Rating
- size:
3 - complexity:
4
Purpose
Practice using AI on policy-heavy logic where polished drafts can hide semantic mistakes and unverified assumptions.
Scenario
A billing policy changes:
- customers get a
7-daygrace period before cancellation - prorated refunds apply if cancellation happens after the grace period
- enterprise customers follow a different renewal rule
AI proposes a single decision tree that looks clean and complete.
Task
Use AI to help critique the proposed decision tree rather than simply accept it.
Identify:
- missing policy questions
- edge cases
- where different customer categories may diverge
- what cannot be implemented safely without policy clarification
Expected output
- annotated decision tree or rule list
- unresolved questions
- verification plan using examples and counterexamples
Good AI uses
- generating boundary scenarios
- surfacing conflicting interpretations
- comparing example cases across policy branches
Verification focus
- every business rule must tie back to source policy
- enterprise exceptions must not be inferred casually
- example-based verification should cover boundary dates and customer types
Anti-patterns to watch
- trusting a complete-looking decision tree
- letting AI collapse different policy branches into one clean story
- implementing before unresolved policy points are escalated
Debrief prompts
- where did the AI draft look strongest while still being unsafe
- which unresolved question mattered most
- what counterexample most stress-tested the policy