Six Questions to Ask Before Approving an AI Project

The hardest part for a decision-maker is judging whether a proposal is sound. These six questions require no technical background, yet they filter out most vague proposals.

1. Who does this today, and how long does it take?

If nobody can answer, the current state has not been measured, and any promised improvement lacks a baseline. Measure first, then discuss improvement.

2. What observable change will show it worked?

Insist on an observable outcome rather than adjectives like "more efficient." For example, "nobody compiles this sheet by hand again" is testable.

3. Which steps will still need human confirmation?

An honest proposal will name the places where human review stays. Claims of fully automatic operation with no human involvement usually mean failure cases have not been considered.

4. What happens when it goes wrong, and who notices?

Look at failure modes, not just the happy path. There must be a defined fallback and an owner, or problems accumulate until they surface expensively.

5. Where does the data live, and who can see it?

Where customer or internal information is involved, settle storage location, access scope and retention period up front — before work starts, not after launch.

6. If we stop in three months, what does it cost?

Exit cost defines your safe margin for experimentation. Whether material can be exported and whether the process can fall back to manual determine how much risk this commitment carries.

Summary

All six questions serve one purpose: converting vague expectations into testable agreements. Proposals that answer them clearly tend to succeed more often.