According to new data from autonomous AI knowledge platform Teradata, a persistent tension remains in enterprise agentic AI adoption. Despite continuous and aggressive investment, many organizations are failing to move from experimentation to enterprise-wide adoption.
Based on a survey of 1,000 senior technology and data leaders, Teradata's 2026 report, Arrested Automation: Why Agentic AI Stalls at the Enterprise Level, identifies misaligned data and measurement structures as a root cause of this ROI gap and offers guidance for enterprises to shift their strategy to maximize returns on their AI investments.
Enterprise AI investment doesn’t automatically lead to enterprise-wide ROI
The report found that although 90% of senior technology leaders expect to increase agentic AI investments over the next 12 months, only 37% of organizations report measurable business impact. Sixty-three percent say they have seen no more than a small or emerging positive return on their AI investments to date.
To show where organizations are on this journey, the report categorizes them into an agentic AI maturity index. About a quarter of organizations (28%) are in the experimenting stage, exploring localized pilot projects that often lead to personal productivity gains. The 40% of enterprises in the developing stage have some successful models and automations but haven’t figured out how to connect knowledge outside of individual team silos. Another quarter of companies are in the building stage. These companies have deployed basic structured governance and workflows but haven’t addressed the data foundations to scale autonomous decision-making across the enterprise.
There are only 7% of organizations in the operationalizing stage. They have implemented dynamic governance and safety rules and harmonized and enriched data to the point where AI can confidently execute multi-step workflows. It’s at this stage that organizations make a meaningful shift to “autonomous knowledge,” a data foundation that stays current, contextually rich, and governed over time without constant manual intervention. According to the report, this shift is what organizations need to begin to make ROI possible.
A key difference in personal AI vs. enterprise AI
The key reason 68% of companies remain in the experimenting or developing stages is that they’re trying to achieve enterprise-level impact with infrastructure designed for personal AI work. The data and measurement layers they’re building with are misaligned with the business outcomes they need.
This misalignment stems from context fragmentation. This is when companies try to move from systems designed for humans separated into business units to an interconnected, context-rich foundation that agents can understand and operate effectively within and across.
When organizations build highly successful, localized proofs of concept using clean, sampled datasets, they still rely on humans to provide additional context to make it work well. The moment engineering attempts to push those models into live, automated, enterprise-scale production, the AI begins to struggle. Each dataset has its own structure and rules, and usually lacks essential context and shared meaning for agents to understand across systems where real enterprise-scale can occur.
In the report, 77% of leaders state that only 20% or less of their enterprise data and knowledge is ready for AI agents to act on reliably, and 78% report difficulty creating the connected data foundation that is crucial for enterprise-wide agentic AI success. Additionally, more than half of the leaders surveyed explicitly cite accuracy and reliability of outputs as a significant deployment barrier.
Forty percent of tech leaders report that more than 40% of their AI pilot projects pause before production because their infrastructure is not ready for autonomy. The data infrastructure they are built on was never designed for agent understanding. It isn’t structured to enable autonomous systems operating at scale.
The report suggests a solution: Change the order of operations by focusing on the data foundation before the software layer. When an organization identifies the critical systems to unify and enrich with the context, lineage, and governance for agents to act on reliably, the company is better positioned to scale its pilots and achieve real ROI.
Ultimately, a foundation of autonomous knowledge with enough context, lineage, and governance that agents can act on reliably, across functions, and on behalf of the whole organization can tip the scales from individual productivity to enterprise-wide ROI.
Methodology
The report, Arrested Automation: Why Agentic AI Stalls at the Enterprise Level, was conducted by Wakefield Research on behalf of Teradata. The study surveyed 1,000 senior technology and data leaders at the vice president level or above, at companies with a minimum of 500 employees, across the United States (500), the United Kingdom (100), France (100), Germany (100), Japan (100), and Saudi Arabia (100). Fieldwork was conducted between March 23 and April 5, 2026.
This story was produced by Teradata and reviewed and distributed by Stacker.
