Get started with Generative AI - Copilot creation

  • 9 minutes

You'll likely be excited to get started with building a bot using Microsoft Copilot Studio. This first lab, Planning your first chatbot, has been written to help provide guidance on how you and your organization can get started thinking about what problems your bot can solve and where it can be deployed. We understand that you want to begin right away, however in this section you'll learn why it's important to first see what can be achieved with Generative AI in Copilot Studio as it might determine what topics you choose to manually author in the studio's authoring canvas.

Section 1: Sign in to create a Copilot

Your first task is to sign in so that you can create your copilot.

  1. Go to https://copilotstudio.microsoft.com/ and type in your email that you'll use to build the copilot. Sign in with your work, school, or developer account if you aren't working with an instructor.

    Note

    Your account might already have been set up. If screens similar to the following images don't show, you can skip to step 7.

  2. Select Next.

    Screenshot of Thank you for choosing Microsoft Copilot Studio with focus on the Continue button.

  3. Select Get started.

    Screenshot of Your account info with focus on the Get started button.

  4. Select Get started under Confirmation details.

    Screenshot of Confirmation details with focus on the Get started button.

  5. Select your country/region and then select Start free trial.

    Screenshot of the Welcome to Microsoft Copilot Studio screen, showing the country/region selected and the Start free trial button.

    Important

    Skip to here if your account was already set up.

  6. You should be redirected to copilotstudio.microsoft.com. If not, go to the maker environment URL for Microsoft Copilot Studio: copilotstudio.microsoft.com. An Environment selector box is located in the upper-right corner. An environment is where your organization stores, manages, and shares the copilot, business data, apps, and Microsoft Power Automate flows. You can select the Environment selector button to display the copilots within that environment.

    Screenshot of the Environment selector button.

  7. If you're working with an instructor, confirm now with your instructor the environment to use, if you didn't already do so. If you aren't working with an instructor, and you're using your own environment, for the purposes of this course, you can use the default environment. Alternatively, if your tenant allows, an admin can create your own environment to use in admin.powerplatform.com.

    For more information, see Manage default environment.

Section 2: Create a Copilot

  1. From the Microsoft Copilot Studio Home page, describe your copilot to create it. Put in a description like the following: I want a copilot named Contoso Customer Service. It's an assistant for Contoso customers, helping to answer questions and help with common tasks like checking an order status.

    Screenshot of describe your copilot to create it window.

  2. You will get redirected to an optional conversational experience to further customize your copilot. You can provide further details to the description and you can also provide a tone of voice.

  3. Start by adding a tone of voice to the copilot. Type Playful tone, joyful, customer focus, but definitely professional in the chat box.

    Screenshot of adding the copilot a tone of voice.

  4. Next, you'll be asked to provide sources for important information. We'll be using publicly accessible websites. Type Information should come from https://learn.microsoft.com/en-us/microsoft-copilot-studio/ and from https://www.microsoft.com/en-us/microsoft-copilot/ in the chat box.

  5. You will also get asked for topics or tasks the copilot shouldn’t help with or talk about. Type We don't want to discuss other brands like Fabrikam. Never provide product comparisons with competitor technologies. in the chat box.

  6. In the right side of the window, you'll see a summary pane that displays all the parameters and information you provided to the copilot.

    Screenshot of summary pane of copilot creation.

  7. Next, above the summary pane, select the ellipses (...) and select Edit advanced settings.

    Screenshot of advanced settings drop-down.

  8. In the Advanced Settings window, change the Schema Name from the default copilot, to ContosoCustomerService.

    Screenshot of advanced settings window.

  9. Finally, select I'm done in the chatbox, then select Create in the top-right corner.

    Note

    After you've selected Create, the process of creating the first copilot within a new environment can take up to 15 minutes. Subsequent copilots are created much faster.

    Note

    After creating the copilot, a pop-up window may appear showing new features in copilot Studio. To continue past this, click Next three times and Done at the end.

  10. After a few minutes, you will be redirected to your copilot's Overview page.

    Screenshot of result of copilot creation.

    After you create your copilot, it appears in the list under the Copilots menu in the left navigation bar.

    Tip

    If you've created a copilot in this environment before, to create another, select the Copilots menu item on the navigation bar and then select New Copilot.

  11. Within the Copilots menu, you can select from the copilots that are available to you. Selecting the ellipses next to a copilot brings up drop-down with controls like open the copilot, export the copilot, and delete. You can also select the name of the copilot, which opens it.

    Screenshot of Copilot menu options.

Section 3: Take a quick tour of the user interface

Microsoft Copilot Studio makes it easier for you to build basic to advanced copilots. The following section reviews the main pages of the maker experience for Microsoft Copilot Studio.

Because the Copilot area was explained in the preceding section, the ensuing section will cover the main options that are available after you select a copilot.

Screenshot of Microsoft Copilot Studio navigation.

