Most enterprises piloting AI agents, but many fail

Nasuni has published research showing that 97% of enterprises have deployed or are piloting AI agents, but 57% of AI projects are not meeting their objectives.

The findings are based on a survey of 1,000 IT purchasing decision-makers in the US, UK, France and the DACH region at organisations with more than 1,000 employees.

The results point to a broad disconnect between the pace of AI adoption and large organisations' ability to support those projects with data, governance and recovery processes. While businesses have moved quickly to test or introduce AI agents, only 18% said they had deployed them at scale.

Data management appears to be a central issue. The research found that 94% of organisations are struggling to manage unstructured data, even though it makes up most of their data footprint. Nearly half, 46%, said their AI initiatives had exposed weaknesses in data quality and governance.

Access problems are also affecting roll-outs. Some 79% of respondents reported inconsistent file access and performance across locations, suggesting companies with distributed operations still find it difficult to give teams and systems reliable access to the same information.

Scaling barriers

Nine in 10 organisations reported barriers to scaling AI. Security concerns were cited by 43% of respondents, while 36% pointed to integration problems and 33% said a lack of trust in data was holding projects back.

Those obstacles help explain why many companies are struggling to show a return on AI spending. Only 43% of projects were reported to be delivering on their objectives, suggesting pilot activity remains far easier than broad operational use.

The figures also indicate a shift in technology spending priorities. Only 16% of respondents currently treat unstructured data management as a core IT investment, but 60% said they plan to invest in it over the next 18 months.

That change comes as IT departments face additional financial strain. Some 62% of organisations expect hardware costs to rise as they expand AI use and update infrastructure for more data-intensive workloads.

Recovery delays

The survey also highlighted weaknesses in cyber recovery. Enterprises use four separate data protection systems on average, and that fragmentation appears to be slowing recovery after disruption.

Only one in four organisations said they were able to recover quickly from a cyber incident. Just 30% reported returning to operation within a week, while the typical recovery period was four weeks.

Those delays suggest that investment in security tools has not necessarily produced simpler recovery processes. Instead, multiple systems may be adding complexity at the point when organisations need speed and clarity.

The research also highlighted employee use of AI tools outside formal oversight. It found that 78% of employees now use AI tools for work, but only 30% of firms said they were concerned about unapproved or untracked AI usage.

That gap points to a relatively relaxed corporate response to shadow AI at a time when many businesses are still dealing with concerns about data security, governance and trust. In practice, organisations may be focusing more on scaling official AI projects than on monitoring how staff already use external tools in day-to-day work.

Sector gaps

The findings also showed differences between industries. In architecture, engineering and construction, 66% of firms identified security as their main concern around unstructured data. Manufacturers were described as facing elevated cyber risk and longer recovery times.

In energy, oil and gas, organisations were divided over whether AI decision-making should sit with the C-suite or IT teams. That split suggests some companies still lack clarity over ownership of AI strategy, a factor that can complicate investment choices and project goals.

Nasuni Chief Executive Officer Sam King said the survey showed many organisations were moving faster on AI than on the data work needed to support it.

"Enterprises are moving fast on AI projects, but most aren't getting the results they want," King said.

"What this report makes clear is that AI success depends on how well you manage and prepare your data. Too many organisations are still relying on outdated approaches to unstructured data management, limiting their ability to unlock its full value. Your proprietary, operational data is your greatest asset, but only if it's accessible and ready for your teams and the AI that supports them. At a time of rising hardware costs, supply risk and growing complexity, getting your data house in order is the ultimate no-regrets move - and it's exactly where we've been investing to help customers turn AI ambition into AI results."