TLDR: A recent survey by Dataiku and The Harris Poll indicates that 95% of senior data executives cannot fully trace AI decision-making, leading to significant concerns about governance, explainability, and trust in AI deployments. The report highlights a disconnect between C-suite expectations and the realities faced by data leaders, with many delaying AI projects due to explainability issues and fearing job loss if AI doesn’t deliver measurable results.
A new survey, ‘Global AI Confessions Report: Data Leaders Edition,’ conducted by Dataiku in collaboration with The Harris Poll, has revealed a critical challenge in the adoption of artificial intelligence: a vast majority of data leaders struggle with tracing AI decisions. The report, based on a study of 812 senior data executives across eight countries including the U.S., U.K., France, Germany, UAE, Japan, Singapore, and South Korea, found that a staggering 95% admit they cannot fully trace AI decision-making processes.
This lack of traceability is raising significant concerns regarding governance, explainability, and overall trust in AI deployments. The survey highlights that 52% of data leaders have either delayed or completely blocked AI agent deployments due to worries about explainability. Furthermore, only 19% consistently require AI agents to ‘show their work’ before granting approval for their use.
The findings underscore a prevailing sentiment that accuracy alone is not sufficient. A substantial 80% of respondents consider an accurate but unexplainable AI decision to be riskier than a wrong but explainable one. This indicates a strong preference for transparency and understanding over blind trust in AI outputs.
The stakes are particularly high for CIOs and CDOs. While 46% are most likely to receive credit for successful AI initiatives, a higher percentage, 56%, are most likely to be blamed for business losses resulting from AI failures. The pressure is palpable, with approximately 60% of data leaders expressing fear that their jobs are at risk if AI does not deliver measurable results within two years.
Practical challenges with AI are already evident. The report notes that 59% of data leaders have experienced business issues in the past year due to AI hallucinations or inaccuracies. Despite this, 82% believe AI can outperform their boss in analytical tasks, though 74% would revert to human oversight if AI errors exceed a 6% threshold. Moreover, 89% of data leaders identified at least one business function they would never delegate entirely to AI.
The survey also exposes a significant disconnect between C-suite executives and data leaders regarding AI understanding and expectations. Only 39% of data leaders believe their C-suite fully comprehends AI. A majority, 68%, feel that executives overestimate AI’s accuracy, and 73% believe executives underestimate the complexity involved in achieving AI reliability before it can be deployed in production.
Also Read:
- CIOs Chart Course Through Evolving AI Regulatory Landscape in 2025
- Enterprise Security Leaders Unprepared for Surging AI Agent Risks, NeuralTrust Report Reveals
Florian Douetteau, Co-founder and CEO of Dataiku, commented on the findings, stating, ‘An alarming revelation of the report is that enterprises worldwide are betting on AI they don’t fully trust. The good news is most failed AI initiatives suffer from common blockers that can be overcome with more explainability, traceability, and governance.’ This quote emphasizes the need for improved frameworks to build confidence and effectiveness in AI adoption.
