Organizations are spending billions to prepare their data for AI, but all that money isn't buying what they really need: trust.
According to research from Fortune Business Insights, the worldwide data mining tools market will hit $1.44 billion in 2026 and grow to $3.49 billion by 2034. During the same time frame, the global data governance market will grow from $5.38 billion in 2026 to just over $24 billion. Despite the projected growth in these markets, most organizations' data quality remains subpar. The issue isn't a lack of data or the inability to find it; the challenge is understanding its history, context and other characteristics practitioners need to evaluate data quality and trustworthiness.
"Finding the data doesn't mean the same as that data being understood," said Tim Anderson, senior managing director at FTI Consulting.
He said that too often, data users don't have the necessary context and definitions. For example, they may not know how "active customer" is defined and whether the definition is used across all datasets within their organization.
A survey by Modern Data Company identified missing or outdated documentation, poor naming conventions and inadequate discovery tools as common obstacles. Even when data practitioners and their organizations can find the data assets they need, they still lack sufficient context to use them confidently.
"Most organizations have data [usable] for dashboards, reporting and advanced analytics. It's good data that falls short for AI models that need AI-ready data to search across the enterprise, interpret business meaning and act within policy," according to a report by global consulting firm KPMG.
Data discovery is no longer sufficient
Data discovery is the process of collecting, mapping and exploring structured and unstructured data to understand what information is available and where it resides.
These related practices – once known as "data mining" and "knowledge discovery in databases" – fall under the broader umbrella of "data discovery." It has become an essential discipline for any organization using business intelligence and other analytics tools.
But in the AI era, knowing what information exists and where to find it is just the starting point.
"Everyone is now realizing that data governance is not a plus but a requirement," said Fabien Cros, chief data and AI officer with Ducker Carlisle, a global management consulting firm.
To use a data asset for AI, an organization must have readily available and accurate information about its ownership, definitions, documentation, lineage and usage history. This information enables data practitioners to evaluate a dataset's accuracy, completeness, consistency, timeliness, validity and uniqueness – known as the six dimensions of data quality – to determine whether it is trustworthy and fit for use.
KPMG's report said AI-ready data has three requirements:
- It must be searchable, allowing the AI engine to find and connect the right data;
- It must be contextual, so that AI can understand meaning instead of simply retrieving records; and
- It must be trusted, with clear lineage, machine-readable permissions and decision logic that makes AI actions traceable and auditable.
Historically, most organizations did not build their data environments with these requisites in mind. Instead, cybersecurity, privacy and regulatory compliance typically drove how organizations organized, stored and controlled access to their data, according to Nader Henein, vice president analyst at research firm Gartner.
"There was no overarching [call, saying] 'Let's label all our data information,'" Henein said.
The work involved with cataloging and labeling data across an enterprise was very expensive and resource-intensive. And because it wasn't typically necessary for business intelligence, analytics and machine learning use cases, organizations had difficulty demonstrating ROI.
"So companies didn't do it, or they only did it on data they wanted to mine," he said.
Today, many organizations know they need better metadata to create organized, searchable and usable data assets. But the sheer volume of data and the speed at which new data is created often overwhelm improvement efforts, Anderson said.
"The data is changing so rapidly, it's hard to get a real-world point-in-time picture into what all that looks like," he said. "It's difficult to stay on top of it from an ongoing data management perspective."
Maturing metadata management
Many organizations still lack the data governance and metadata management practices required to create AI-ready data.
A November 2025 study from the IBM Institute for Business Value found that while 81% of chief data officers prioritized investments to accelerate AI initiatives, only 26% were confident their data could support AI-driven revenue. The 2026 AI Impact Survey by consulting firm Grant Thornton, which polled 950 business leaders, found that "insufficient data readiness" is the third most common reason for AI project underperformance or failure, at 23%.
Organizations are taking steps to clean and label more of their data. Henein said most are doing so with the data assets they've identified as required for the specific AI initiatives they're pursuing.
That's because they can now see the financial benefits of doing so.
"In the past it was very hard to make a business case around data governance and improving data quality. But with AI, we can now quantify the ROI," Cros said.
Not all data will be refined; instead, organizations will focus on preparing the data needed for specific AI endeavors.
"It will continue to be task-oriented," Henein said, adding that the cost of making all data within a typical organization AI-ready remains an overwhelmingly challenging, expensive and resource-intensive task. "But as more and more tasks develop, valuable information will float up to the top and get organized and labeled."
Henein said he expects organizations will make significant progress in refining their data as vendors improve supporting technologies and offer them as standard services rather than today's expensive, limited add-ons.
"Eventually it's going to be folded into the standard offering, and organizations will turn it on and get better information on the data they have," he said. "We can expect to see this as a commoditized feature by 2030."
These advancements won't be enough on their own to improve metadata management and data governance, Cros said. Organizations must also implement processes and practices, which include assigning data owners, data stewards and data champions.
Mary K. Pratt is an award-winning freelance journalist with a focus on covering enterprise IT and cybersecurity management.
