Data issues stall AI projects at most organisations

Iron Mountain has published research indicating that data issues are delaying or cancelling artificial intelligence projects at most organisations. The study found that 86% of respondents had faced disruption to AI work over the past 12 months.

The findings point to a gap between companies' stated AI ambitions and the condition of the information needed to support wider deployment. While 95% of organisations said using data to improve business outcomes is a top priority, many are still struggling with fragmented systems, unstructured content and older information stores.

The survey, conducted by Vanson Bourne and commissioned by Iron Mountain, found that 81% of organisations believe AI cannot scale beyond pilot projects unless siloed and unstructured content is turned into AI-ready data. It also showed that only 39% of respondents fall into what the report describes as a "digitized and structured" category, where dark data is centrally managed, indexed and available for analytics and AI use.

Manual preparation remains a significant burden. Four in 10 organisations said they spend at least 41% of AI project timelines on manual data preparation, suggesting much of the work required to move AI projects into production still lies in data handling rather than model development or deployment.

Data barriers

The research identified disconnected systems and departmental silos as the most common obstacle, cited by 56% of organisations. A further 53% pointed to legacy formats or physical archives as barriers to making information usable for AI projects.

Those constraints become more significant as businesses look beyond generative AI tools to more autonomous systems that can act across workflows. In that context, the ability to trace and govern information becomes more important, particularly in sectors dealing with regulated records, audit demands and large stores of historic content.

The report found that 70% of organisations cannot effectively track the lifecycle of the data used in their AI models, including its origin and movement. That creates problems not only for model performance but also for compliance, security and legal discovery, especially when organisations cannot show whether information is current, authorised for use or properly governed.

Iron Mountain, which focuses on information management services, framed the issue as one of operational readiness rather than enthusiasm for the technology itself. It said businesses are moving towards greater AI autonomy without the information foundations needed to support that shift.

Varun Gupta, Senior Vice President, Global Product & Solutions Marketing, Iron Mountain, said: "AI readiness is about much more than deploying the latest technology. Enterprises need to know that the information powering their AI is accessible, current and governed, with clear ownership and permissions. As AI agents move from generating answers to taking actions across business workflows, these foundations become essential to scaling AI with confidence."

Governance pressure

The findings add to a wider debate in the technology sector over whether companies' AI spending is running ahead of their data governance and infrastructure. Many organisations have spent the past two years testing generative AI applications in customer service, internal search, document handling and software development, but broader deployment often depends on whether underlying records can be found, cleaned, structured and governed.

The challenge is especially acute for large businesses with information spread across multiple departments, geographies and storage systems. Physical records, archived documents and older file formats can make it difficult to assemble complete and reliable data sets for AI systems, while different access rights and retention policies can further complicate deployment.

For suppliers and service providers in the channel, the results suggest potential demand for systems integration, information management and governance work linked to AI projects. Businesses trying to move from pilots to production may need support connecting separate repositories, preparing data for specific use cases and setting rules around ownership, permissions and lifecycle management.

Iron Mountain said organisations need continuous lifecycle governance that provides an auditable view of information from creation to destruction. It argued that such oversight can help businesses verify that the data used for AI is current, compliant and trusted.

The scale of Iron Mountain's customer base gives the findings relevance beyond a narrow group of early AI adopters. It said it works with more than 240,000 customers across 61 countries, including about 95% of the Fortune 1000.

Among the clearest findings was the extent to which basic information management remains unresolved: 56% of organisations cited disconnected systems or departmental silos, 53% cited legacy formats or physical archives, and 70% said they cannot effectively track the lifecycle of data used in AI models.