CCAR-F scenarios › Customer Support Resolution Agent · Question 9 of 60
Scenario 1 · Customer Support Resolution Agent
You're building a support agent with the Claude Agent SDK that resolves ambiguous customer requests such as returns, billing disputes, and account problems. It reaches your backend through custom MCP tools: get_customer, lookup_order, process_refund, and escalate_to_human. The goal is to resolve at least 80% of contacts on first touch while recognizing the cases a human should take over.
Context Management & Reliability
A customer says "I'm Maria Lopez, I was double-charged last week." get_customer returns three active accounts named Maria Lopez in different cities. Logs show the agent picks the account with the most recent order and proceeds to investigate the charge. What should the agent do instead?
Show answer and explanation
Answer: D — Ask the customer for an additional identifier, such as email address, phone number, or an order number, to determine which account is theirs before taking any action.
Multiple matches call for clarification with an additional identifier, not heuristic selection (task 5.2). Picking by matching amount (a) is another heuristic, and confirming after a refund is too late if it was the wrong account. Escalating every multi-match case (b) sends a routine clarification to humans and hurts first-contact resolution. Investigating all three accounts and showing the results (c) exposes other customers' account data to an unverified person.
More from this scenario
- You are writing explicit escalation criteria with few-shot examples for the system…
- Company policy caps agent-issued refunds at $500; anything larger must go to a human…
- Your loop around the support agent ends the conversation whenever Claude's response…
- Human agents who receive cases from escalate_to_human report that they spend the first…
- When a customer asks for a refund on an order that is outside the 30-day return window…