The use of unstructured data in analytics ratcheted up after big data systems emerged in the mid-2000s, but structured data continued to hog the spotlight in most organizations. Now, unstructured data is increasingly critical to the success of AI applications — and ensuring it’s AI-ready is becoming an essential task for data teams.
“Unstructured data is where true competitive advantage now lies,” said William McKnight, president of McKnight Consulting Group. Documents, emails, transcripts, images, videos and other forms of unstructured data provide valuable information and nuanced business context AI systems need to generate accurate results, McKnight told TechTarget. Without that, he said, “AI initiatives risk delivering shoddy, ungrounded outputs.”
But Stephen Catanzano, a senior analyst at Informa TechTarget’s Omdia research and advisory division, said many organizations need to rethink their data management and governance strategies if they are to make trustworthy unstructured data available at AI speed — and avoid errors that proliferate at the same pace. When data quality, lineage or metadata shortcomings send AI astray, “the errors don't stay small,” he said. “They get amplified quickly as AI systems consume and act on that data at scale.”
Here are five things data leaders should keep in mind as they work to make unstructured data assets ready for effective use by AI models and agents.
1. Data teams often don’t know about all the unstructured data in their organization.
Many data teams don’t know exactly how much unstructured data their organization has or where it’s stored. In a survey conducted by analyst firm BARC in early 2026, only 29% of 255 respondents who are responsible for data, analytics or AI strategies said they knew where all the unstructured data potentially relevant to AI use cases resided in their systems. “You cannot govern what you cannot find,” BARC analysts Kevin Petrie and Merv Adrian wrote in a June 2026 report on the survey.
“You cannot govern what you cannot find.”
Kevin Petrie and Merv Adrian
BARC analysts, in a June 2026 report
Anne Marie Smith, principal consultant at Alabama Yankee Systems, said part of the problem is that unstructured data is commonly distributed across various platforms and repositories. That could include databases, data lakes and lakehouses, file systems, SharePoint sites, document management systems and more. In many cases, Smith said, organizations “don’t know what they have or why they have collected it.”
2. Making unstructured data accessible and usable is more complex than managing structured data.
The BARC survey also found that 70% of respondents said less than half of the unstructured data they did know about was discoverable and usable for analytics and AI applications. Most data teams “have significant work to do” to make the data AI-ready, Petrie and Adrian wrote.
“Unstructured data is where true competitive advantage now lies.”
William McKnight
President, McKnight Consulting Group
McKnight said unstructured data usually lacks consistent metadata and built-in quality guardrails, both of which must be added. In addition, he said Word documents, PDFs, emails and other forms of text require complex parsing, chunking and embedding workflows to add context, break them into more manageable pieces for AI tools and make them AI-searchable.
Donald Farmer, principal at consultancy TreeHive Strategy, said traditional data quality metrics such as accuracy, completeness and consistency aren’t enough with unstructured data. For example, there’s also whether AI training data is representative of the full datasets and free of embedded bias. Measuring those things can be a challenge for teams used to managing structured data, Farmer said.
3. Data governance programs can’t afford to ignore unstructured data anymore.
In a panel discussion during Dataversity’s recent AI Governance Online conference, Kelle O’Neal, founder and CEO of consulting firm First San Francisco Partners, said many data governance leaders previously talked about addressing both structured and unstructured data. “The reality was unstructured data was ignored,” she said. “It was a second-class citizen, as far as that goes. That can no longer be the case.”
Smith agreed that most organizations have applied limited — or no — governance to unstructured data until now. For example, she said they’ve rarely assigned data stewards to oversee it. They need to do so in the AI era, she added, and data governance programs must ensure there is broader accountability for managing unstructured data assets and establish appropriate controls on their use by AI.
While many companies do have longstanding information governance programs for documents, emails and other unstructured data, Farmer said those initiatives typically focus on retention policies, records management, e-discovery and other legal and compliance functions—not on governing the data itself. AI is forcing that to the fore, he said.
4. Adding unstructured data assets to data catalogs is a key step.
Data catalogs that provide a unified inventory of enterprise data assets are a key component of data management and governance programs. In BARC’s survey, though, only 38% of the respondents said they were cataloging unstructured data as part of efforts to prepare it for AI use.
That isn’t a recipe for success, Petrie and Adrian wrote: “Without a clear inventory, AI projects will lack the contextual inputs they need to perform reliably at scale.”
In a data catalog, teams can automatically collect metadata and data lineage documentation, classify sensitive data and apply access controls. McKnight said doing so supports semantic search and vector indexing capabilities that help AI models and agents find contextually relevant data. It also enables governance managers to trace AI outputs back to the upstream data sources to verify data freshness and ensure compliance with internal policies and regulations such as the EU AI Act, he said.
5. Fully consolidating unstructured data in a centralized platform isn’t practical in most cases.
Catanzano said the traditional approach of consolidating structured data in a data warehouse, lake or lakehouse for analytics uses “breaks down quickly with unstructured data” because its volume and variety make migration work complex and costly. Data sovereignty requirements and the fact that much of the data is stored in systems built for other purposes, such as document repositories or collaboration tools, further complicate consolidation efforts.
The more practical approach, Catanzano said, is a federated model in which unstructured data largely stays where it is and is made discoverable through a data catalog with a shared metadata and governance layer. For example, a data fabric architecture underpinned by a catalog is a potential option.
“That shift, from centralizing data to centralizing visibility and control over distributed data, is the real operational challenge most organizations are still working through,” he added.
Craig Stedman is a senior reporter on the TechTarget Data news team, covering analytics, data management and data backup topics. He has more than 40 years of experience as an enterprise IT writer and editor.
