Share via

Copilot Studio agents report

The Copilot Studio agents report can help you understand the adoption and impact of agents built in Microsoft Copilot Studio that are deployed across various channels within your organization.

This report can help you answer questions such as:

  • What are the top agents being used?
  • What are top agents' high-level KPI-like sessions, satisfaction scores, and success rates?
  • What is the impact of individual agents (conversational and autonomous), such as the split of engaged sessions and topics against actions and triggers distribution, as well as maker-led inputs?

Example Agent KPIs

What's covered in this report:

Agents built in Microsoft Copilot Studio that are deployed in channels such as:

  • Custom and demo websites
  • Microsoft Teams and Microsoft 365 Copilot
  • Mobile apps
  • Facebook
  • Azure bot service channels such as Skype, Twilio, and Microsoft Cortana

What's not covered in this report:

  1. Declarative agents built in Copilot Studio

  2. Agents built using Agent Builder in Microsoft 365 Copilot

  3. Agents built outside of Microsoft Copilot Studio like within Microsoft SharePoint or third-party apps

Prerequisites

This report is enabled for tenants who meet the following eligibility criteria:

  1. Tenant must have at least 50 Copilot licenses.
  2. Tenant must have at least one Microsoft Copilot Studio license, either standalone, pay-as-you-go, or bundled with Copilot, and have created at least one agent and published it within a "Production" and "default" environment, and should have usage. Learn more.

In addition, before you can run the query and populate the report in Power BI, you’ll need to:

Report setup

Run query

  1. In the Viva Insights analyst experience, select Create analysis.

  2. Under Power BI templates, navigate to Copilot Studio agents report and select Set up analysis. The template can also be found under the Copilot section.

  3. Under Query setup:

    1. Enter a Query name.
    2. Select a Time period, which defaults to Last 1 months.

    Note

    To start, you can only run an analysis covering the last month, but eventually you'll be able to select a wider time period as Microsoft Copilot Studio collects more data.

    1. Set Auto-refresh (optional). You can set the query to automatically update by selecting Auto-refresh. When you select this option, your query automatically runs and computes a new result every time Viva Insights gets updated agent data.

    2. Enter a Description (optional).

    Note

    The More settings pane also contains Group by and Metric rule settings. This Power BI query is set to Group by Week, and you can't change this setting, and neither can you change the Metric rule setting.

  4. Under Predefined template metrics, view a list of preselected metrics, which appear as gray tags. These metrics are required to set up the Power BI report and you can't remove them. You can't add other metrics for this report. Learn about the availability of data for Copilot metrics.

  5. Under Select which agents you want to include in the query, add filters to narrow down the agents in scope for your report. You can filter by agent name, agent surface, and agent type.

  6. Under Select which agent attributes you want to include in the query, attributes are selected automatically. Once the query runs, you can use these attributes to group and filter the reports.

    Important

    This Power BI query needs some specific attributes to run, and we've preselected them for you. These attributes appear in gray and you can't remove them. You can't add other attributes by selecting Add attributes.

  7. Select Run on the upper right. The query might take a few minutes to run.

    Note

    You can only run the query if there's agent usage data (agent built in production and default environment and have usage data in terms of sessions) for at least four weeks.

Access query results

You can access the report in two ways: 

  • View the report in the browser. Use this option if you only want to view the report. 

  • Open the Power BI template in Power BI desktop and connect to your query results. Use this option if you want to customize the report or share it with others in your organization by publishing the report to the Power BI Service.

View report in the browser 

To view the report in the browser, go to the Query results page and select the eye icon in the View column. Select Open in new tab if you want to keep the report in the background while doing other tasks in the analyst experience.

Open the Power BI template in Power BI Desktop

  1. Go to the Query results page and select the Power BI icon in the Actions column to download the Power BI template and get the query and partition identifiers. You'll need these identifiers later. 

  2. Open the downloaded template. 

  3. If you're prompted to select a program, select Power BI. 

  4. When you're prompted by Power BI:

    1. Paste in the partition and query identifiers. 
    2. Select Load to import the query results into Power BI.

  5. If prompted by Power BI, sign in using your organizational account. Power BI then loads and prepares the data. For large files, this process might take a few minutes.

