A five-step routine for reviewing AI-assisted decisions.
THINK helps a named human frame the decision, challenge the recommendation, own the call and know when to escalate before approving AI-assisted work.
What THINK solves
“Human review” is too vague to guide a consequential decision. One manager may check the numbers, another may inspect the reasoning, and another may simply add a signature. THINK gives reviewers the same five moves without pretending a checklist can make the decision for them.
The five moves
T — Tune In. Regulate before you prompt. A reviewer in reactive mode — rushed, defensive, impressed — rubber-stamps. The first move is noticing which one you are.
H — Hypothesize First. Define the real decision before reading the AI’s answer. If you meet the recommendation before you’ve framed the question, the recommendation frames it for you.
I — Interrogate Everything. Evaluate like a decision scientist: what would have to be true, what’s the strongest case for the option the AI did not choose, and which single condition would flip the call.
N — Narrow to the Call. Make the one defensible decision — and say out loud who owns it. Not “the committee approved it.” A name.
K — Know How to Land It. Deliver it with authority: hold the position under challenge, name the reasoning, and know in advance when you’d escalate instead.
Where it shows up
Every PAVNESS workshop uses THINK inside the simulation, not as a slide. Your team runs it while a constructed decision is under pressure, then applies the questions to a real category of work in its own words. By the end of the room, the four anchors—what AI may do, who owns the call, what reasoning stays visible and when to escalate—have been answered out loud by your people.
What it is not
THINK is not software, not a certification, and not a compliance artifact — we don’t issue seals and we don’t attest to anything. It is a working routine, and the way we prove it works is simple: run 30 seconds of the method yourself, free, before you talk to anyone.