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

By the end of 2025, half of all companies had already pushed AI into at least three business functions. Finance, supply chains, HR, customer ops — everyone seems to be running some kind of copilot or agent. The hype is real, and the adoption numbers back it up.

But here’s the thing nobody wants to admit publicly: the hardest part isn’t the model. It’s not the GPU shortage either. It’s the data. Specifically, it’s the lack of context around that data.

Irfan Khan, who runs SAP’s data and analytics business, puts it bluntly: “AI is incredibly good at producing results. It moves fast, but without context it can’t exercise good judgment.” And good judgment, he reminds us, is what actually creates return on investment. Speed without judgment is just noise — or worse, bad decisions at scale.

I’ve seen this play out in real projects. A team builds a fancy AI assistant, feeds it clean inventory data, and watches it recommend actions that technically make sense but are operationally idiotic. Why? Because the system knows stock levels but doesn’t know which customers are strategic accounts, which contractual obligations are non-negotiable, or which products management has flagged as strategic priorities.

This is the context problem. And it’s worse than most executives realize.

The aggregation trap

For twenty years, the data strategy playbook was simple: extract everything from operational systems, dump it into a warehouse or lake, and build dashboards on top. That worked fine for human analysts. Humans bring their own context — they know the politics, the priorities, the unwritten rules. But AI doesn’t. When you strip data from its operational home, you strip away the semantics that tell you what it actually means.

Consider two companies both using AI to manage supply chain disruptions. One feeds in raw signals — inventory levels, lead times, supplier scores. The other adds context: which customers are strategic, what tradeoffs are acceptable during shortages, how extended supply chains interconnect. Both systems will analyze the data fast. Only one will make decisions that don’t piss off your most important client.

“Both systems move very quickly, but only one moves in the right direction,” Khan says. He calls this the “context premium.” I’d call it the difference between a useful tool and an expensive liability.

The maturity gap

The numbers are sobering. Only one in five organizations consider their data approach to be highly mature. Just 9% feel fully prepared to integrate and interoperate their data systems. That’s a lot of companies running AI on shaky foundations.

And the stakes are higher now because AI doesn’t just display information — it acts on it. When a human analyst looks at a dashboard, they can say “this looks wrong” and double-check. An AI agent just executes. If it doesn’t understand the business context, it optimizes for the wrong outcome. Inventory numbers might be accurate, but they don’t tell the model which customer gets the last unit when supply runs short.

Don’t consolidate, integrate

The emerging answer is a data fabric — not another repository, but an abstraction layer that sits on top of your existing mess. It connects data across applications, clouds, and operational systems while preserving the business semantics that matter. For agentic AI, the fabric becomes the primary interface. Agents query business knowledge, not raw storage.

Knowledge graphs are a big part of this. They let agents navigate enterprise data using natural language and understand relationships that flat tables can’t express. It’s not a new idea — knowledge graphs have been around for years — but the AI boom is finally giving them a real job to do.

Is a data fabric a silver bullet? No. Building one is hard, requires organizational buy-in, and takes time. But the alternative — throwing AI at poorly contextualized data and hoping for the best — is already failing in ways that cost real money.

Speed without judgment doesn’t help. It just gets you to the wrong answer faster.