本案例依原書 Table 11.2 完整整理。
用五個 Real-world Query 檢查 Chatbot 在 Transparency、Payment、B2B、Authenticity 與 Customer Service 上的回答能力。
原書將常見 Stakeholder Query 輸入 Chatbot，觀察回答是否清楚、符合規範並與情境相關。
| Query | Outcome |
|---|---|
| How do you handle incorrect or damaged deliveries? | 說明 reporting steps、timelines、sensitive product conditions，以提高 transparency／trust |
| What payment methods do you accept? | 列出 COD、UPI、insurance reimbursement compatibility 等 payment options |
| Can hospitals place bulk orders? | B2B volume discount、automated restocking options |
| How is medicine authenticity ensured? | Licensed suppliers、QR verification、blockchain-enabled traceability |
| How to contact customer service? | Toll-free helpline、live chat、email 等 multi-channel support |
建立 Query Test Set，分欄評估 Intent、Required Evidence、Clarity、Compliance、Contextual Relevance、Escalation Need。
Table 11.2 未提供各回答的數值分數，因此不自行建立 Accuracy／Compliance Score。
Sample Query 可測 Functional Coverage，但不能單獨證明 System-wide Performance；需與 Table 11.3 Operational KPI 搭配。
請依 Table 11.2 為每個 sample query 標記 intent、required evidence 與 potential escalation condition；不得自行給模型分數。