The 10 Questions Before Any AI Project

Why this matters

These questions are a gate. If you cannot answer them clearly, you are not ready to build or buy. They are the difference between a project and a purchase order with hopes attached.

What you will be able to decide after this

• Whether a proposal is ready to proceed, needs prerequisites fixed, or should be killed.
• Which questions to ask the next person who brings you a proposal.
• Why "proceed" is a stronger word than "try it."

Core lesson

The ten questions:

1. Can a human already do this job reliably with the data you have today?
2. Do you make this decision or perform this task often enough to justify the cost?
3. What does success look like in 90 days — in numbers, not slogans?
4. Who owns this when it breaks?
5. Can you measure accuracy or usefulness in production?
6. What happens when the AI is wrong?
7. Would a simple rule or process solve most of the cases?
8. Do you have real support for the actual timeline (not the optimistic one)?
9. Is the data clean and traceable enough to trust?
10. Can you staff the monitoring and maintenance after launch?

Question three is where proposals die. "Faster" is not a number. "Better" is not a number. If the answer to question three is a slogan, the project is a hobby.

Question ten is where they die quietly. The launch is exciting; the monitoring is not. If nobody is staffed to watch it, the answer to question four is fiction.

Worked example

A proposal: "Use AI to summarize all internal documents so employees can find answers faster." Run it through the gate. Question three — what does success look like in 90 days? The proposer says "faster search." That is not a number. Pressed: "reduce time-to-answer for the top 20 questions by half." That is a number, and it changes the project: it no longer means "summarize everything," it means "curate twenty answers." The gate just saved the team from building the wrong thing.

Practice

Take one proposed AI project you have heard about or are part of. Answer questions 3, 4, and 10 in writing. Notice which ones you cannot answer without asking someone.

Apply — produce the artifact

Take one real proposed AI project or use case. Answer all ten questions in writing. End with a clear recommendation: Proceed / Fix prerequisites first / Kill it.

Verify

Give the answered list to someone who will have to live with the consequences. If they find an unanswered or weakly answered question that would change the decision, address it.

Sources

This module is original practice guidance based on the authoring standard and does not depend on a specific external factual claim. Editorial review is still required.