72% of AI Decision-Makers Cite Data Foundations as Root Cause

72% of AI decision-makers identified poor data foundations as the root cause when AI initiatives failed to meet expectations. This finding comes from a survey commissioned by software company Collibra with Harris Poll, covering 300 data management, privacy, and AI decision-makers, released in September 2026.

What the Survey Reveals

According to the survey, decision-makers are consequently restructuring their operating models to align intelligence with data governance. The breakdown by category is as follows.

• 53% of decision-makers are shifting AI function reporting lines closer to their data organizations.
• 58% of companies are focused on clarifying internal accountability for AI outputs.
• 87% of decision-makers said their teams regularly verify that the contextual information used by AI agents is accurate and up to date.
• More than half of respondents said employees spend hours each week reviewing and correcting AI agent outputs.

How Agents Use Data Differently

Felix Van de Maele, co-founder and CEO of Collibra, said companies need to know which agents are running, who owns them, what data and systems they can access, and what they're authorized to do. Without that visibility, governance gaps only surface after an incident occurs.

He explained that enterprise data has traditionally been structured for humans reading dashboards and running analyses. When data looked off, people would question it and fill in context that the data itself didn't capture. Agents change that premise. Even when definitions are ambiguous or context is missing, an agent will still produce an answer and act on it with confidence.

So the task isn't to clean the data further, he said, but to make it clear—in a form machines can understand and use—what the data means, whether it can be trusted right now, and what it's permissible to do with it. Concrete approaches he mentioned include aligning definitions across the business and translating policies into code that can be checked at the moment an agent acts.

The scope of access itself is also a constraint. According to an August report from Google Cloud and MIT Technology Review Insights, AI currently has access to an average of only 45% of enterprise data, and only half of organizations trust that AI agent outputs are appropriate and accurate.

Regulatory readiness also featured in the survey. In the Collibra/Harris Poll survey, nine out of ten business leaders said they are actively preparing for AI regulations in the US and globally, and 51% of enterprise leaders said they are investing in data lineage and documentation.

Implications for AX at Korean Enterprises

The bottleneck this survey points to isn't model performance—it's data definitions and authority. Before scaling up pilots for agent adoption, there are things worth checking first.

• Whether the same metric names carry the same definitions across departments
• Whether the scope of data and systems agents can access is formally documented
• Whether it's organizationally clear who corrects errors and who is accountable when outputs are wrong

Start by putting together a single page listing all agents in operation along with each one's owner, access scope, and authorized actions. If that page can't be filled in, the priority isn't cleaning up data—it's clarifying accountability.

Source: AI failures often trace back to poor data foundation: survey