Why AI can't reach the data that matters

Many firms’ AI strategies are yet to deliver game-changing business outcomes because they aren’t built on the right foundation: advanced data access and governance capabilities. In a nutshell, organisations still can’t surface the right data for the AI model to gain a real competitive edge.

 

While many C-level executives view AI models, GPUs, and compute capacity as weaknesses in AI, these are often the parts of the company tech stack that are improving fastest. The fundamental barrier is the organisation’s existing data infrastructure, which could meet yesterday’s information retrieval demands, but cannot reliably deliver the always-on, organisation-wide context that AI tools require.

 

For years, companies have regarded file storage as a steady-state layer that needs to be maintained and refreshed in cycles by IT teams when capacity and access capabilities need improving. Data was requested intermittently, mainly by human users who could fill capacity gaps when needed.

 

In contrast, today’s AI agents remove that margin for error by demanding continuous, system-level access to complete and consistent data across the organisation. IT teams never envisaged such demand and their data infrastructures were never designed to service it.

 

 

AI obstacles

Our survey of 1,000 US, UK & European organisations has revealed a raft of obstacles to companies improving their AI projects’ performance and boosting business value – data governance issues, risk to their data infrastructure and ongoing dilemmas over funding their AI projects and maintaining their existing technology stack

 

The good news is that modern, strategic approaches to data access, particularly identifying data’s value, can provide a pathway for resolving these challenges and delivering stellar AI performance.

 

 

Data governance issues

The first trend in our research is the data quality or governance challenges affecting almost half of organisations, with these issues increasing in importance the greater the organisation’s AI maturity – two-thirds of firms with AI tools embedded in their processes say it’s a problem.

 

Strategic thinking can address these data governance questions and provide stronger guardrails for integrating AI tools. A global construction firm recently paused an AI copilot pilot for its design teams after discovering the models were surfacing project content across confidential client engagements - a real legal and commercial risk.

 

With external input, the firm is applying its existing Active Directory permissions to AI queries at the point of retrieval, so the AI only returns content the requesting user is entitled to see. The aim is an enterprise-wide rollout of AI across design teams, giving teams faster access to institutional knowledge while keeping project IP properly isolated.

 

 

Emerging risks

A second major finding is IT teams crossing their fingers over their data infrastructure’s performance: while 70% say their file data infrastructure can support upscaled AI, nearly all (94%) are struggling to manage their organisation’s unstructured data – which typically accounts for more than 90% of an organisation’s data assets.

 

Strategic improvement of data infrastructure can reduce enterprises’ data risks and pave the way to better AI project performance.

 

A global manufacturer held petabytes of unstructured file data across dozens of sites and legacy NAS systems, with assets comprising engineering documents, quality records, and standard operating procedures. Its early AI initiatives could reach only a fraction of that estate, limiting model accuracy and pushing teams toward ungoverned workarounds.

 

The company is now consolidating data onto a single cloud-backed file platform with a governed index that AI applications can query directly. The eventual outcome is faster, more reliable answers for engineers and shift supervisors, drawn from authoritative content across every site, rather than isolated pockets of data.

 

 

AI funding dilemma

The third trend from the research was organisations exhibiting “chicken and egg” indecision over how to fund a data infrastructure for the AI age.

 

Half of the survey (50%) say they are having to increase spending on storage because of rapid data growth, but almost as many (43%) are struggling to balance their budget given competing demands from AI and creaking storage infrastructures. Even keeping data storage infrastructures running in steady state remains a headache for IT teams - 43% say their biggest concern is the strain that regular storage hardware refreshes or expansion cycles put on their daily operations.

 

Practical advances are helping organisations resolve these funding dilemmas. A European energy company faced rising storage costs as its data volumes grew, while its board pressed for faster returns on AI investment. Rather than fund a legacy NAS refresh and a parallel AI data pipeline simultaneously, the company is working to consolidate distributed file storage onto a single cloud-native platform, retiring hardware, backup, and disaster recovery costs in the process.

 

The enterprise’s objective is to reallocate the savings from the consolidation directly into the AI programme, so the same platform that is reducing infrastructure spend also feeds governed file data to AI applications, avoiding a second and costly build.

 

 

A data governance layer

The survey findings and examples of data-value-led storage / AI implementations chime with industry experts who have argued that a simpler, better-governed data access layer is the missing link in optimal AI performance. This approach enables large language models to interact directly with both structured and unstructured data “in place” using the right permissions to better harness their power.

 

The organisations that design strategic data layers will provide effective data governance, transforming infrastructure performance and treating data storage management as an IT budget priority. This in turn will underpin risk-contained AI implementations that achieve transformative innovation and workflow efficiency while avoiding future IT investments being undermined by outdated approaches to storage infrastructure and capacity.

 

 

Surfacing governed and secure data for AI

Forward-looking enterprises are shifting from passive management of storage capacity to actively prioritising data’s value for AI. This will move organisations towards data management infrastructures that ensure critical data remains consistent and usable, whatever the demands that AI models put on it.

 

Companies that fail to re-engineer their data infrastructures as a strategic data layer could find AI models’ performance is undermined - however much they spend on cutting-edge AI tools and compute innovations.

 

---

 

Nick Burling is chief product officer at Nasuni

 

Main image courtesy of iStockPhoto.com and amgun