If You Have to Explain It Tomorrow, You Think Differently Today

A sales manager is looking at a discount recommendation. The AI tool estimates a 78% chance of closing before quarter end with a 12% concession, and 43% without it.

Two versions of this decision are possible.

Two versions of the same decision

In the first, the discount is applied unless the manager overrides it.

In the second, the manager can approve it, but tomorrow at 9:00 she will walk the CFO through the evidence behind the decision and explain why she approved it.

The recommendation is identical. So are the model, the deal and the economics.

What changes

In the second version, different questions are more likely to appear.

Where does the 78% come from? Did the customer actually ask for a discount? Is price the obstacle, or is the deal stalled for a reason a discount won't fix? What does this customer learn about quarter-end concessions?

Nothing about the recommendation changed. What changed was the expectation that someone would have to defend it. That expectation changes what people do before they decide, and that's the point at which judgment either happens or doesn't.

Accountability is not merely the consequence of judgment. It is one of its conditions.

This only works if explanation means examination, not justification after the fact. If people learn that the job is simply to defend whatever they approved, accountability turns into theatre. The useful question is not “Can you make a case for it?” Almost anyone can. It is “What did you test before you accepted it?”

The friction AI removes

AI removes friction from decisions. Research takes seconds. Recommendations arrive immediately. The explanation is already written and the conclusion is already polished.

Most of that is welcome. A great deal of what slowed decisions down was waste.

But some of it wasn't. Having to build the case yourself used to show you where it was thin. Having to explain the evidence forced you to separate what you knew from what you were assuming. When a recommendation arrives finished, the decision-maker hasn't had to build the reasoning, so some of its weak spots can remain invisible.

The leadership question isn't how to add friction back. It's which friction was waste and which was making people think.

Friction belongs where the consequences are

Nobody needs to defend the subject line AI suggested for a follow-up email. Demanding that would just slow people down and teach them to resent the process.

But a discount, a hire, a forecast, a resource decision: the more consequential the decision, the less acceptable borrowed confidence becomes. That's where someone should expect to answer for it.

What leaders can do

Before a consequential AI-supported decision gets acted on, ask the person proposing it:

“If I asked you tomorrow why this was the right decision, could you show me the reasoning you were willing to own?”

Not why the model recommended it. Why you accepted the recommendation.

If they can, the AI did its job. It helped someone think.

If they can’t, nobody has made the decision yet. It has only been passed along.

Judgment Labs help teams examine where AI-supported decisions need more than a recommendation and an approval. Explore how we can work together →