AI is part of the work.
The responsibility
is still yours.

You don’t need a government contract to encounter that question.

Consider an illustrative situation. Your leadership team is reviewing a proposal to expand an AI pilot. The recommendation is clear, the financial case is attractive, and the brief anticipates the obvious objections.

You ask what supports the projected saving.

The executive presenting it points to a benchmark in the appendix. Someone on the team found it. The supplier’s presentation used a similar figure. Finance has checked the arithmetic.

You still need to know whether the benchmark describes work sufficiently like yours. The calculation may be correct while the assumption carrying it remains untested.

This is where I would slow the conversation.

A document can become more persuasive as it passes through an organization. Each person improves something: the analysis, the wording, the presentation. By the time it reaches you, it can be difficult to see which parts were independently established and which parts acquired credibility through repetition.

Your approval gives the recommendation another kind of authority. It also gives the organization permission to act.

Before that happens, stay with the person making the recommendation.

“What persuaded you?”

Give them time to answer. They may explain why a particular piece of evidence matters. They may identify a trade-off they believe is acceptable. They may also discover that the brief is more definite than their own judgment.

That discovery is useful, provided the room can tolerate it.

If uncertainty is treated as evidence of poor preparation, people learn to arrive with smoother explanations. You can end up with greater confidence in the meeting and less understanding of the decision.

Try it on one decision this week.

Ask its owner to identify the claim most likely to change the decision if it proved wrong. Then ask for the original evidence behind that claim and a short explanation of why it applies here.

Suppose a proposal says an AI service will reduce handling time by 30 percent. Finding the source of that figure is the beginning. You still need to understand what was measured, what work was excluded, and whether the comparison includes the time people spend checking and correcting the output.

Those are questions about the decision you are making.

You may conclude that the evidence is strong enough to proceed. You may approve a smaller test or require another check. You may decide that the opportunity warrants accepting some uncertainty.

Whichever choice you make, name the uncertainty you are accepting and the evidence that would cause you to revisit it. That leaves the executive with a recommendation they can own, and you with an approval you can explain.

Before the next brief enters the meeting pack, try this question:

“Which claim are we relying on most, and who has checked that it deserves the weight we’re putting on it?”

The answer may be straightforward. If it isn’t, you have found work worth doing before the decision travels any further.

Donal

This briefing develops a theme explored in The AI Didn’t Do It: how leaders establish understanding and ownership when AI participates in the work.