Currently there are two topics that are omnipresent for me: Transition to real-time marketing in Customer Insights Journeys (obviously) and the agent with Copilot Studio (thanks to Microsoft Ignite all around). The first one I already know a little about, but the second one i must admit i haven’t yet looked into so much. Copilot Studio seems to be a big topic but i thougth now might be finally the time for me check it out. And of course I fell into that rabbit hole..
So why not creating an agent for the transition to real-time marketing? Users may find the shift from outbound marketing to real-time marketing in Dynamics 365 overwhelming. Many of them ask questions about new features, best practices, and adapting their strategies. An agent assists users instantly, smoothing and streamlining the transition.
In this blog, I’ll guide you step-by-step through the process of creating your own agent using Copilot Studio and integrating it into Microsoft Teams.
What is an agent in Copilot Studio?
Before we dive in, let’s get on the same page: AI-powered agents answer questions, provide guidance, and perform tasks tailored to user needs. Think of it as a smart helper for your team.
For this article, our agent will guide users through the transition from Dynamics 365 Outbound to Real-Time Marketing in Customer Insgihts by answering FAQs and providing resources. So a simple agent to start with.
1. Step: Create the agent in Copilot Studio
Go to Copilot Studio to create your first agent. If you don’t have a subscription yet, you can also create a trial of course.
Create a new agent and specify the first details about it.
- Preferred language (the primary language cannot be changed later. Also when you use generative AI for selecting topics and actions, the agent is only supported in English yet.)
- Name
- Icon (The icon should be in PNG format and less than 30 KB in size.)
- Description
- Instructions (Direct the behavior of the agent, including its tasks and how it completes them. I will tell you more about this in the next chapter.)
- Knowledge (I will tell you more about this in the next chapter.)
2. Step: Add description and instructions to the agent
In the Description field, you can specify the purpose of the agent and how it can assist users. While in the Instructions field, you define the agent’s primary role and desired behavior. Think of the instructions as when you are interacting with ChatGPT or Copilot. Here you can give the agent the context and how it should behave. In the real-time transition for Customer Insights case I add the following instruction to my agent in Copilot Studio:
“You are an expert for the transition from outbound to real-time marketing in Dynamics 365 Customer Insights Journeys. Maintain a polite and professional tone while assisting with questions about the user’s business.”
With Generative AI functionality in Copilot Studio you can use the advanced language models (LLMs) to create original content and understand natural language. Either you can start with a blank agent and let it generate answers based on knowledge sources like websites or files. Then there is no need to manually set up topics. Or by enabling the feature Allow the AI to use its own general knowledge your agent can answer general questions, even if they aren’t tied to specific knowledge sources.
3. Step: Add knowledge to the agent
Microsoft Copilot Studio’s generative answers feature allows agents to pull information directly from internal or external sources. By this you can reduce the need to create specific topics. Agents independently handle all kinds of user questions.
As knowledge sources you can add public website, Sharepoint, Documents, Dataverse, Enterprise data using Microsoft Graph connectors.
I keep it easy for my little transition agent in Copilot Studio and use my website and the Microsoft Learn Page for Dynamics 365 Customer Insights Journeys: https://learn.microsoft.com/dynamics365/customer-insights/.
Note: When adding a new source based on a public website you can only add a URL that does NOT have more than two levels. I noticed though that you change this after you saved the source for the first time.
When you check the topics of the agent you can see one that is added and called Conversional boosting. The generative answers node is included there and uses the data sources for the answeres.
But the using the Generative AI option requires some time and specialized knowledge to set up. Otherwise the answers are not precise or correct, but on the other hand your agent is up and running from day one.
And also very important: In my case I only use data sources that are publicly available and they don’t contain secret data. But you could also add a PDF will non-public information to the knowledge base. Ensure users have permission to view the data from the PDFs.
Now you can test your agent in the test pane of Copilot Studio before we publish it to Teams.
4. Step: Publish and connect the agent to Teams
When you are done with creating your first agent you can now publish it for the first time. You can find the publish button in the top right corner or in the bottom of the Overview tab.
After publishing we want to connect the agent to Teams.
Go to the Channels tab in Copilot Studio and choose Microsoft Teams. In the assistent that pops up choose Turn on Teams. Once this is done you have few options to make the agent available in Teams.
- You can click on Open agent to install the agent only for you.
- Under availability options you can select Copy Link and share the link with users so they can install the agent in Teams.
- You can show it in the Teams app store so users can find and install it.
- either show to teammates and shared users that you can define yourself. Confirm the Visible in Copilot extensions so it is visible in the Teams store.
- or show to everyone in your org by submitting it for admin approval and it will then be featured in the Built for your org section of the Teams app store.
I decided to publish it to everyone in my organisation by choosing the section in the availability options. You can see the submission status and version of your agent here as well.
When you are sure of the content of your agent and everything is ready select Submit for admin approval. Then submit the agent app to the Teams admin. And after the admin approves, it updates the agent’s status to Approved.
This is also the way to send updates of the agent to Teams. Then a notification in the Teams admin center shows up.
Please note: Everytime you publish a new version of your agent you also need to approve the new version in the teams admin center. It might take some time until the new version is visible in Teams.
5. Step: Use the agent in Teams and ask questions to Customer Insights
Now that your agent is published and approved it will appear in the Teams app store. Users can install it themselves or you can also set up a policy that installs and pin the agent to chosen users.
No matter which way you choose in the end you and your users can now use the agent.
Summary
An there you have it! You’ve built your first agent with Copilot Studio for Customer Insights and integrated it into Microsoft Teams. It might not be the perfect agent and it certainly needs to learn a little bit more.
But as you can see, building an agent doesn’t need to be intimidating. Starting with a focused purpose—like helping users transition to real-time marketing in Dynamics 365 Customer Insights Journeys—keeps things manageable and impactful. Plus, embedding it in a tool like Teams ensures it’s readily available where users are already collaborating. While this was a simple example, it’s a powerful starting point.
There is a lot more to explore and think about, like security and fallback actions. Also we can extend the agent with triggers, actions or skills in the future. But I hope this project introduces you to the possibilities of Microsoft Copilot Studio, opening the door to creating more sophisticated agents in the future. We might dive more into this in future blog posts.
Until then, have fun exploring, and happy agent building!
***Please be aware: The content is accurate at the time of creation. It may be that Microsoft has made changes in the meantime.***
Check out the FAQ section of my blog as well: Short questions with quick answers! Go to FAQs
