Start Here

Field Guide

From Recommendation to Accountability

AI can predict, rank, optimize and recommend. But when a recommendation can move capital, safety or regulatory exposure, somebody still has to decide — and be able to explain why.

Five questions, one decision

Bring one real, live decision. Work through these in your own words.

01

What decision or action will change if we accept this recommendation?

What will someone actually do differently if we say yes?

02

What are we relying on — and what remains uncertain?

Which parts are evidence, which are assumptions — and where are we still using judgment?

03

What has to be true for this recommendation to be usable here?

What conditions make this recommendation usable here, at this level of consequence?

04

What would make us pause, challenge it, or stop relying on it?

What would make a reasonable person here say “not yet”?

05

Who can actually say yes, no, or not yet — and owns what follows?

Does the person accountable for the outcome actually have the authority to challenge or override it?

Recognition

Where did you hesitate?

You do not need five perfect answers. Look for the first question where the answer became vague, conditional, disputed — or depended on “we assume,” “usually,” or “someone will decide.”

The recommendation is clearer than the decision.

You are relying on more than you can easily name.

It works — but only if certain conditions continue to hold.

People can challenge it, but the stopping point is unclear.

Someone is accountable, but the authority is less obvious.

Start with the first answer you cannot make explicit.

Reliance is not binary. It expands as the conditions for relying become explicit and the evidence earns it.

Will the decision hold?

Stress-test one AI-influenced decision before it becomes load-bearing. One recommendation. One focused session. One concise Decision Readiness Brief.

Can You Defend This Decision? →

The Brief will tell you which kind of gap you are looking at — a vendor-side buyer-trust problem, or an enterprise consequential-decision problem — and route you accordingly.

Or keep working it yourself

These questions are yours to reuse on any live decision, with no obligation to bring them to us.

Revisit the five questions ↑

Want the next case? Subscribe to The Scale Gap.