If you are waiting for broad enterprise AI adoption to happen, you have already missed it. According to new research from cloud file data company Nasuni, published this week, 97% of organisations have either deployed AI agents or are actively piloting them. The AI wave has arrived.

The problem is that it largely is not working.

The same report — the State of Enterprise File Data Annual Report 2026, based on a survey of 1,000 purchasing decision-makers across the US, UK, France, and DACH region — found that 57% of AI projects are not delivering their stated objectives. Only 18% of organisations say they have deployed AI agents at actual scale.

That gap between activity and outcomes is the story. And the culprit, hiding in plain sight throughout the report, is data.

The Data Readiness Problem Nobody Wants to Admit

Nearly half of organisations surveyed said that their AI initiatives have directly exposed gaps in their data quality or governance. Almost all enterprises — 94% — report struggling to manage unstructured data, which comprises the majority of their overall data footprint.

Here is the uncomfortable part: most companies know they have a problem but have not acted on it. Only 16% currently treat unstructured data management as a core IT investment. Yet 60% say they plan to prioritise it over the next 18 months — a classic case of planning to fix the thing they should have fixed before they started.

What makes this worse is the confidence gap. 70% of organisations believe their file data infrastructure can support AI scaling. The data in this same report repeatedly contradicts that belief. Fragmented storage, governance gaps, inconsistent data access, and weak recovery systems are cited as persistent issues. Organisations are overestimating their readiness, and it shows in results.

You Cannot Build a Smart Agent on Dumb Data

This is not a technology problem. Every major AI platform — whether you are building with Claude, GPT-4o, Gemini, or open-source models — can produce genuinely useful results when given accurate, well-structured, accessible information. The model is not what fails. What fails is the data that the model is given to work with.

AI agents retrieve information to answer questions, make decisions, and complete tasks. When that information is siloed across disconnected systems, lacks governance, contains outdated records, or exists in formats agents cannot easily process, the agent fails — not because the AI is bad, but because it is working with a broken foundation.

The Nasuni findings align with broader industry research. A separate Cloudera and Harvard Business Review Analytic Services report found that only 7% of enterprises say their data is completely ready for AI. The gap between AI ambition and data reality is not a niche concern — it is the defining operational challenge for businesses in 2026.

What This Means for Business

The organisations getting results from AI agents are not the ones that moved fastest. They are the ones that did the unsexy groundwork first: cleaning data, building governance frameworks, consolidating fragmented storage, and ensuring their teams could actually understand and work with data before handing control to automated systems.

A few patterns separate the 18% that have reached scale from the 57% that are struggling:

They treated data as a product. Rather than leaving data as a byproduct of business operations, high-performing organisations actively manage it as a strategic asset with ownership, quality standards, and governance processes.

They invested in data literacy alongside AI tools. Deploying an agent on top of a team that does not understand how to interpret its outputs or question its assumptions is a recipe for decisions made by confident-sounding AI on bad information. Companies that trained their people in data skills saw measurably better outcomes from AI adoption.

They started with narrow, well-defined use cases. Instead of deploying agents across the business simultaneously, they identified one process with clean, accessible data, proved the value, and expanded from there. Small wins built the foundation for scale.

They measured outcomes, not activity. 97% adoption sounds impressive until you ask what “adoption” actually means. Companies with genuine AI ROI tracked specific business outcomes — time saved, errors reduced, revenue influenced — not licence counts or tool rollouts.

The Upskilling Gap Is Part of the Same Problem

Separately, enterprise AI training data tells a consistent story. 82% of enterprise leaders say their organisation provides some form of AI training, yet more than half still report an AI skills gap. The issue is not access to training — it is the quality and relevance of what is being taught.

Video-based generic AI courses do not translate into practical capability when an employee sits down to work with a company’s actual data systems. The organisations closing the skills gap are the ones combining structured data literacy training with hands-on applied practice against their real tools and datasets.

This matters beyond the individual employee. When a company’s data team cannot assess the quality of the outputs an AI agent is producing, no amount of automation will save them from building on bad foundations.

The Bottom Line

The AI adoption race has already been run. Almost everyone entered. The question now is who will finish — and the answer will be determined almost entirely by who has the data foundations and human capability to support what AI agents actually need to do their job.

If your organisation is in the 57% where projects are not delivering, the answer is almost certainly not a better model or a different vendor. It is a data quality problem, a governance problem, or a skills problem — most likely all three.

The companies that treat those as separate from AI strategy will keep spending and keep getting disappointing results. The ones that recognise data readiness and team capability as the foundation of AI success will be the ones still standing when the dust settles.

This finding aligns with concurrent research from HCLTech: a survey of 467 senior executives found that 43% of enterprise AI initiatives are expected to fail — with the same root cause: the gap is not in the technology, it is in the organisations deploying it.

---

Enterprise DNA helps organisations build both pillars of this equation — data literacy through EDNA Learn and AI-ready systems through Omni by Enterprise DNA. If your AI projects are not delivering the outcomes you expected, let’s talk about why.