  • Overview - Displays your home page, including a summary of the copilot's Details, Knowledge, Topics, Actions, and Publish status.

  • Knowledge - Allows you to provide a knowledge base as grounding data for your copilot that enables it to pull information from specific data you provide.

  • Topics - Provides a list of all the Topics available to your copilot. These Topics are conversation paths you develop to determine how your copilot responds to specific scenarios.

  • Actions - Allows you to build Actions that your copilot can complete. These actions are similar to Power Automate flows like automating a database entry.

  • Analytics - Provides view metrics to monitor how well your copilot is operating. The various Analytics sections include: Customer satisfaction, Sessions, Billing, and Boost conversations.

  • Channels - Allows you to publish and connect your copilot with various websites, web-services, and integrations into your own website.

Section 3: Take a tour of Analytics

Microsoft Copilot Studio makes it easy to analyze the results of your copilot's sessions with users.

To view the analytics of your selected copilot, select Analytics from the top navigation bar of your copilot's home page.

Screenshot of the analytics page.

Note

The following screenshots display the analytics data from a sample copilot. When you first create a copilot, there is no session history to pull data from. To review analytics for your copilot, you will need users to interact with the copilot first.

When you first explore the Analytics section, you will be taken to the Summary page. This is where a summary of your copilot's sessions is displayed.

Screenshot of the summary page.

The Summary page displays key performance indicators (KPIs) from your copilot's session logs. These KPIs include:

  • Total sessions - The count of sessions, starting when a user sends a message and ending when the problem is solved, the chat is abandoned, or escalation is triggered.

  • Engagement rate - The percentage of sessions where a user-authored topic or escalation topic is triggered.

  • Resolution rate - The percentage of engaged sessions that are resolved, indicated by the user reaching the ‘end of conversation’ topic and responding positively or not at all to the survey.

  • Escalation rate - The percentage of engaged sessions ending in escalation.

  • Abandon rate - The percentage of engaged sessions abandoned within 1 hour of interaction.

  • CSAT - The average Customer Satisfaction score from surveys rated on a scale of 1-5.

Additional visuals in the Summary include graphs comparing Engagement over time and Session outcomes over time, as well as tables detailing the impact and rate drivers of Resolution, Escalation, and Abandon rates for each triggered topic.

Next to Summary is the Customer satisfaction Analytics page. This page displays the history of CSATs and the relationship between CSATs triggered by different Topics.

Screenshot of the Analytics Customer satisfaction page.

The Sessions page makes the session logs for your copilot available for download. Sessions are only able to be pulled from within the past 28 days, and are limited to 7 days of data at a time.

Screenshot of the Analytics Sessions page.

Note

The session log is downloaded in CSV format, with the SessionID, StartDateTime, InitialUserMessage, TopicName, ChatTranscript, SessionOutcome, and TopicID fields represented for each individual session.

Screenshot of the session log downloaded from the Sessions Analytics page.

The Billing page displays the total number of billable sessions and compares billed sessions over time.

Screenshot of the Analytics Billing page.

The Boost conversations page displays the analytics for individual questions answered through the OpenAI GPT service. The data covers both the rate of which GPT was able to answer questions outside of the copilot's grounding data, and compares the reasoning for questions going unanswered (filters, errors, etc.)

Screenshot of the Analytics Boost conversations page.

Section 4: Take a tour of Settings

The settings panel contains all the settings available to a copilot, allowing you to quickly make security changes or add skills.

Screenshot of settings window.

  • Copilot details – Where you can update the copilot display name, icon, and modify advanced settings (e.g. configure the Azure Application Insights integration)

  • Generative AI – Where you can choose to replace the more classic natural language understanding approach for topic triggering and entity extraction with one that’s based on a large language model to do multi-intent detection and more complex entity extraction. This is also where you can configure content moderation setting for knowledge sources (to reduce risks of hallucinations).

  • Security – Where you can share your copilot with other users (to co-author it) or with security groups (to use it). This is also where you configure end-user authentication settings (the type of authentication and whether it is enforced or not), and web channel security that allows you to further secure the Direct Line channel that is used for any web or custom application deployment.

  • Entities – Copilot Studio comes with a lot of pre-built entities to help identify key information in a user utterance (e.g. a city, date, number, etc.). This menu is also where you can define your own closed-list entities or regular expression entities.

  • Skills – Where you register external Bot Framework skills that your Copilot Studio copilot can call, or where you can configure how existing Azure Service Bot can use your Copilot Studio copilot as a skill.

  • Languages – Where you can configure additional languages your copilot can be used in and localized into.

  • Language understanding – Where you can configure custom language models developed and trained on Azure AI Language, in Azure Conversational Language Understanding (CLU). When configured, this effectively replaces the out-of-the-box natural language understanding model (NLU) for intent detection, and can also replace entity detection and extraction.