問題文
A retailer routes 2 million inbound emails per month into one of six queues. The routing task is a short classification with clear category definitions, and an evaluation on 1,000 held-out messages shows that a fast, low-cost model grade reaches 97.1 percent accuracy while the most capable grade reaches 97.6 percent. Latency matters because customers see an acknowledgement screen. Which model selection is best justified?
選択肢
- Use the most capable grade but cache responses so that repeated emails are cheap.
- Run both grades on every message and take the answer they agree on.
- Use the most capable grade for routing, because higher accuracy is always the correct architectural choice when it is available.
- Use the fast, low-cost grade for routing, because the accuracy difference is negligible for this task while the cost and latency advantages apply to every one of the 2 million messages.