Question

Your team is refining a customer-support chatbot built in Azure AI Foundry prompt flow. Two engineers propose different system prompts, and you need to determine objectively which prompt produces higher-quality responses across a curated test dataset before promoting one to production. You want to keep both prompt versions traceable and repeatable. What should you do?

Accepted Answer

Create two prompt variants within the same prompt flow node and run a batch evaluation against the test dataset, comparing metrics such as groundedness and relevance.

Answer

Deploy each prompt as a separate managed online endpoint and route 50 percent of live production traffic to each to compare user satisfaction.

Answer

Manually paste each prompt into the chat playground, submit a few sample questions, and pick the prompt whose answers look better.

Answer

Provision additional provisioned throughput units so both prompts can be tested at higher concurrency before choosing one.