New Research: Why Enterprise Agentic AI Stalls Before It Scales

The Agentic AI ROI Gap Is a Data Architecture Problem, Not an Ambition Problem

Teradata dropped a research report this week that, if you strip away the vendor framing, contains one of the more clarifying diagnostics of where enterprise AI actually stands in mid-2026. The study — 1,000 senior technology and data leaders across six global markets, conducted by Wakefield Research — arrives at a conclusion that will feel uncomfortable to anyone who has been approving AI budget lines for the past two years: nearly two-thirds of senior technology leaders have seen no more than a small or emerging positive return on their agentic AI investments, even as nine in ten plan to increase spending over the next twelve months. That is not a rounding error. That is a structural problem hiding behind a confidence narrative.

The report calls this dynamic the ROI Gap. I would call it something slightly sharper: enterprises are funding organizational AI ambitions on top of personal AI infrastructure, and then expressing surprise when organizational returns fail to materialize. Teradata’s CTO Louis Landry puts it well in the report: individual productivity gains — faster code, better drafts, quicker research — are real, but they do not show up on the P&L in a way that justifies significant infrastructure investment. The returns executives are chasing require agents operating at the organizational level, automating decisions, executing workflows, and driving measurable outcomes. Most organizations are not there. Only 7% have reached what the report calls the “Operationalizing” stage of its Agentic AI Maturity Index — the point where AI is actually executing multi-step workflows with measurable business impact. The remaining 93% are, to varying degrees, still watching pilots.

The Perception Gap Inside the C-Suite Is Its Own Warning Signal

Before getting to the data architecture argument, there is a secondary finding in this report that deserves more attention than it will probably receive. Sixty-nine percent of C-suite executives say their organization is already operating with agentic AI. Only 57% of VPs say the same. That 12-point gap between what the boardroom believes and what the people closer to implementation are experiencing is not a communications problem — it is an accountability problem. When executives believe deployment has already happened, the pressure to actually solve the underlying infrastructure challenges dissipates. The budget has been spent; the announcement has been made; the strategy slide says “AI-enabled.” The VP managing the pilot that never reached production is in a structurally awkward position to deliver that correction upward. CIOs and CDOs reading this should treat that gap as an organizational risk metric, not a data footnote.

Context Fragmentation Is the Actual Bottleneck

The report’s central technical argument revolves around what it calls context fragmentation — enterprise data that exists in volume but lacks the meaning, lineage, and governance that AI agents need to act reliably. The numbers here are striking: 77% of executives report that 20% or less of their enterprise data is sufficiently described and contextualized for agents to use. Seventy-eight percent find it challenging to unify data and knowledge across business functions so agents can reason across the full enterprise. The top two barriers cited are data lacking necessary metadata, context, and relationships (43%) and data fragmented across systems that cannot be connected in real time (42%).

This is the crux of the argument, and it is correct. The challenge enterprises face is not a shortage of data. It is that the data infrastructure built over the past two decades was optimized for human users — analysts who bring their own contextual judgment, who know which data source to trust, who understand what a given field label actually means in practice. AI agents have none of that ambient organizational knowledge unless it is explicitly encoded. When it is not, agents either hallucinate or stall. Forty percent of tech leaders report that more than 40% of their AI pilot projects fail to reach production because infrastructure systems were never built for autonomous use. That failure rate is not a technology problem in the narrow sense. It is a data readiness problem that was always there; agentic AI just exposed it.

The “Fix Everything” Trap and Why Ruthless Selectivity Wins

The natural organizational response to a data readiness problem is to launch a data quality initiative — which, in practice, often means an expensive, multi-year effort to contextualize the entire data estate before any agents go live. Teradata’s Chief Data and AI Officer Josh Fecteau makes a pointed argument against this approach in the report: the goal of contextualizing your entire data estate is likely the wrong goalpost, and chasing it is part of why organizations stall. The prescription he offers is ruthless selectivity — identify the highest-value portion of your data, get that portion fully described, governed, and agent-ready, and start there.

