全球AI报告：数据领导者版（英）

Global AI Confessions Report: Data Leaders EditionDATAIKU/HARRIS POLL SURVEYExecutive SummaryThe Global AI Confessions Report: Data Leaders Edition from Dataiku, based on a Harris Poll survey of 800 data leaders worldwide, reinforces a key notion: AI agents are no longer experimental. A staggering 86% say their organizations now rely on agents in daily operations, with nearly half (42%) embedding them so deeply that dozens of core processes depend on them.Yet beneath this rapid adoption lies a revealing tension: Leaders admit significant gaps in trust, explainability, and readiness that combine to inhibit AI performance and impede rollouts at scale.Agents in Action, Trust in QuestionTwo-thirds of data leaders (64%) say their company’s AI agents are better at automating operational tasks than at making analytical, higher-order business judgements. This reveals a serious lag in operationalizing AI for critical business decisions. Why? Data leaders admit to a pervasive lack of confidence in explainability and accuracy of answers generated by their AI systems and apps, because they either don’t require or simply cannot explain how their AI generated its outputs. Only one in five (a mere 19%) always requires agents to “show their work,” and the vast majority, 95%, shockingly admit they could not fully trace AI decisions end-to-end if they were asked to provide this reasoning to regulators.In fact, just 11% would consider AI agents for any business function, including sensitive, high-stakes functions like hiring, compliance, or ethical decisions. That disconnect highlights the core issue: While AI is fast becoming the default for automating repetitive work, leaders lack the conviction it can be trusted with critical business decisions where the technology can become truly transformative.And the stakes are personal: CIOs and CDOs are most likely to be credited for gains when AI delivers ( 46% ) and even more likely to be blamed when it fails ( 56% ).KEY FINDINGS46%56%DATAIKU1GLOBAL AI CONFESSIONS REPORT: DATA LEADERS EDITIONTrust Is on Trial That worry runs parallel to a growing reliance on the technology. Nearly 70% of leaders acknowledge that AI-generated business suggestions carry more weight than those provided by human employees, even as they question the accuracy of those same outputs.In fact, 82% of data leaders believe AI can provide more accurate analysis than their own boss, even as doubts about explainability persist. Yet, reluctance to trust hasn’t slowed deployment. Even though 52% admit to delaying or blocking an agent rollout over explainability concerns, many data leaders feel they can’t afford to wait for perfect accuracy. With competitors racing ahead, the urgency to stay ahead often outweighs hesitation about reliability.Are AI agents stalling out? Half of data leaders say fewer than half of their agents make it beyond POC, a fragility echoed by MIT research showing 95% of GenAI pilots fail to deliver ROI. The problem isn’t the models