    Important

    You need to sign in to Power BI with the same account you use to access Viva Insights. If available, select Organizational account from the left. You might have to sign in more than once.

    Screenshot that shows signing into to Power BI on the Organizational account tab

Page-level settings

Each report page includes the following settings at the top:

  • Time period for the report – Select the start and end date for the data you'd like to view.

  • Apply filters – Select the Category, Agent name, and Agent usage type to filter the Agents shown in the report.

About this report

The Copilot Studio agents report includes the following report pages to help you better understand agent adoption and KPIs across your organization.

Agent Overview

This page provides an initial overview of adoption of the agents across the organization for the selected filters. The card on the left shows the Agent landscape as the total number of agents within the tenant, split by conversational and autonomous agents. It also gives the split of agents with the total number of sessions and number of engaged sessions, which is split by conversational and autonomous.

The card in the middle of the page shows the trend line of active agents, or the number of sessions split by conversational and autonomous agents.

The card on the right identifies the top five agents based on the number of monthly engaged sessions.

At the top of the page, you can also filter by Category and Usage type. For Usage type, select Internal to see metrics for internal employee-facing scenarios. Select External to see metrics for agents used externally, such as on public websites.

FAQs

Q1. What are Autonomous agents?

Autonomous agents are agents built in Copilot Studio with defined event triggers. Learn more.

Q2. How many agents were active this period?

An "active agent" is defined as any agent with whom a user has engaged in at least one session during the selected time period.

Q3. What's the total number of engaged agent sessions?

The total number of engaged agent sessions indicates the number of times users engaged with the agents identified in question 1 above. 

Q4. How has the number of active agents and number of sessions changed over time?

The trend lines illustrate the change in key metrics over time. Use the dropdown menu to toggle between the metrics. Rising trend lines over time indicate a growth in the number of agents.

Pay attention to the percentages at the top of each chart, which indicate the change over the previous month. Review the percentages for number of sessions, and how they compare to the number of agents. If the number of agents is growing faster than these usage and impact metrics, it might be a sign to focus more on promoting and enabling the existing agents.

Q5. What are the top five agents based on monthly usage?

These are the most popular agents currently in use at your organization. Select Explore more to view more in-depth metrics and view a full list of available agents. 

Identify top agents used across the organization

This page lets you focus on top agents sorted in descending order based on the number of monthly engaged sessions, and explore the key Agent KPIs.

You can filter the agents on this page using the Category, Agent name, and Usage type such as internal or external to the organization.

Top agents are split by conversational and autonomous agents.

Interpretation Guide

How to use the number of engaged sessions, satisfaction score, and resolution rate for actionable insights

Use the table below to evaluate the effectiveness and satisfaction of your agents and determine potential next steps. By analyzing successful agents, you can boost the performance of those that aren't doing as well.

Also consider the session volume in your evaluation. Is the agent being used enough to draw meaningful conclusions? If you have agents with low usage, you might want to think about promoting them more, retiring them, or reworking their design.

You can filter the agents on this page using the "Category," "Agent name," and "Usage type" fields, like internal or external to the organization. 

Use the "Category" field such as Sales, Support, or Internal to compare performance within similar agent types. This helps avoid unfair comparisons and can help suggest best practices within each area.

Resolution Rate Satisfaction Score - Low Satisfaction Score - High
Low This agent isn't delivering value

  • Review the agent's configuration to confirm it has been set up correctly
  • Prioritize for redesign or retirement
  • Interview users or review logs to understand failure points
  • Consider whether the agent's purpose is still relevant
    		•  Users like the experience but aren't getting what they need

  • Investigate if the agent is too limited in scope or lacks access to needed data
  • Review the agent's configuration to confirm it has been set up correctly
  • Expand capabilities or improve escalation paths
  • Consider retraining the model or refining intents
    		•
    High Tasks are being completed, but the experience might be frustrating or confusing

