AI’s Real Bottleneck Isn’t Models—It’s Data Context

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. The rollout is real, not just pilot projects anymore.

But here’s what’s catching people off guard: the hardest part isn’t the model. It’s not the compute either. It’s the data quality and—more importantly—the context behind it.

Irfan Khan, president and chief product officer of SAP Data & Analytics, 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 is what creates a return on investment for the business. Speed without judgment doesn’t help. It can actually hurt us.”

That’s the core tension. AI can generate answers at machine speed, but if it doesn’t understand which customers are strategic, which contractual obligations apply, or what tradeoffs are acceptable during a shortage, those answers might be technically correct and operationally disastrous.

The Context Problem Is Real

We’ve spent two decades building data warehouses, lakes, and dashboards. The goal was always aggregation—pull everything into one place so you can run reports and monitor performance. But in that process, a lot of meaning got stripped out. The relationships between data points, the policies behind them, the real-world decisions they’re supposed to inform—all that context got lost in translation.

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 accounts, what contractual obligations exist, what tradeoffs are acceptable during shortages. Both systems will analyze the data quickly. But only one will make decisions that actually help the business.

“Both systems move very quickly, but only one moves in the right direction,” Khan says. “This is the context premium.”

In the old days, humans bridged that gap. You’d have a supply chain manager who just knew which accounts were critical. But when AI acts autonomously—placing orders, adjusting prices, rerouting shipments—it doesn’t have that human intuition. It has whatever context you gave it. If that context is missing, the system optimizes for the wrong outcome.

The numbers back this up. Only one in five organizations considers its data approach highly mature. Only 9% feel fully prepared to integrate and interoperate with their data systems. Most companies know they have a problem.

Don’t Consolidate, Integrate

The emerging answer is a data fabric—an abstraction layer that sits across all your infrastructure, clouds, and operational systems. It doesn’t just move data into one place; it connects information while preserving the semantics that describe how the business actually works.

For agentic AI, this fabric becomes the primary interface. Instead of agents talking directly to databases or data lakes, they interact with business knowledge. Knowledge graphs are central here—they let agents query enterprise data using natural language and business logic, not SQL or API calls.

Khan breaks the value into three components:

• Intelligent compute for speed
• A knowledge pool for business understanding and context
• Agents for autonomous action grounded in that understanding

What makes this powerful is how they work together. Speed without context is dangerous. Context without action is useless. Action without speed is irrelevant.

This isn’t just a technical shift. It’s a strategic one. Companies that get this right will have AI that doesn’t just answer questions but makes good decisions. Companies that don’t will have fast, confident systems making bad calls at scale.

I’ve seen this pattern before—organizations chasing the latest model while ignoring the data plumbing underneath. The models keep getting better. But if your data foundation is a mess, better models just produce better garbage faster.

The data fabric approach isn’t new. It’s been around for years in various forms. But the rise of agentic AI makes it essential. You can’t have autonomous systems making business decisions if they don’t understand the business.

The real question isn’t whether your AI can generate answers. It’s whether those answers will be right.