Extend Copilot with plugins and measure its impact
Configuring Copilot with a knowledge base and filters handles most representative assistance needs. But some scenarios require Copilot to reach beyond static knowledge articles—retrieving live data from external systems, running custom AI prompts, or surfacing information from non-Microsoft tools. Plugins make this possible.
At the same time, knowing whether Copilot is actually being used—and whether it's helping—requires analytics. The Copilot analytics report gives you that visibility.
What are Copilot plugins?
Plugins extend Copilot by connecting it to external data sources and custom logic. When a representative asks Copilot a question that triggers a plugin, Copilot calls the plugin, retrieves the relevant data, and incorporates it into the response. From the representative's perspective, this looks like a normal Copilot answer—the data access happens in the background.
Two plugin types are available in Dynamics 365 Contact Center:
Prompt plugins connect Copilot to Dataverse and external AI Builder prompts. You build a prompt that defines what data to retrieve or what operation to perform. For example, a prompt plugin might retrieve a customer's subscription tier from Dataverse and surface it when the representative asks about the customer's account.
Custom connector plugins connect Copilot to any external system reachable through a Power Automate custom connector—such as an order management system, a billing platform, or a non-Microsoft CRM. Representatives can ask Copilot for order status, invoice details, or any other data the connector can access.
Both plugin types require a Copilot Studio license and are created in Copilot Studio before being enabled in the admin center.
Note
Plugins are a preview feature. Test and validate plugin behavior in a nonproduction environment before enabling for service representatives.
Enable a prompt plugin
Before enabling a plugin in the admin center, create the prompt in Copilot Studio using AI Builder. Ensure the prompt description is specific and unique—Copilot uses the description to identify which plugin to call at runtime.
Once the prompt is created and published in Copilot Studio:
- In the Copilot Service admin center, go to Support Experience > Productivity > Plugins for generative AI (Preview), then select Manage.
- Select the prompt plugin and choose Turn on from the command bar.
- Select the access type:
- All agents who have Copilot: The plugin is available to every representative with Copilot enabled.
- Specific user roles: Restrict the plugin to selected roles.
- Optionally, configure inputs to make the plugin context-aware. When inputs are configured, Copilot automatically pulls field values from the current record—such as the customer name or account ID—and passes them to the plugin without requiring the representative to type them manually.
- Choose whether to store plugin interaction data in Dynamics 365 Copilot transcripts.
- Select Turn on plugin, then Publish changes. Changes can take up to 15 minutes to take effect.
Enable a custom connector plugin
The setup flow for a custom connector plugin follows the same general steps, with a few important additions:
- After selecting the plugin and choosing Turn on, you must link it to a Power Automate connection.
- Choose a User authentication method:
- Admin: All accesses use admin authentication. Simpler to configure but appropriate only for nonsensitive data.
- Agent: Representatives authenticate individually at runtime. Required for sensitive or restricted data sources.
Tip
Use agent authentication for any external system that contains customer-specific or confidential data. Admin authentication is convenient but reduces the auditability of individual representative access.
Measure Copilot adoption with the analytics report
Enabling Copilot is the first step. Understanding whether representatives are using it—and whether it's delivering value—requires the Copilot analytics report.
Access the report from Copilot Service workspace by selecting Copilot analytics from the navigation menu. The report is filterable by day, week, or month.
Usage metrics
| Metric | What it tells you |
|---|---|
| Daily active users | How many representatives are actively using Copilot. Low numbers suggest adoption challenges or awareness gaps. |
| Total Copilot AI responses | Total responses generated across all features. |
| Number of responses used | How many responses representatives copied. |
| Percentage of Copilot AI responses used | The core adoption indicator—what proportion of generated responses are actually put to use. |
A high daily active user count with a low percentage of responses used is a signal that representatives are engaging with Copilot but not finding its output useful enough to apply. This typically points to a knowledge source quality issue, poor response instructions, or missing filters.
Productivity metrics
The report also shows conversation-level productivity data, including average handle time and conversation throughput—with separate views for interactions where Copilot was used versus where it wasn't. This comparison gives you the clearest evidence of whether Copilot is reducing handling time, which is often the primary business case for adoption.
Representative satisfaction ratings
Representatives can rate Copilot responses with a thumbs up or thumbs down. These ratings accumulate in the Agent ratings metric and give you a qualitative signal about response quality that complements the quantitative usage data.
Tip
Review the Copilot analytics report on a regular cadence—weekly during initial rollout, monthly once adoption stabilizes. Use low response usage percentages or declining ratings as triggers to review and update your knowledge sources, custom instructions, or filter configuration.
With plugins extending Copilot's data reach and analytics informing your continuous improvement cycle, you have the full set of tools needed to run an effective, evidence-based AI program in your contact center.