Nearly all (95%) data leaders admit they lack full visibility into AI decision-making, according to a report from Dataiku.
This was based on a study conducted in August 2025 by The Harris Poll for Dataiku, surveying 812 senior data executives across the United States, United Kingdom, France, Germany, United Arab Emirates, Japan, Singapore and South Korea. Respondents work for large companies with an annual revenue of or more than $1 billion or regional equivalents.
The report reveals that, while 86% of respondents estimate AI has become embedded into their daily operations, concerns about governance gaps, lack of explainability, and misplaced confidence remain.
Only 19% of data leaders always require AI agents to “show their work” before approval, and 52% have delayed or completely blocked agent deployments due to explainability concerns, specifically.
CIOs and CDOs carry the weight of responsibility: 46% are most likely to be credited for AI gains, but over half, 56%, are most likely to be blamed for business losses due to failed AI.
With 60% fearing their own jobs are at risk if AI doesn’t deliver measurable results within two years, the stakes for data leaders have never been higher.
Despite optimism about potential, AI is already causing headaches for global enterprises, as 59% report that AI hallucinations or inaccuracies have already caused business issues in the past year.
Also, 82% believe AI can outperform their boss in business analysis, but 74% would revert to human-managed processes if AI error exceeds just 6%.
Further, 89% say there is at least one business function they would never delegate to AI.
An earlier report from Dataiku showed how bullish CEOs are on AI. Data leaders, however, are more skeptical. Only 39% say their C-suite truly understands AI, 68% believe executives overestimate its accuracy, and 73% say they underestimate the difficulty of achieving AI reliability prior to production.
The stakes are high, with 56% of data leaders expecting a CEO will be ousted by 2026 due to a failed AI strategy. This disconnect, CEO optimism versus data leader caution in pushing unfit AI into production, could explain why so many projects remain stuck in POC.
“An alarming revelation of the report is that enterprises worldwide are betting on AI they don’t fully trust. The good news is that most failed AI initiatives suffer from common blockers that can be overcome with more explainability, traceability, and governance. That’s how AI moves from hype to real business impact,” said Florian Douetteau, co-founder and CEO of Dataiku.
