Poor data infrastructure — not AI itself — is the biggest barrier to AI success

Full text

We’re in what’s being dubbed the ‘year of AI ROI’. Four years on from the AI boom - ignited by the launch of ChatGPT - many businesses now believe that they’re AI-ready, using successful early-stage chatbot and copilot rollouts as evidence.

Yet there is a significant difference between experimenting with AI and embedding it across complex business processes and operations.

The real barrier to this transformation sits beneath the models themselves. While most AI outcomes-related conversations focus on model performance, GPUs, and compute capacity, organisations are increasingly realising that it’s their data infrastructure holding AI projects back.

Up until now, businesses have got away with operating successfully despite their disconnected file environments, inconsistent data governance, and information spread across multiple repositories.

This is because historically, data has been accessed sporadically, mostly by human employees who could compensate for shortcomings when data lacked context or wasn’t where it should be.

Simply put, data infrastructure didn’t need to be perfect, so it wasn’t. However, these fragmented file environments are now impossible to ignore as AI demands 24/7 consistent access to information, strong governance, and trusted data.

This puts enterprises in an uncomfortable situation: under pressure to deploy AI, but missing the foundations needed to be truly successful with AI projects at scale.

AI makes the data puzzle more complicated

Traditional file environments were built for a world where information lived in separate locations, individual teams managed access, and gaps in governance or data availability could be worked around.

This means that enterprises are now operating with fragmented information spread across multiple locations, with inconsistent governance and varying levels of accessibility. When data is accessed primarily by human employees, these inefficiencies are a source of frustration rather than a barrier.

This is where AI changes things. Models depend on fast, reliable access to information. If data is difficult to find, lacks context or cannot be accessed consistently, its value diminishes rapidly. In this instance, enterprises have data, but it’s completely unusable.

Overconfidence is coming at the expense of real success

Most organizations realize that their data infrastructure is far from perfect. Yet, the pressure to demonstrate progress on AI is intense. If businesses aren’t investing in and experimenting with AI, they’re seen as behind the curve. This is creating a strong incentive to deploy new tools as quickly as possible.

The problem is that organizations' early-stage AI learnings, if not outright success, are creating an overconfidence in their wider AI readiness and ability to scale up. Chatbots and copilots, which provide a lower barrier to entry to AI, are enabling organizations to demonstrate early, tangible AI outcomes.

When these projects deliver value, enterprises start to think that perhaps their infrastructure is future-proofed after all, despite a lack of sustained attempts to address and resolve deeper structural issues around their data accessibility, governance and infrastructure. These same businesses are also struggling with data recovery after cyber incidents - the warning signs are there, but current overconfidence is blurring reality.

As a result, businesses are moving rapidly onto major agentic projects, only to face delays, questions over ROI, and often, an implementation that fails. The irony is that an urgency to deploy AI is actually slowing long-term progress.

Organizations hellbent on deploying AI are risking overlooking the deeper risk management and data management work required to support it, while those that invest in strong data foundations are more likely to achieve sustainable success as their AI ambitions grow.

Shifting data from a cost to an asset

Most businesses don’t suffer from a lack of data; if anything, they have the opposite problem. Unfortunately, this is still viewed as an operational cycle driven by capacity planning, refresh cycles and expansion requirements, rather than a strategic asset. Industry research consistently shows IT teams still struggle to get a grip on their unstructured data - where most of their data assets reside - despite most of them believing their file data setup is robust.

The difference now is that AI systems operate continuously, drawing on information created today, yesterday, and sometimes decades ago. To deliver meaningful outcomes, they require consistent access to trusted data, rich context, and clear governance.

This means that organizations need to rethink their approach to enterprise information. When data is treated purely as a storage refresh challenge, AI quickly exposes the limitations of that approach. However, when it’s treated as a strategic asset, organizations can create environments where information is accessible, governed and ready to behave in the way that AI expects.

C-Suites’ goal should be to move from steady-state storage capacity to data utility, creating centralized environments with fewer systems, where data can be used no matter where it is located. This shift reduces fragmentation, simplifies data management and creates a unified view of the enterprise’s information, and with it the path to competitive edge.

The fewer barriers that exist between AI and the data it needs, the easier it becomes to move from isolated AI successes to meaningful deployment at scale for productivity and efficiency benefits.

Fix the foundations before scaling AI

The success of chatbot or copilot rollouts does not indicate AI readiness. It is instead measured by data utility: whether the data underpinning AI tools is accessible, secure, and fit for purpose at scale.

The organizations that succeed with AI in the coming years will not be those deploying the most advanced models or the biggest number of agents, but those who can provide AI with consistent access to trusted, well-governed data.

Those boards that keep investing in sophisticated AI tools without upgrading their data layer and governance will find their operations shackled for years to come by the same underlying data access and governance limitations.

It’s time to stop viewing data as an operational burden and start treating it as a strategic asset, helping unleash organizations' true IP and powering their ability to innovate. Building strong data foundations today will set organizations on the path to realizing long-term AI value, instead of discovering down the line that AI ambitions outpace their ability to support them.

We've featured the best data removal service .

This article was produced as part of TechRadar Pro Perspectives , our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit