Current pathHuman + AI Workflow Design
Step 4 / 6PracticeLab15 min 50 🪙
Human and AI Decision Boundaries
Design what an agent may recommend, do, stop, and escalate.
Scenario
Foundation caseAn e-commerce company receives 18,000 refund requests monthly. Refunds below TRY 2,000 currently require agent approval and wait 19 hours on average. Seventy-two percent have verifiable reasons such as delivery delay or damage. Fraud flags include more than three refunds from one card in 30 days, a different delivery address, and high-value electronics.
Your role
You are the product owner for refund operations.
Your task
Define what an agent may automate, recommend, and escalate.
Volume and risk signals
- There are 18,000 monthly requests and 19 hours of waiting.
- 72% have verifiable reasons.
- More than three refunds per card is a risk signal.
- High-value electronics carry higher consequence.
Your response
There is no single correct answer; we will evaluate the reasoning chain you build.
0/4000 · min 25
0/4000 · min 25
0/4000 · min 20
0/4000 · min 25
0/4000 · min 20