The Confidence Gap
Leadership believes AI oversight is working. The people who use AI tools every day, largely, do not agree. In a survey of 770 professionals across healthcare, finance, and manufacturing, managers and executives were significantly more confident than individual contributors on six of the ten most consequential dimensions of AI governance — sometimes by more than twenty points on the identical statement. The gaps cluster tightly around verifiability: whether documentation of human review could actually be produced, whether someone could explain how an AI tool reaches its outputs, whether training changed real behavior rather than just getting completed. Every gap reported as significant held under statistical testing.
The report goes well beyond establishing that the gap exists. It tests the most obvious objection directly — that the gap might just reflect managers' greater familiarity with AI governance jargon rather than a genuine seniority effect — and finds the gap survives the test, holding at nearly identical size whether or not a respondent recognized the terminology. It tracks whether organizational experience with AI closes the gap over time, finding that confidence does rise with tenure, but unevenly: verification and vendor-diligence habits improve with practice, while ownership clarity and explainability do not, exposing two different problems that are usually mistaken for one. And rather than stopping at a topline summary, the report breaks the gap apart sector by sector, publishing the complete ten-statement breakdown for healthcare, finance & insurance, and manufacturing & operations, each with its own within-sector seniority split, representative quotes, and sector-specific implications — before pulling those threads back together to identify, by name, which specific segment of professionals is most confident in their organization's AI oversight, and which is least.
The stakes are not abstract. A confidence gap this size and this consistent means the evidence a regulator, auditor, or incident review would ask to see — proof of review, a verified vendor claim, a documented and closed-out incident — is often not something leadership can actually produce, even when leadership believes it can. The report closes with four specific, evidence-backed moves that separate organizations that treat AI oversight as a checkable practice from organizations that treat it as a policy document, and maps each of the report's core findings to the specific diagnostic question it raises for any organization reading it.
Leadership believes AI oversight is working. The people who use AI tools every day, largely, do not agree. In a survey of 770 professionals across healthcare, finance, and manufacturing, managers and executives were significantly more confident than individual contributors on six of the ten most consequential dimensions of AI governance — sometimes by more than twenty points on the identical statement. The gaps cluster tightly around verifiability: whether documentation of human review could actually be produced, whether someone could explain how an AI tool reaches its outputs, whether training changed real behavior rather than just getting completed. Every gap reported as significant held under statistical testing.
The report goes well beyond establishing that the gap exists. It tests the most obvious objection directly — that the gap might just reflect managers' greater familiarity with AI governance jargon rather than a genuine seniority effect — and finds the gap survives the test, holding at nearly identical size whether or not a respondent recognized the terminology. It tracks whether organizational experience with AI closes the gap over time, finding that confidence does rise with tenure, but unevenly: verification and vendor-diligence habits improve with practice, while ownership clarity and explainability do not, exposing two different problems that are usually mistaken for one. And rather than stopping at a topline summary, the report breaks the gap apart sector by sector, publishing the complete ten-statement breakdown for healthcare, finance & insurance, and manufacturing & operations, each with its own within-sector seniority split, representative quotes, and sector-specific implications — before pulling those threads back together to identify, by name, which specific segment of professionals is most confident in their organization's AI oversight, and which is least.
The stakes are not abstract. A confidence gap this size and this consistent means the evidence a regulator, auditor, or incident review would ask to see — proof of review, a verified vendor claim, a documented and closed-out incident — is often not something leadership can actually produce, even when leadership believes it can. The report closes with four specific, evidence-backed moves that separate organizations that treat AI oversight as a checkable practice from organizations that treat it as a policy document, and maps each of the report's core findings to the specific diagnostic question it raises for any organization reading it.