Nearly every enterprise sits on a pile of data nobody uses, and that stockpile is now the biggest obstacle standing between AI pilots and AI that actually works, according to a new report from Everpure and analyst firm Omdia.

The infographic report, titled “Exploring the enterprise data readiness gap” and published on 8 September, found that 99 percent of organisations hold so-called dark data — information that’s collected and stored but rarely reused, much of it redundant, obsolete or trivial. More than half said this dark data makes up over a third of their total enterprise data.

AI pilots stall on messy data

The consequences show up further down the pipeline. Some 97 percent of organisations said they’re struggling to move AI projects from pilot to production, with 62 percent describing the obstacles as moderate to severe. Sixty-eight percent of IT leaders named data management their top challenge in getting AI into production, and 63 percent pointed specifically to storage silos and data sprawl.

Visibility is part of the problem. While 76 percent of IT leaders view dark data as a significant business risk, 58 percent admitted they still lack basic visibility into their own data environment.

“Scaling AI at speed but velocity without visibility is risky,” said Nathan Hall, Everpure’s VP and General Manager for APJ.

Hall said organisations that can accelerate innovation will be the ones that can answer three questions about any dataset: where it is, who is using it, and whether it can be trusted. Simon Robinson, Chief Analyst at Omdia, said the upside for those that close the gap is significant — turning fragmented data into intelligence that supports faster AI deployment and better decisions.

A framework for sorting the data

Everpure recommends organisations map their data along two axes — business value and risk — to decide what to activate, govern, retain or delete. High-value, high-risk data should sit under strict access controls; high-value, low-risk data should be prioritised for AI use; low-value data, whatever the risk, should generally be minimised or deleted.

The company frames this as an ongoing exercise rather than a one-off clean-up, arguing AI systems are only as reliable as the data feeding them.