Question

A production Copilot Studio agent orchestrates three downstream tools: a Dataverse query action, a custom Azure Function connector, and a Foundry-hosted model for summarization. Users report intermittent slow responses, but overall agent analytics show acceptable average latency. As the solutions architect, you need to identify which specific component causes the tail-latency spikes without disrupting production. Which monitoring approach should you implement?

Accepted Answer

Enable Application Insights distributed tracing with correlated end-to-end telemetry across the agent and each tool call, then analyze per-span duration percentiles (P95/P99)

Answer

Rely on the built-in Copilot Studio analytics dashboard's average session duration and escalation metrics to locate the bottleneck

Answer

Add a manual timer variable inside the topic and log elapsed time to a Dataverse table for later review

Answer

Increase the Foundry model's provisioned throughput units, assuming the model is the slowest component