Deploy Personalized Shopping Agent (Preview)

Preview sign-up

This release is in preview and available only to customers participating with the Microsoft product engineering team.

Fill out the Microsoft for Retail agent sign-up form to request access and entitle your tenant to Personalized Shopping Agent (Preview).

Prerequisites and environment setup

Before deploying Personalized Shopping Agent (Preview), ensure you have the following prerequisites in place:

• Microsoft Copilot Studio: Personalized Shopping Agent (Preview) is based on Microsoft Copilot Studio and requires admin access to install the agent on your tenant and enable end-user access across the medium of your choice (for example, web-based, app-based, or Microsoft 365–based). If you already have access to Microsoft 365 Copilot, Copilot Studio is included with the Microsoft 365 Copilot service at no extra cost. For trial purposes, you can use the free trial license of Microsoft Copilot Studio.

• Power Platform environment: The agent is built and hosted in a Power Platform environment (with Microsoft Dataverse) within your tenant. Use a sandbox or development environment and enable the Dataverse database. If you don’t have one, create a new environment via the Power Platform admin center. Keep in mind that Power Apps Developer Plan environments have limitations. For example, Dataflows can't be created in purely developer-plan environments, so a standard trial or sandbox environment is better for full functionality.

• User accounts and permissions: Perform the deployment as a user who is both a Copilot Studio admin/user and a Power Platform maker. Set up Dataflows in the Power Apps maker portal, and ensure the account has permission to create Dataflows to bring your data into the Dataverse index that powers Personalized Shopping Agent (Preview). This action requires a Power Platform license that includes Dataverse (for example, Power Apps per user or a trial license) and access to Copilot Studio. Put this account in the Copilot administrator role or an equivalent role to manage the agents.

• Governance and compliance: This solution uses AI and can affect customer-facing scenarios, so involve your IT governance team early:

  • Legal/regulatory compliance: Verify the agent complies with your company's policies and any industry rules governing AI use or data handling. Consult Microsoft guidance and your legal or compliance teams to confirm obligations for your scenario.

  • Data residency: Your Dataverse environment stores the ingested retail data, so be mindful of where that environment is hosted (EU, US, etc.) to comply with data residency requirements.

  • Security: Plan who has access to the agent and its data. The agent operates under the context of the user chatting with it. For instance, you could restrict access to internal employees in Microsoft Teams during a pilot; alternatively, embedding the agent on a public, customer-facing site would provide broader access.

  • Connection security: When connecting to external data sources via Dataflows or connectors, use secure service accounts or OAuth connections. Don't hard-code any credentials. All Power Platform connectors handle authentication via connection references, which you can manage securely in the environment.

Install Personalized Shopping Agent (Preview)

You can create Personalized Shopping Agent (Preview) in your environment using Copilot Studio. Microsoft provides a ready-made template for this agent, which greatly simplifies deployment. To install the agent:

1. Navigate to the Microsoft Copilot Studio and sign in.

   Note

   Copilot Studio can also be accessed via the Power Apps maker portal in some previews.

2. Select the correct environment from the top-right corner where you want to install the agent. Use the environment switcher if you need to move out of the default environment.

3. Select Create on the left navigation. Look under the section Install a managed agent and find the Personalized Shopping Agent (Preview) card.

   

4. Select the card for more details. The agent is provided as a managed solution that includes initial topics, flows, and an out-of-the-box connection to the personalization microservice. A single quick-click installation provisions required components such as the Dataverse database and microservice connections needed for the agent to operate.

5. Select Install to add the agent to your environment. You're prompted to confirm installation. The installation process guides you to add related details like connection references at the appropriate steps.

   • You see a preview banner since the solution is in preview. If your tenant is enrolled in the preview program, you can proceed with the installation.

   • You can select "Demo mode" to provision a demo data index for testing. This mode recreates the index with demo data and can take up to an hour. You can change the selection later from Copilot Studio after you have actual data ingested.

6. Verify that the agent appears in Copilot Studio’s agent list. Select it to open the detailed Agent Builder view. Here you see sections like Cover (Name, description, icon), Instructions, Sample utterances, Knowledge, Capabilities, and Topics. The template shows prefilled fields:

   • Name and description: The agent is named Personalized Shopping Agent (Preview) by default. You can rename it to something like Shopping Assistant if desired. The description explains its purpose, for example: “Helps users find and recommend products from our catalog through conversation.”

   • Instructions: The system prompts that define the agent’s behavior. The template includes a base set of instructions. For example, the template instructs the agent to greet politely, ask what the customer is looking for, and then use the product data to find matches. Review this text and adjust it to better fit your organization's tone. Changes to the instructions affect functionality.

   • Topics: The agent provides prebuilt conversation topics (dialog flows) that handle common scenarios. For example, there might be a Greeting topic (“Hello, how can I help you today?”) triggered when the conversation starts, or a Product Search topic that takes the user’s query and returns product suggestions, etc.

At this point, Personalized Shopping Agent (Preview) is deployed in your tenant environment; the next step is to import and index your product data so the agent can answer queries with actual product information.

Next steps

• Ingest data for Personalized Shopping Agent in Copilot Studio (Preview) to connect your product data and customize the agent behavior.