Your AI is only as smart as the data fabric it sits on

By the end of 2025, half of all companies had AI running in at least three business functions—finance, supply chain, HR, customer ops. Copilots, agents, predictive systems—they’re everywhere. And yet, if you ask business leaders what’s actually holding AI back, they don’t say “not enough GPUs” or “models aren’t good enough.” They say the data is a mess. Not in the sense of missing or dirty data, but in a subtler way: the data lacks context.

AI can answer questions at lightning speed. But without understanding the business context behind the numbers, it can produce answers that are technically correct and operationally disastrous. Irfan Khan, who runs SAP Data & Analytics, put it bluntly: “Speed without judgment doesn’t help. It can actually hurt us.”

That’s the core problem. And the solution, according to Khan and a growing number of practitioners, is a proper data fabric—not just another integration tool, but a semantic layer that preserves meaning across systems.

The context gap is real

For the last twenty years, data strategy has been about aggregation. Extract data from operational systems, dump it into a warehouse or a lake, run dashboards on top. That approach works fine for reporting. But when you strip data from its original context—policies, processes, relationships, business rules—you lose the very things that give it meaning.

Khan gives a good example: two companies using AI to handle supply chain disruptions. Both have inventory levels, lead times, supplier scores. But one company also feeds in context—which customers are strategic, what trade-offs are acceptable during shortages, the status of extended supply chains. Both AI systems will analyze the data quickly. Only one will make the right call.

“Both systems move very quickly, but only one moves in the right direction,” Khan says. He calls this the “context premium.” I think that’s a useful way to frame it. In the old world, humans filled in those gaps automatically. But AI doesn’t have that instinct. If you don’t tell it why a piece of data matters, it will optimize for whatever it can measure—which might be the wrong thing entirely.

The numbers back this up. Only one in five organizations consider their data practices highly mature. Only 9% feel fully ready to integrate and interoperate their data systems. That’s sobering when you consider how many companies are already deploying AI agents that act autonomously.

Don’t consolidate, integrate

The emerging answer isn’t another data lake. It’s a data fabric—an abstraction layer that sits across your infrastructure, clouds, and operational systems. For agentic AI, the fabric becomes the primary interface. Instead of agents poking around raw databases, they query a layer that understands business semantics.

Knowledge graphs are a big part of this. They let agents navigate enterprise data the way a human would—following relationships, understanding hierarchies, respecting policies. It’s not about moving everything into one place. It’s about connecting things while preserving meaning.

This is a fundamentally different approach from the “one big repository” mindset that dominated the last decade. And frankly, it’s overdue. AI agents that don’t understand context are just fast, confident idiots. A data fabric that embeds business logic and metadata gives them the guardrails they need.

Of course, building this isn’t trivial. It requires mapping processes, defining semantics, and maintaining that layer as the business evolves. But the alternative—letting AI make decisions without context—is worse. Speed without judgment isn’t just unhelpful. It’s dangerous.