Share via


Measure agent value using Copilot Studio Kit (preview)

[This article is prerelease documentation and is subject to change.]

The Agent Value component is a modular solution designed to classify conversational and autonomous agents in Copilot Studio based on static metadata and prompt-driven logic such as type, behavior, and value benefit. This classification enables organizations to:

  • Understand the role and purpose of agents.
  • Align agents with strategic goals.
  • Quantify the value each agent delivers.

The component consists of:

  • A classification Agent Value table.
  • A custom page that visualizes the spread and distribution of classified agents.
  • Power Automate flows for automated classification using AI tools.
  • An environment variable to enable the component.

Set up the Agent Value Component

  1. Review the prerequisites.
  2. Install the kit.
  3. Sign in to Copilot Studio.
  4. Set the environment variable Enable Value Component to Yes for the automated flows to operate.
  5. Turn on all Power Automate cloud flows.
  6. Run the Value Summary | Initialize agent value inventory cloud flow manually.

Dataverse table

The Agent Value component includes a custom Dataverse table to store agent classification metadata.

Agent Value

The Agent Value table is the core metadata table. This table stores the classification output for each agent, including:

  • Agent type
  • Primary behavior
  • Value benefit
  • Date of classification

The table is automatically populated by Power Automate flows using AI tools prompt responses.

Table schema

Column name Type Description
AgentId Guid ID of classified agent
Name Text Agent display name
Agent Type Text One of: Collaborator, Performer, Retriever, Orchestrator, Assistant, Advisor
Agent Behavior Text One of: Sense, Decide, Act, Collaborate, Reflect, Respond
Agent Value Benefit Text One of 10 defined value benefits
Classification date Date/Time Date of agent classification
EnvironmentDisplayName Text Display name of agents environment
EnvironmentId Text ID of environment

Power Automate flows

The component includes two Power Automate flows that classify agents by using AI tools.

Value summary | Initialize agent value inventory

Important

To populate the Agent Value table, enable this flow and run it manually. Only agents with completed metadata (description and instructions) are classified.

Trigger: Manually

Purpose: Iterates through all agents in the Agent Details table and classifies them if both the description and instructions are present.

Actions:

  • Filter agents with missing metadata
  • Call "Run a prompt" action with the embedded classification prompt
  • Parse and update:
    • Agent Type
    • Agent Behavior
    • Agent Value Benefit
    • Environment ID
    • Environment Display Name
    • Classification date

Value summary | New agent classification

Trigger: When a new row is added to Agent Details.

Purpose: Runs classification immediately for new agents.

Actions: Same logic as the previous flow, scoped to a single agent row.

AI Tools prompt: Classify agent

The "Run a prompt" action uses a structured system prompt to classify the agent. The embedded prompt includes a prompt objective.

Given the agent's instructions and description, assign it:

  • A primary Agent Type
  • A dominant Behavior
  • A Value benefit it most likely delivers
  • A simple formula or signal to measure this value