This is the right strategic posture, and it maps well to how successful technology transformations actually happen inside large organizations. Enterprises that tried to boil the ocean on data warehousing in the 1990s, or on master data management in the 2000s, largely failed. The ones that picked a domain, proved value, and expanded iteratively did not. The same logic applies here. CFOs and COOs who are being asked to fund broad “AI-ready data” initiatives should push back with a specific question: which three workflows, if agents could execute them reliably, would generate measurable P&L impact in the next 12 months? Start the data readiness work there, not everywhere.

The Action Bridge Problem Is About Trust, Not Technology

The report surfaces one more friction point that does not get enough coverage in mainstream AI coverage: even when organizations make progress on context fragmentation, getting agents to take consequential action is still hard. Sixty percent of leaders report decision paralysis on durable infrastructure decisions. Fifty-one percent cite accuracy and reliability of outputs as a significant deployment barrier. There is also what the report describes as a location problem — AI output currently lives outside the systems where consequential work actually happens. When intelligence is surfaced inside a tool or application where someone is already working, action follows. When it lives in a separate dashboard, it usually does not.

The location problem is underappreciated. Enterprises have spent considerable energy building AI capabilities and then surfacing them in new interfaces — portals, dashboards, standalone applications — that sit adjacent to but outside existing workflows. This is backwards. Agents that act on behalf of the organization need to be embedded in the systems where organizational decisions are made and executed: ERP systems, CRM platforms, ITSM tools, financial close workflows. The integration challenge is real and non-trivial, but it is the right problem to solve. Building beautiful AI interfaces that no one uses because they require a context switch is a costly detour.

The Vendor Framing Is Transparent But the Diagnosis Still Holds

It would be naive not to acknowledge the obvious: this is a Teradata-commissioned study, and Teradata sells a platform that it describes as a “context foundation, governance layer, and performance backbone” for AI agents. The study’s conclusions — that enterprises need better data contextualization, governance embedded in the data layer, and architectural portability — map with notable convenience onto Teradata’s product positioning. CIOs evaluating this research should read it with that lens applied.

That said, vendor-sponsored research can still be valuable, and this report clears that bar. The underlying data points — the 7% operationalization rate, the 77% figure on insufficiently contextualized data, the 12-point C-suite perception gap — are specific enough and troubling enough to be directionally credible even if one applies a skepticism discount to the precise numbers. The structural argument, that enterprises built data infrastructure for human judgment and are now surprised it does not support autonomous agents, is analytically sound and not a Teradata-specific claim. It is an accurate description of why the gap between AI enthusiasm and AI P&L impact remains as wide as it does heading into the second half of 2026.

What Executives Should Do With This

For CIOs and CDOs, the immediate action is the audit Fecteau describes: not a comprehensive data quality initiative, but a targeted assessment of which portions of the data estate are actually agent-ready today, and which high-value workflows could be unlocked if they were. That scoping exercise should drive both the infrastructure investment prioritization and the conversation with the board about realistic timelines for organizational-level AI returns.

For CEOs and CFOs, the relevant question is whether the ROI metrics being used to evaluate AI investments are actually measuring organizational AI or personal AI productivity. If the primary evidence of AI value in the organization is anecdotes about faster drafting and better meeting summaries, the measurement framework is lagging the investment thesis by at least one maturity stage. Closing that gap requires being honest about where on the Agentic AI Maturity Index the organization actually sits — which, based on this research, is almost certainly one or two stages behind where the C-suite believes it to be.

The companies that will have a meaningful competitive advantage from agentic AI in three years are the ones making deliberate, selective data readiness investments today. The ones that are waiting for the technology to mature, or for a platform vendor to solve the context problem on their behalf, are going to find that the gap between their AI strategy slides and their AI P&L impact has not closed — it has compounded.

Based on reporting from New Research: Why Enterprise Agentic AI Stalls Before It Scales, originally published 2026-07-07 08:30:00.