Question

A business solutions architect is designing an agent flow in Copilot Studio for a claims-processing team. The flow must take unstructured email text describing a customer complaint and produce a consistent, machine-readable result containing a category, a severity score (1-5), and a short summary, which downstream Power Automate steps will use to route the claim. The architect wants to use a prompt action but ensure that the output can reliably feed the branching logic that follows. Which approach best meets this requirement?

Accepted Answer

Configure the prompt action to return structured JSON output with defined fields for category, severity, and summary, so downstream actions can reference each property directly.

Answer

Use a prompt action that returns free-form text and add a fallback topic to reparse the text if the branching logic fails.

Answer

Replace the prompt action with a conversational language understanding (CLU) intent model that classifies the email into predefined routing intents.

Answer

Configure the prompt action to summarize the email as plain text and store the whole summary in a single variable that the branching condition evaluates with keyword matching.