IBM published a survey of AI leadership authority on November 17. Only 18% of respondents say they fully control AI investment decisions.

Conducted with Oxford Economics, it surveyed 1,667 executives across 33 geographies and 20 industries during April–June 2026, covering different AI leadership roles.

What the comparison shows

The group reporting stronger outcomes says 63% of projects deliver measurable results, versus 43% among others. Self-reporting and analytical grouping do not establish the cause of that difference.

The useful observation is a potential mismatch between responsibility and decision-making authority. A person can be accountable for performance while budgets, workflow design and the final decision to stop a project belong to other participants. That relationship is what the research brings into focus.

Turning a dashboard into a decision

Our editorial assessment is that measurement becomes useful when its consequences are defined in advance. In an illustrative document-processing project, a team might track accuracy, human review time and cost per accepted result. If automated processing becomes faster while verification takes longer, the overall benefit may differ from what a demonstration suggests.

That is a hypothetical scenario, not a customer example from the survey. It illustrates why intervention thresholds should precede the results: who can reduce scope, request additional checks or end the trial? Without an answer, reporting remains a description of the problem rather than a mechanism for responding to it.

Another useful distinction separates model quality from deployment quality. A correct answer can still enter the wrong workflow. Conversely, a well-designed process can contain the consequences of a predictable error. Neither assessment substitutes for the other.

A team could also specify which evidence would justify expanding a successful pilot. More usage alone is not necessarily more value. If the accepted result is clearly defined, the team can compare performance without changing the goal after seeing the numbers. This is an editorial recommendation, not a causal finding attributed to the study.

The survey does not provide a universal instruction to transfer every budget to one job title. It offers a reason to make decision rights around AI explicit. The more useful follow-up is not simply how many projects an organization has, but who can change the decisions when the evidence changes. That is where a dashboard can become a management tool.