The 10 Questions as an Organizational Gate

Why this matters

Individual judgment is not enough when money, policy, and organizational risk are involved. The same ten questions from Core Judgment now function as a gate for any serious proposal. The difference is the answer level: an individual answers "who owns this when it breaks" for their own output; an organization answers it with a budget line, a role, and a real person.

What you will be able to decide after this

• Whether a proposal survives the same filters that apply to a single AI use — at org scale.
• Where the proposal is disqualifying, and where it is fixable.
• What the honest recommendation is: proceed, fix prerequisites, or kill.

Core lesson

The ten questions, at the organizational level:

1. Can a human already do this job reliably with the data we have today? If yes, the AI has to beat an existing, working baseline — measured, not assumed.
2. Do we make this decision or perform this task often enough to justify the cost? Infrequent work pays for integration, monitoring, and training slowly, if ever.
3. What does success look like in 90 days — in numbers, not slogans? A slogan cannot be checked at day 90.
4. Who owns this when it breaks? A named role with authority, not "the AI team."
5. Can we measure accuracy or usefulness in production? If the answer is "we will know it when we see it," there is no production standard.
6. What happens when the AI is wrong? The failure mode, traced to an action. If it is undefined, the risk is unowned by definition.
7. Would a simple rule or process solve most of the cases? The cheapest correct answer beats the expensive impressive one.
8. Do we have real support for the actual timeline (not the optimistic one)? People, budget, and attention — committed past the demo.
9. Is the data clean and traceable enough to trust? Unclean data does not get fixed by the model; it gets amplified by it.
10. Can we staff the monitoring and maintenance after launch? The launch date is the start of the cost, not the end of it.

Worked example

A team proposes an AI system that screens incoming support tickets and auto-drafts replies. The ten-question gate finds the weak answers where they usually sit: on question 4, ownership of bad replies is unclear — "the team" is not a person with authority; on question 6, the failure mode for angry or high-value customers is undefined; on question 9, ticket data quality is uneven, with no plan for the bad third; on question 10, no one is assigned to monitor drift after launch. Those gaps are disqualifying until fixed. The recommendation is not "kill it" — it is "fix prerequisites first": name the owner, define the failure mode, sample and clean the data, assign the monitoring role. The proposal is not ready; it is also not hopeless.

Practice

Take the proposal you will evaluate. Read the ten questions and mark each as strongly answerable, weakly answerable, or not answerable in your current knowledge. Note the three you would least want to defend to a budget holder.

Apply — produce the artifact

Take one real AI proposal currently under discussion (or a realistic one). Answer all ten questions in writing at the organizational level. End with a clear recommendation: Proceed / Fix prerequisites first / Kill it.

Verify

Share the answered list with someone who has budget or operational authority. If they identify an unanswered or weakly answered question that would change their decision, address it before proceeding.

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.