WCRC Analysis
An executive can ask AI to examine a proposal in seconds. The harder task is explaining why its answer deserves to influence a decision that affects other people.
That distinction matters in light of Charter's AI Pulse, released on 21 October. The research reports that executives are roughly twice as likely as employees to trust AI over their own judgment. Its introduction describes a survey of 5,002 US desk workers, managers and executives who use AI, alongside 91 additional interviews. These are reported attitudes among AI users, not an audit of boardroom decisions. Charter research introduction.
The finding does not establish that executives are making worse decisions. It does create a useful question for every organisation adopting these tools: what must remain the leader's responsibility when part of the reasoning comes from a model?
A recommendation needs a responsible owner
WCRC's view is that accountability becomes meaningful at the point where someone can explain the choice without leaning on the authority of the tool that suggested it.
Consider a hypothetical expansion proposal. AI could organise market information, compare scenarios and identify assumptions worth testing. The executive still needs to establish whether the underlying demand estimate is sound, whether the company can absorb a delayed return, and whose interests the proposed trade-off serves. A fluent recommendation cannot settle those questions on its own.
The distinction should appear in the decision record. Who checked the evidence? Which assumption would change the recommendation? Who accepted the remaining uncertainty? Recording these answers gives colleagues something concrete to challenge before commitment and something useful to revisit afterwards.
An AI-generated rationale copied into a presentation is a poor substitute. It may explain the recommendation elegantly while leaving nobody responsible for testing it.
Challenge should match the consequence
A leader does not need a committee to review every use of AI. Turning a meeting agenda into a clearer document calls for a different level of scrutiny from evaluating a plant closure or recommending a candidate for promotion.
The organisation should therefore decide which consequences require additional review. The criteria can include the effect on people's opportunities, the financial exposure and how easily the decision can be reversed. For consequential choices, an independent reviewer should inspect the underlying evidence rather than merely ask the same tool to produce another answer.
This is WCRC's proposed management practice, not an intervention tested by the Charter survey. It is consistent with the broader purpose of the US National Institute of Standards and Technology's voluntary AI Risk Management Framework: incorporating trustworthiness into the design, use and evaluation of AI systems. NIST AI Risk Management Framework.
The cost of making disagreement difficult
A further risk arises when a model's answer arrives with the authority of the most senior person in the room. A team may find it difficult to question the output because doing so also means questioning the leader who chose it.
Executives can reduce that pressure by making uncertainty visible. They can show the evidence they verified, identify what the model could not establish and invite a colleague to make the strongest case against the preferred option. Disagreement then has a defined job in the process.
Charter's public findings also report a gap between managers' accounts of the support they provide and employees' recollections of receiving it. That reinforces the need to examine what teams experience, while leaving the causal explanation open. Charter AI Pulse findings.
For leadership teams, the useful review question is straightforward: could another person reconstruct this decision from the evidence and assumptions we have kept? If the answer is no, faster analysis has left an unfinished management task.
The executive who signs off on a decision must remain able to defend it, amend it and take responsibility when the evidence changes.