Seventy percent of organisations say less than half their unstructured data is discoverable and usable for analytics or AI, according to new research from analyst firm BARC. The finding lands as enterprises push AI agents into production on top of documents, emails and images they cannot fully see.
The study, Harnessing Unstructured Data for AI Innovation, surveyed 225 data, analytics and AI leaders across North America and Europe in February and March 2026.
The research exposes a sharp gap between confidence and capability. Four out of five respondents are confident they can extract value from unstructured data without compromising governance controls. Yet roughly one third report inconsistent or non-existent controls for data bias and data lineage. About one third cannot consistently enforce data access and usage policies.
"More troubling still, some AI adopters still 'don't know what they don't know'," the report states. "Automated decisions and actions by agents only compound the potential damage of poorly governed unstructured data."
The stakes are rising as autonomous agents act on data without a human in every decision loop. The authors warn that laggards "must pay close attention and move fast to catch up, because unstructured data provides the deep context that AI agents need to take smart, safe actions."
The findings echo a recent Cloud Security Alliance survey, which found more than half of organisations have only partial visibility into where unstructured data is stored.
Despite the gaps, BARC found surprising progress on governance. Across every category, more respondents have implemented or optimised governance controls for unstructured data than a year earlier. Data quality controls jumped to 47 percent implemented or optimised, up from 36 percent last year.
Maturity remains uneven. North American respondents outpace Europeans in every governance category, including regulatory compliance at 58 percent implemented or optimised versus 48 percent. "This begs the question of whether General Data Protection Regulation (GDPR) and the EU AI Act have achieved their privacy and safety objectives," the report says.
Asked how they measure the AI readiness of unstructured data, respondents pointed to accuracy (54 percent), consistency (48 percent), business relevance (44 percent), completeness (39 percent) and timeliness (31 percent). The report notes these measures require probabilistic logic and judgement rather than deterministic quality rules. That contrast with structured data keeps humans in the mix, with human validation ranking high at 41 percent.
Skills and training top the list of obstacles, cited by 36 percent of respondents, ahead of data privacy (30 percent) and data quality (29 percent). Data quality also dominates success measures at 48 percent, as organisations recognise the risks of faulty model inputs.
Classification (60 percent) and validation (56 percent) are the most widely adopted tasks for preparing unstructured data, well ahead of formatting, cataloguing and indexing. Vectorisation trails at just 17 percent. "You need your map before you begin the voyage," the authors write.
Generative AI dominates model deployment, with 73 percent of respondents using generative models with unstructured data. Microsoft Azure leads platform choices at 52 percent, ahead of AWS (38 percent) and Google Cloud (32 percent). More than half of databases containing unstructured data reside on-premises or in hybrid environments.
Given that distribution, BARC recommends organisations build toward a platform-independent semantic layer covering federated querying, aggregated data views and integrated metadata. The firm expects adopters to favour platform-independent tools that support open ecosystems and eliminate lock-in risk.
BARC identifies a cadre of AI "Leaders" with mature, cross-functional programs, comprising 23 percent of the market. Leaders tackle more sophisticated projects, reach production faster and achieve better business results. They show higher maturity than other adopters in every governance category.
The full report is available here.