  • Review conversation logs for friction points such as tone, length, or complexity
  • Improve user experience elements like clarity, empathy, or speed
  • Consider user feedback mechanisms to pinpoint dissatisfaction
    		•  This agent is performing well and meeting user needs

  • Use it as a benchmark and analyze what makes it successful such as tone, flow, or content
  • Consider scaling or promoting it more broadly
  • Replicate its design patterns in underperforming agents
    		•

    Conversational Agent Impact

    This page provides a deep dive into specific KPIs for particular agents using your selected filters. The chart on the left shows the number of engaged sessions over time, and you can toggle between this and Agent session outcomes based on Resolution, Abandons, and Escalations.

    The chart in the middle of the page also shows the Agent satisfaction score for the selected timeframe, as well as the number of Agent assisted hours that could be gained by deploying the agent across the organization. For each task in a conversational agent session, the metric compares the difference between the time a human would take and the time an agent actually took, and it sums these differences across all tasks and sessions.

    The chart on the right shows the number of sessions per topic and sessions per action for the agent in the selected timeframe. Learn more about topics in Copilot Studio.

    Agent assisted impact (time)

    Agent assisted hours estimates the amount of time employees saved by using an agent to perform tasks. It uses the number of engaged agent sessions, number of knowledge source references, session outcomes, and a time savings multiplier to calculate the total agent assisted hours. This metric helps businesses understand efficiency gains and time savings by deploying agents.

    Here's the formula for the metric:

    Agent assisted hours = (Number of knowledge source references + Weighted number of sessions without knowledge sources) x Time savings multiplier

    A given agent might have multiple sessions, some with references to knowledge sources and others without these references. For each knowledge source reference, the amount of time saved for the employee is estimated to be equal to the value of the time savings multiplier. For the sessions without knowledge source references, there might also be time savings, which are estimated based on the outcome of the session. Resolved sessions count as a full session, while escalated and abandoned sessions are weighted at 0.7. These weights are determined by estimating the extent of task completion based on a Microsoft analysis of session duration for each outcome type.

    The time savings multiplier is customizable, with a default value based on research on AI usage, including studies by Microsoft's Office of the Chief Economist (see "The strain of searching") showing an average of six minutes of time saved for tasks involving information retrieval (see study #4 in section 2 of the blog post), as well as studies highlighting six-minute time savings for writing tasks (see study #1 in section 2 of the blog post). For additional guidance on how to maximize these productivity gains, see Agents of Change. The default multiplier is an estimate and might be refined over time as new data become available.

    To edit the value of the multiplier, select the calculator, where you'll see a breakdown of how this metric is calculated for a specific agent. If you change the value, it updates for all agents in the report.

    Agent assisted impact (cost)

    This is an estimate of the value of an agent's impact over the given time period, calculated by multiplying Agent assisted hours by an average hourly rate. By default, this hourly rate is set to $72, based on data compiled by the U.S. Bureau of Labor Statistics. To update the average hourly rate, select the Calculator, where you'll see a breakdown of how the total value is calculated with your updated setting.

    To toggle between Agent assisted hours and Agent assisted value, select the appropriate symbol at the top right of the card.

    FAQs

    Q1. How have usage and satisfaction changed over time for this agent?

    These two metrics indicate the number of people using the agent over the last three months, as well as the average agent satisfaction rating.

    Use these sections to gain a high-level view on the success of the specific agent.  Based on the trends you see, use the table below for potential actions to take.

