Most firms not ready for agentic AI, Teradata study finds

Teradata and Wakefield Research have published a study on barriers to scaling agentic AI in large organisations. It found that only 7% of enterprises have reached a stage where the technology is delivering measurable business outcomes.

The survey of 1,000 senior technology and data leaders across six markets, including the UK, identifies enterprise data infrastructure as the main obstacle to wider deployment. It points to a gap between early AI trials and broader organisational use, where systems, governance and data lineage become more important.

Most respondents remain in the early phases of adoption. Under the study's four-stage maturity index, 68% of organisations are still in the experimenting or developing stages rather than operational use, where AI runs multi-step workflows with a measurable business impact.

A central finding is that many companies do not believe their data is ready for AI agents to use at scale. Some 77% of respondents said 20% or less of their enterprise data is sufficiently contextualised for agents to use effectively, while 78% said it is difficult to unify data and knowledge across business functions.

This points to a broader issue in how corporate data has been built and managed. The study found that 43% of leaders see data lacking metadata, context and relationships as a key barrier, while 42% cited fragmentation across systems that cannot be connected in real time.

Investment gap

The findings also suggest that spending plans are rising despite limited returns so far. Nine in ten senior technology leaders said they expect to increase investment in agentic AI over the next 12 months, yet 63% said they have seen only a small or emerging positive return on those investments to date.

Louis Landry, Chief Technology Officer at Teradata, said the mismatch stems from expecting enterprise-wide returns from tools that mostly improve individual productivity.

"Individual productivity gains - faster code, better drafts, quicker research - are real benefits, but they don't show up on the P&L in a way that justifies significant infrastructure investment. The ROI executives expect requires agents operating at the organizational level: automating decisions, executing workflows, driving measurable business outcomes. Most organizations are measuring enterprise AI ROI against personal AI infrastructure - and wondering why the numbers don't add up," said Louis Landry, Chief Technology Officer at Teradata.

The study draws a distinction between personal AI, such as chatbots and writing assistants used by individuals, and organisational AI, where agents work across shared company data under defined access controls and governance rules.

Pilot problems

Many AI projects are also struggling to move beyond the trial stage. The survey found that 40% of technology leaders said more than 40% of their AI pilot projects fail to reach production, while only 15% said their organisations are getting 80% or more of pilots into live use.

That shortfall is reflected in attitudes to infrastructure decisions. Some 60% of respondents reported decision paralysis over durable infrastructure choices, and 30% said they were concerned about vendor lock-in. Just over half, 51%, cited accuracy and reliability of outputs as a significant barrier to deployment.

The results also highlight differences in perception within senior ranks. While 69% of C-suite executives said their organisation is already operating with agentic AI, only 57% of vice presidents said the same, suggesting leaders do not share a common view of how far implementation has progressed.

Teradata argues that companies should focus on preparing the most valuable parts of their data estate rather than trying to overhaul everything at once. The report uses the term Autonomous Knowledge for data with enough context, lineage and governance for AI agents to act on it reliably.

Josh Fecteau, Chief Data and AI Officer and Chief Information Officer at Teradata, said organisations should narrow their focus.

"The goal of contextualizing your entire data estate is likely the wrong goalpost, and chasing it is part of why organizations stall. Instead, identify the highest-value portion of your data, structured and/or unstructured, and focus on getting that portion fully described, governed, and agent-ready. If most of the data is unusable, the answer isn't to fix all of it at once. It's to be ruthlessly selective about where you start," said Josh Fecteau, Chief Data and AI Officer and Chief Information Officer at Teradata.

The survey covered companies with at least 500 employees in the United States, United Kingdom, France, Germany, Japan and Saudi Arabia. Respondents were senior technology and data leaders at vice president level or above.