HALDEMAN Notes · 03 ·

Agree on the evidence before debating the number.

Set the rules first. The argument gets smaller.


Most AI arguments aren’t about the number. They’re about the standard: what counts, measured how, compared to what. Settle that after the numbers arrive, and everyone argues backward from the answer they want.

Self-grading makes it worse. The University of Melbourne and KPMG found 57 percent of employees who use AI at work say they have used it without being open about it. What leadership hears is not what’s happening.

Sometimes the room still disagrees. Then the sponsor picks the standard for this decision, and the other view is written beside it. A measure nobody can define yet goes on the unknowns list, with an owner and a date. The decision then goes ahead on the sponsor’s judgment, within agreed limits: the evidence it needs, the downside it can carry, and a gate where it can be undone.

So: before any disputed number is pulled, agree in writing which sources count, how each measure is defined, and what result triggers which decision. One meeting. One page. Signed. If it won’t sign, the sponsor decides, and the dissent stays on the page.

Sources checked October 2026

Source

KPMG and the University of Melbourne, Trust, attitudes and use of artificial intelligence: A global study 2025. Read the original

See how findings get challenged before they’re final Ask Albert: how do we agree on the evidence? Next note
Ask Albert