Building Copilot Agents with User Experience in Mind

Understand where and how the user will be accessing the Copilot agent

Create multiple-modal experiences for your Copilot using voice

To enhance the accessibility and user-friendliness of your copilot agent, consider developing a voice agent for voice interactions. This hands-free experience is not only inclusive but also particularly beneficial in customer scenarios where associates are using their hands and need to interact with the agent quickly and effectively without typing. Before deploying the agent to your user base, ensure thorough testing in the work environment to validate a good end-user experience, especially considering the loudness of the environment. For step-by-step instructions on testing the voice agent experience, you can view this blog

Determine the proper channel deployment

Selecting the appropriate channel for deployment is crucial. Ensure you reach users where they perform their work, whether it's through M365 Copilot, M365 Copilot chat, embedding into a website, or other channels. It's essential to deploy the agent to the channels that best suit the use case.

Creating multilingual agents

Creating multilingual agents is essential for providing a personalized and inclusive experience. By enabling your agent to interact in multiple languages, you can cater to a diverse user base and ensure effective communication without language barriers. This feature enhances accessibility, user satisfaction, and expands the agent's reach in global contexts.  For more specific information on multilingual agents see below.

 

Take Time During the Configuration to Set Up the Agent for Success

Creating a name for the agent

In a world where multi-agents are on the rise, it's essential to be as descriptive as possible in the name and description for the intended use. This helps users understand the agent's purpose and functionality. Planning out your multi-agent scenarios early and defining clear names is especially important to avoid confusion.

Add starter prompts

Starter prompts guide users on how to interact with the agent effectively. These prompts should be clear and concise. When selecting starter prompts, consider defining the most common use cases for the agent. For example, if you are creating an IT Support agent, you might include prompts like "create a support ticket" or "check request status" to cover the most asked questions.

Provide description and general instructions

A detailed description of the agent's capabilities and intended use helps users understand what they can expect from the interaction. This description also assists the language model in identifying and utilizing your agent for specific tasks or situations. Ensure that the description is concise yet informative.

For the instructions (declarative and generative mode), you should be as detailed as possible. Include the purpose, guidelines on what the agent should and should not do, tone, and the skills the agent will possess. Below are resources that provide detailed instructions on how to construct the verbiage effectively.

 

Add key knowledge resources

Adding key knowledge sources to the agent is crucial for its effectiveness and reliability. By integrating relevant documentation and connecting to essential business systems, the agent can provide accurate and timely information, ensuring users receive the support they need. These connections enable the agent to perform specific tasks efficiently, leveraging up-to-date data and resources. This not only enhances the user’s experience but also ensures the agent remains a valuable tool in various scenarios, from customer support to internal business operations.

 

Offer alternative ways to answer questions

Connecting to a live agent

Provide users with the option to connect to a live agent for more complex queries or issues that the agent cannot resolve.  Customizing the escalation topic is another way to track potential opportunities for agent improvement.  Creating a step in the topic to record the escalation to ensure it is captured for agent performance is key to ensuring agent success and continued feedback on areas of improvement.

Create adaptive cards

Using adaptive cards in a Copilot agent enhances user interaction by providing a visually appealing and interactive interface. They facilitate structured data collection, ensuring consistency and ease of processing. Adaptive cards offer customization and flexibility, allowing them to be tailored to various scenarios and use cases. They improve communication by presenting information clearly and concisely, reducing ambiguity. Additionally, adaptive cards integrate seamlessly across different platforms and devices, ensuring consistent user experience. Overall, they significantly enhance the functionality and user experience of Copilot agents.

Provide documentation references or links

In the agent's response, include links to additional resources. Offer links to documentation or external sites where users can find more detailed information.

 

Automate, Automate, Automate

Enhancing efficiency with automation

Using automation in Copilot agents (agent flows and agent actions) can significantly improve efficiency and reduce manual workload by identifying key processes that can be automated. Incorporating messages that inform users when the agent is processing their request helps manage user expectations and enhances the overall experience. By automating repetitive tasks and providing real-time updates, Copilot agents can streamline workflows, allowing users to focus on more complex and value-added activities. This approach not only boosts productivity but also ensures smoother and more transparent interaction with the agent.

 

Monitor Agents for Insights

Actively review the analytics for the agent

Regularly reviewing analytics is essential for assessing the agent's performance and identifying areas for improvement, ensuring a high-quality user experience. By analyzing metrics such as engagement outcomes, knowledge source usage, action success rates, and user feedback, you can gain valuable insights into how effectively your agent is meeting user needs. Resources like Microsoft Copilot Studio Analytic, Topic Analytic Autonomous agent health, and Application Insights telemetry provide detailed guidance on measuring and enhancing agent performance. For more information see the discussion on monitoring performance and effectiveness of M365 Copilot agents.

 

Collect Feedback

Integrate feedback mechanisms

Use adaptive cards to collect user feedback after every interaction with your agent. This allows you to assess and enhance the quality of agent responses while ensuring user satisfaction. Implementing this feedback mechanism helps in obtaining valuable insights into how well your agent is performing and identifying areas for improvement.

 

Conclusion

To create an effective Copilot agent, focus on providing easy accessibility by developing voice interactions and multilingual capabilities. Ensure you set up the agent properly in a world of multi-agents to decrease complexity and avoid confusion. Be creative in how you answer prompts, using adaptive cards and alternative ways to respond. Automate crucial business processes associated with the agent's purpose to enhance efficiency and reduce manual workload. Finally, actively collect feedback and apply it to continuously improve the user experience, ensuring the agent remains a valuable tool.

 

A special thanks to IvicaIvancic ((3) Create your first custom AI enabled Copilot Studio Voice Agent – step-by-step guide | LinkedIn)  and TrevorNorcross (Monitoring performance and effectiveness of Microsoft 365 Copilot agents | Microsoft Community Hub) for their incredible support in allowing me to link to their insightful blogs.