    Trend Potential interpretation Recommendation
    Rising session volume Increased adoption or successful rollout
  • Investigate which agents or topics are driving growth
  • Reinforce successful patterns and replicate them across other agents
  • Review satisfaction scores to ensure quality is keeping pace with volume
    		•
    Declining session volume Possible user disengagement, technical issues, or seasonal drop-off 
  • Check for recent changes in agent behavior, broken triggers, or user feedback
  • Re-engage users with targeted communications or improvements
  • Review satisfaction scores to identify if declining quality is contributing to the drop
    		•
    Flat or stagnant volume Stable usage, but possibly underutilized potential 
  • Consider promotional efforts, training, or expanding the agent's capabilities to drive more value
  • Analyze satisfaction scores to determine if users are content or indifferent
    		•
    Low volume Very few sessions might indicate low adoption or discoverability
  • Promote the agent in internal communications such as newsletters, Yammer, or Teams
  • Educate and train users
  • Review satisfaction scores from the limited sessions. Positive scores suggest the agent is on track and you might need to increase awareness. Negative scores indicate improving the agent before wider deployment.
    		•
    Sudden spikes or dips in volume  Could indicate a campaign, outage, or a new feature rollout 
  • Coordinate with internal events or releases
  • Use these insights to plan future launches or mitigate issues
  • Check satisfaction scores to assess user sentiment during the spike or dip
    		•

    Q2. What are the session outcome trends for agent maturity?

    Select the bar graph in the upper right corner of the section trend to toggle between the session outcome trends.

    It's good to see a positive trend while moving away from "Abandoned" toward "Resolved." Ideally, satisfaction scores should increase as resolution rates go up and decrease when there are more cases of abandonment or escalation. Think of these scores as a way to confirm or question the volume-based interpretations.

    Trend Potential interpretation Recommendation
    Increasing resolved sessions Your agent is effectively handling user prompts/queries
  • Reinforce successful patterns and analyze which topics or flows are contributing to high resolution and replicate them
  • Review satisfaction scores to confirm that resolution quality aligns with user expectations
    		•
    Low resolution rate A low percentage of successful outcomes suggests the agent isn't meeting user needs
  • Review unresolved sessions to identify common issues like misunderstood intents, missing content, or ineffective responses
  • Enhance training data and refine conversation flows to better support user goals
  • Analyze satisfaction scores for unresolved sessions to understand user sentiment. Low scores might indicate frustration, while neutral or high scores could show users found value even without a formal resolution. Use these insights to prioritize improvements. 
    		•
    Rising escalations The agent is hitting limitations in handling prompts/queries
  • Review escalation triggers. Are they firing too early?
  • Could the agent be trained to handle more before escalating?
  • Review the agent's configuration to confirm it has been set up correctly
    		•
    High or growing abandonment Users are leaving before getting help, possibly due to confusion, delays, or irrelevant responses
  • Audit conversation paths. Look for friction points or dead ends.
  • Consider configuration such as adding fallback logic
  • Review the agent's configuration to confirm it has been set up correctly
    		•
    Persistent abandonment outcomes Sessions are starting but not engaging, and users might be testing or unsure how to interact
  • Improve onboarding or configuration. Use fallback logic or experience cues to guide users into meaningful engagement.
  • Review the agent's configuration to confirm it has been set up correctly
    		•

    Topics recommendations

    To analyze topic effectiveness, select an individual topic under Number of Sessions Per Topic to learn more. Some suggestions for further analysis include: 

    1. Track resolution rates by topic: Observe how frequently sessions on a specific topic conclude with a "Resolved" status, indicating the topic is well understood and effectively handled. 

    2. Monitor escalation and abandonment: Identify topics that lead to escalations or users giving up, signaling possible content gaps or user frustration. Compare these trends with satisfaction scores to determine if users are dissatisfied or simply need more guidance. 

    3. Evaluate engagement metrics: Assess the duration of user engagement with each topic and their interaction with multiple steps or follow-up prompts. High engagement and positive satisfaction scores typically signify a well-designed topic flow. 

    4. Use satisfaction scores for sentiment analysis: Review satisfaction scores for each topic to understand user sentiment beyond resolution. Quick resolutions with low scores might indicate a rushed or incomplete experience. 

    5. Compare topic performance over time: Examine how topic effectiveness evolves over time to prioritize updates or training for underperforming areas.

    Autonomous Agent Impact

    This page provides a deep dive into specific KPIs for particular autonomous agents (agents with pre-defined event triggers) using your selected filters. The chart on the left shows the number of sessions over time, and you can toggle between this and agent session outcomes based on successful and unsuccessful outcomes.

