data across multiple clouds, data lakes, warehouses, and hundreds of SaaS applications.
More than two-thirds of companies now cite data silos as a top challenge in AI adoption. More than half of enterprises are managing 1,000 or more data sources. The sprawl does not shrink when a company pivots to AI deployment. It tends to grow.
The answer Khan points to is a semantic or knowledge layer — one that sits across multiple platforms, encodes business rules and relationships, and delivers harmonized data to autonomous agents regardless of where that data physically lives. “What’s really making a distinction now,” he says, “is the way that we harmonize the data and harvest the value of the data across multiple sources of content.”
The next few months to years, Khan says, will be the window in which companies either build that foundation or fall behind the companies that did.
This article is a curated summary based on third-party sources. Source: Read the original article
