Data fragmentation blocks ROI for enterprise agentic AI

The ROI bottleneck

Organizations are facing a significant hurdle in turning agentic AI from experimental pilots into profitable business tools. While 90% of senior technology and data leaders expect to increase their investments in agentic AI over the next year, the actual results are lagging. As the Garden Island reported, only 37% of these companies currently see strong positive returns from their efforts.

The primary culprit for this stagnation is data fragmentation. Most companies have massive amounts of information, but that data was originally designed for human consumption. Humans can infer meaning and context from a messy spreadsheet or a disorganized folder, but AI agents require much more explicit structure to act independently.

This gap between having data and having usable data creates a bottleneck for scaling. Many companies are stuck in the 'experimenting' or 'developing' stages, where they see small personal productivity gains but struggle to achieve organizational-wide impact. Without a way to unify knowledge across different business functions, agents cannot reason effectively across the entire enterprise.

The context gap

The problem is not just that data is stored in different places; it is also that the data lacks sufficient description. The report reveals that a majority of executives report that 20% or less of their enterprise data is actually contextualized enough for an agent to use. This means the vast majority of corporate information remains 'dark' or unusable for autonomous systems.

When leaders were asked to identify specific barriers, they ranked the lack of metadata, context, and relationships as a major issue. In fact, this lack of context (43%) was ranked nearly as high as the physical fragmentation of data across different systems (42%). This suggests that cleaning up the data is only half the battle; the other half is defining what that data actually means.

To move forward, businesses need to create what researchers call 'Autonomous Knowledge.' This is data that carries enough inherent context, governance, and meaning for an AI agent to act on it reliably and repeatedly. Without these layers, agents are limited to simple tasks rather than the complex, multistep workflows that drive real business value.

Maturity and roadblocks

The report introduces an Agentic AI Maturity Index to categorize how different companies are progressing. Currently, 28% of enterprises are in the experimenting stage, focusing on localized pilots that only offer personal benefits. Another 40% are in the developing stage, where they have found some success with automation but are hitting walls due to poor data connectivity between teams.

A smaller group of 25% has reached the building stage. These companies have established some basic structured governance and workflows, but they still haven't solved the problem of knowledge fragmentation. They have more connected data than their peers, but they still lack the critical context layers needed for an agent to make autonomous decisions at a high level.

Only 7% of enterprises have reached the operationalizing stage. These are the companies that have successfully implemented dynamic governance and safety rules alongside a solid data foundation. These organizations are the ones finally starting to see measurable business impacts because their AI can execute complex, multistep processes across the entire company.

A pragmatic roadmap

Many leaders feel paralyzed by the scale of the task, with 60% reporting decision paralysis regarding technology choices. There is a common fear of vendor lock-in, with 30% of leaders citing it as a major concern. To solve this, the report suggests moving away from trying to fix every single piece of data at once, which can cause a company to lose momentum.

Instead, the recommended strategy is to be selective. Organizations should pinpoint the highest-value portion of their data—perhaps just 20% to 50%—and focus on making that specific portion fully described and governed. By prioritizing the most important datasets, companies can see faster results without getting bogged down in a total infrastructure overhaul.

The goal is to build data into reusable, governed assets that any model can understand. By embedding the necessary context and connections within the foundational architecture, the data becomes portable. This allows a company to switch out different software tools or AI models without having to rebuild the underlying knowledge base every time the technology changes.

Next steps for leaders

To reach the operationalizing stage, the report outlines a clear sequence of actions. First, leaders should audit their data estate to identify high-value datasets. They must then apply consistent definitions, lineage, and meaning to those data points across all departments, regardless of who owns the information.

The second step involves embedding dynamic governance directly into the foundational data architecture. By doing this rather than at the software or application layer, the company ensures that guardrails and context stay with the data. This allows any agent to reason in a way that mimics human logic while ensuring that all decisions remain traceable throughout a multistep process.

Finally, building for portability ensures long-term flexibility. When data is connected and contextualized correctly, the underlying governance remains intact even as the organization evolves. This approach enables enterprises to realize the full ROI of agentic AI while maintaining the freedom to adopt new tools as the market changes.