    The chart in the middle of the page also shows the average completion time for the selected timeframe, as well as agent assisted impact that could be gained by deploying the autonomous agent across the organization.

    The chart on the right shows the number of sessions per trigger for the agent in the selected timeframe Trigger Overview as well as the option to select the number of sessions per action for the agent in the selected time frame with the Actions/Tool Overview.

    Agent assisted Impact (Autonomous Agents)

    Agent-assisted Impact for autonomous agents is calculated in two ways.

    1. Maker-led inputs: Maker-led inputs are passed to Viva Insights and Viva Insights reflects the updated value. Maker-led values are not editable in Viva Insights. Learn more.

    2. Inferred Agent-assisted Impact (AAI) for Agent assisted hours (AAH): Inferred Agent-assisted impact for Autonomous agents estimates the amount of time and cost savings by using an agent to perform tasks. Autonomous agents save time for the end user both when performing information retrieval from knowledge sources and while performing actions.

    AAI for an agent would be sum of the following:

    • Time saved in information retrieval: This represents the number of times knowledge source references were used, multiplied by information retrieval time savings (such as six minutes).

    • Time saved in performing actions: This represents the number of times an action was performed multiplied by the time savings multiplier for that action type.

    • Time saved in successful sessions which used neither independent actions nor knowledge source: These sessions have actions within a topic; therefore, the specific action performed is unknown. Instead, a generic time savings multiplier is used.

    Here's the formula for the metric:

    Agent assisted hours = (Number of Knowledge source referenced x Time saving in information retrieval (in minutes) + Total time saved in automating actions (in minutes) + Number of successful sessions without presence of independent actions x Generic time saving estimate (in minutes))/60

    The time savings multiplier for information retrieval is customizable, with a default value based on research on AI usage, including studies by Microsoft's Office of the Chief Economist (see "The strain of searching") showing an average of six minutes of time saved for tasks involving information retrieval (see study #4 in section 2 of the blog post), as well as studies highlighting six-minute time savings for writing tasks (see study #1 in section 2 of the blog post). For additional guidance on how to maximize these productivity gains, see Agents of Change. The generic time savings estimate is derived from the time savings estimate for commonly automated actions using telemetry, research, and large language model estimates. The time savings estimates will be refined over time as new data become available.

    To edit the value of the multiplier, select the calculator, where you'll see a breakdown of how this metric is calculated for a specific agent. If you change the value, it updates for all agents in the report.

    Agent assisted Impact (Cost)

    This is an estimate of the value of an agent's impact over the given time period, calculated by multiplying Agent assisted hours by an average hourly rate. By default, this hourly rate is set to $72, based on data compiled by the U.S. Bureau of Labor Statistics. To update the average hourly rate, select the Calculator (it's visible for agents when maker-led values aren't there), where you'll see a breakdown of how the total value is calculated with your updated setting.

    Add business outcome data (preview)

    Important

    This feature is for public preview customers only. Features in preview might not be complete and could undergo changes before becoming available in the broader release. This preview feature allows personal data processing. See the "Previews" section of the Microsoft Products and Services Data Protection Addendum for more information.

    To upload business outcome data, select Add business outcome data.

    Keep the following best practices in mind:

    • The URL should be a OneDrive or SharePoint (ODSP) link.

    • You can only use a .csv file.

    • The first column should be the date.

    • There should be at least one Business KPI column.

    • Use the Agent name exactly as it's shown in the report's Agent name field.

    Delete Copilot Studio agents data

    The admin can delete the existing agents data for the tenant using PowerShell. Learn how.

    After the data is deleted, the query won't auto-refresh, and you can't run any new queries for one month.

    Where to find Power Platform admin center settings

    1. Microsoft 365 Service setting
    2. Setting for Data flow from Microsoft Copilot Studio to Viva Insights
    3. Cross Geo setting

    Power BI tips, FAQs, and troubleshooting

    Learn more about how to share the report and other Power BI tips, troubleshoot any issues, or review the FAQ.

    Filters

    • Last updated on