CopilotScopeWork in progress
StudioLensAvailable—

StudioLens

Available

Agents & Studio

Deep agent performance and evaluation for Copilot Studio. StudioLens reads your agents' conversation transcripts and turns them into answers: how agents perform, resolve, contain, and cost — quality, topics, errors, feedback, and message-credit consumption, in one Power BI template.

2builds of one template
2required inputs to start (Dataverse path)
~30days of Dataverse transcript retention

What it shows

StudioLens reads your agents' conversation transcripts and turns them into answers.

  • Sessions & turns

    Agent sessions, turns, errors, and sub-agent calls.

  • Quality & performance

    Resolution, containment, abandonment, and escalation.

  • Topic analysis

    What users actually ask, and where conversations succeed or break down.

  • Knowledge usage

    Which knowledge files agents actually draw on.

  • User feedback

    Signal from the people talking to your agents.

  • Message credits

    Copilot Studio credit consumption, for the cost side of the story.

What's inside

Two builds of one template — shared transcript model, different page availability and data plumbing.

Dataverse (Direct)

Start here
  • Parses transcripts live in the Power BI model
  • Open the template and set the two required inputs
  • One or many Dataverse environments, zero infrastructure

Fabric

  • Transcripts and credits land as Delta tables in a Lakehouse
  • Scheduled ingestion for larger volumes and longer history
  • Unlocks the PPAC message-credit pages

Data requirements

SourceDescriptionRequired
Dataverse environmentWhere Copilot Studio logs conversation transcripts.Required
Org data CSVUser and department context for the reports.Required
Fabric capacity + LakehouseStorage and compute for the Fabric build.Fabric path only
Note Dataverse's ~30-day transcript retention — the Fabric path is the answer for historical analysis at scale.

Get this lens

Two paths, shared transcript model:

  • Dataverse (Direct)

    Start here

    The simplest, lowest-footprint build.

    • Transcripts parsed live in-model
    • One or many Dataverse environments
    • No infrastructure to stand up
    Download template

  • Fabric

    Scheduled ingestion for scale and history.

    • Scheduled Spark ingestion
    • Larger volumes and longer history
    • Adds the message-credit pages
    Download template

Both paths include documentation and setup guidance.

Experimental template reading Copilot Studio transcripts from Dataverse. Not an official Microsoft-supported product — questions and issues via the GitHub repo.