The Four Human Inputs
What is actually happening here: the politics, the history, the unwritten rules. AI knows only what you typed.
Who trusts whom, who will champion or resist, the human reality the decision lands in.
The responsibility for the outcome. This is yours and cannot be delegated to a tool, ever.
Whether this can actually be executed given your real constraints: capacity, team, budget, timing.
The Calibration Routine
AI is more reliable on common, well-documented topics. Less reliable on novel, niche, or interpersonal situations specific to your organization.
The higher the stakes, the more verification it needs before you act on it. Match scrutiny to consequence.
Time pressure is exactly when automation bias strikes. Flag it and slow down deliberately. Rushing past this question is the mechanism of the bias itself.
Sam's story
Sam is deciding how to restructure responsibilities on the team and gets a clean, confident AI recommendation. Sam runs the calibration: the domain is this specific team (not well-documented for the AI), the stakes are high (people's roles), and Sam is rushing before a deadline: all three flags up. Then Sam checks the swing factor and realizes it is relationship dynamics. One of the people the AI proposes to shift responsibilities onto is already overloaded and quietly looking to leave, which the AI had no way to know. The recommendation looked complete and was missing the one thing that mattered. Sam supplies it and decides differently.
Apply It Today
On your next AI-assisted decision, run the three-question calibration out loud, then name the one human input that is the swing factor. Make sure it is supplied by a human.
Reflect