全球AI坦白报告：数据领导者版

Global AIConfessions Report:Data Leaders Edition Agents in Action,Trust in Question Executive Summary The Global AI Confessions Report: Data Leaders Edition from Dataiku,based on a Harris Poll survey of 800 data leaders worldwide, reinforcesa key notion: AI agents are no longer experimental. A staggering 86% saytheir organizations now rely on agents in daily operations, with nearly Two-thirds of data leaders (64%) say their company’s AI agentsare better at automating operational tasks than at makinganalytical, higher-order business judgements.This reveals aserious lag in operationalizing AI for critical business decisions. Why?Data leaders admit to a pervasive lack of confidence in explainability admit significant gaps in trust, explainability, and readiness thatcombine to inhibit AI performance and impede rollouts at scale. their work,”and the vast majority, 95%, shockingly admit they couldnot fully trace AI decisions end-to-end if they were asked to provide thisreasoning to regulators. And the stakes are personal: CIOsand CDOs are most likely to be function, including sensitive, high-stakes functions like hiring,compliance, or ethical decisions.That disconnect highlights the coreissue: While AI is fast becoming the default for automating repetitivework, leaders lack the conviction it can be trusted with critical business Trust Is on TrialThe most telling signal comes from a single stat:75% of data leaders say trust in Performance Bottlenecks their AI agent deployments is a concern. Are AI agents stalling out?Half of data leaders say fewer than half of their agentsmake it beyond POC,a fragility echoed by MIT research showing 95% of GenAIpilots fail to deliver ROI. The problem isn’t the models, it’s the strategy. Three in fourdata leaders (75%) admit their AI efforts are driven by tech ambition, not business of data leaders say trust intheir AI agent deployments isa concern. 75% What Comes NextThis report surfaces a candid truth: Data leaders are pushing AI into the heart of daily acknowledge that AI-generated business suggestions carry more weight than thoseprovided by human employees, even as they question the accuracy of those same business, but they’re doing so with limited control and shaky confidence. And that riskis far from hypothetical:59% of data leaders have already faced a business issueor crisis stemming from AI hallucinations or inaccuracies in just the past year. their own boss,even as doubts about explainability persist. Yet, reluctance to trusthasn’t slowed deployment.Even though52%admit to delaying or blocking anagent rollout over explainability concerns,many data leaders feel they can’t affordto wait for perfect accuracy. With competitors racing ahead, the urgency to stay ahead faced a business issue or crisisstemming from AI hallucinations orinaccuracies in just the past year. outcomes, organizations will never unlock its full impact. The pressure is mounting:77% of data leaders already believe a competitor has deployed a stronger AI strategy than their own company.The data leader mandate is clear: Reinforce trust,sharpen explainability, and ensure AI agents don’t just work fast, but work right. explainability concerns. Diving Into Data Leaders’ GlobalAI Confessions This report captures unfiltered admissions from the people on the front lines of AI execution:data leaders themselves. Their confessions reveal what’s really driving (or derailing) enterpriseAI: performance gaps, hidden risks, and the uneasy truth about trust, reliability, andexplainability of AI agents. From shadow AI and governance blind spots to what’s working at released in March 2025, to highlight where executives and data leaders align and where theirperspectives diverge. Finally, we break down regional results across the U.S., U.K., Germany,France, the United Arab Emirates, APAC, and Japan, revealing the unique pressures and responsible and most accountable for both driving AI forward and confronting the practicalbarriers that will define whether AI in 2026 is a story of trust and control or one of blind spotsand missed potential. Ultimately, the winners and losers will be determined very publicly on I Let AI Agents Make CriticalI MUST CONFESS... Business Decisions, Even ThoughI Don’t Fully Trust Them decision or answer from an AI agent because it lacked a clearexplanation, and two inthree(65%) have questioned anagent’s decision outright. in the loop for their existing AI agents,so answeraccuracy often becomes the sole safety net. AI decision is more dangerous to their organization than a wrong but traceable one. leaders would allow an AI agent to make a critical business decisionwithoutexplanationof how the result was reached, and 81% say they’d stake their jobs onthose calls. Data leaders may lean on AI for speed and scale, but without explainability, oversight,or guardrails, every decision carries hidden risks that can derail their AI success asmuch as their company's performance. G