20 questions from 5 pages

According to RAND's 2024 study of AI project failures, what is the first root cause it lists?

Which discovery question is most likely to produce reliable information?

A workflow map shows 9 minutes drafting, 12 minutes lookup and a 1.5-day approval queue. The sponsor's complaint is that customers wait too long. What does this tell you about a drafting assistant?

Why must the baseline be measured before the build starts?

In the support-drafting example, which change would improve ROI the most?

A team says the assistant 'saves 2,100 hours a month, worth $94,500'. What should a reviewer ask?

Discovery shows customers wait four days because claims sit in a supervisor approval queue. Is this a good LLM use case?

Why present ROI as a sensitivity table instead of one number?

Which section of a design doc usually gives reviewers the most help in finding a better design?

Six months after accepting ADR 0004 (hosted API), the team decides to self-host. What should happen to ADR 0004?

Why is 'model lifecycle' a non-functional requirement for systems built on hosted models?

What two things should every ADR include beyond the decision itself?

You have 5 minutes with a CFO to get approval for a pilot. What do you say first?

Which statement presents an eval result best to a non-technical sponsor?

Why include a deliberate failure in a demo?

A team lead says 'This will replace my team.' What is a good response?

What most often turns a pilot into 'pilot purgatory'?

A use case drafts responses to insurance complaints. Which stage should the pilot start at?

Handling time fell 30% during a pilot that ran from November to January. Why is that not yet strong evidence?

Name three things that must have a named owner at handover, and why.