What You Will Learn

• Build a functional AI agent in Copilot Studio from scratch
• Add and manage knowledge sources (documents, SharePoint, web)
• Design conversation flows, topics, entities, and confirmations
• Create adaptive cards and integrate with Power Automate for order emails

Study Guide

Introduction

Welcome to the comprehensive guide on building an AI agent using Copilot Studio. This course is designed to empower you with the knowledge and skills to create a functional AI agent for a fictional pizza shop, "Slice Slice Baby." The course is not just about building an agent; it’s about understanding the potential of AI in automating business processes, enhancing customer interactions, and ultimately driving efficiency. By the end of this course, you'll be equipped to create an AI agent that can handle customer orders, provide information, and integrate seamlessly with other tools, all without requiring deep technical expertise.

Introduction to Copilot Studio

Copilot Studio is a platform that democratizes the creation of AI agents for businesses. With its user-friendly interface, it allows users to start building an agent by simply describing its purpose. Imagine a scenario where your favorite pizza place, "Slice Slice Baby," faces staffing challenges. By leveraging Copilot Studio, you can build a co-pilot agent to assist customers in ordering pizzas, with orders emailed directly to the chef. You can access the platform at copilotstudio.microsoft.com using a Microsoft 365 account with a Copilot license.

For example, when you log in, you can initiate the creation of an agent by describing its functionality. Type, "We own a pizza shop called Slice Slice Baby, and we want to build a co-pilot agent to handle new customer orders. Customers choose the type of pizza they want, including the base, cheese, and toppings, and the order is emailed to our chef." Copilot will suggest a name for the agent, like "Pizza Order Assistant," and confirm the instructions.

Initiating an Agent with AI Assistance

One of the standout features of Copilot Studio is its ability to assist in building an agent by providing suggestions and automating the initial setup. This is particularly beneficial for users who may not have extensive coding knowledge. By simply describing the agent's functionality, Copilot can generate a basic configuration.

For instance, when you input the purpose of your agent, Copilot might suggest a name and provide a preliminary setup for your agent. This step is crucial as it sets the foundation for your AI agent, ensuring it aligns with your business needs. The initial configuration can include basic conversation flows and default responses, which you can customize further.

Adding Knowledge Sources

To ensure your AI agent provides accurate and specific information, it needs to be fed with relevant knowledge. Copilot Studio offers various options for adding knowledge sources, such as public websites, SharePoint sites, Dataverse, and uploaded documents. Initially, the agent might rely on its general knowledge, but this can lead to generic answers. To avoid this, you can disable the agent's general knowledge and rely solely on the provided sources.

For example, you can upload a Word document containing details about your pizza shop's address, opening hours, delivery information, and payment methods. This ensures the agent provides precise answers to customer inquiries. Another example could be linking a SharePoint site that contains your menu and special offers, allowing the agent to access up-to-date information.

Customizing Greetings and Conversation Flow (Topics)

The default greetings of an AI agent can often sound robotic. Copilot Studio allows you to customize these greetings through "Topics," making the interaction more engaging and aligned with your business's brand voice. System topics like "Conversation Start" can be edited to create a more friendly and business-specific greeting.

For example, you can change the default greeting to: "Hiya, how are you doing? That pizza last night, by the way, was fantastic! Pineapple on a pizza, okay, I get it! Anyway... I am here to help you order one of our delicious pizzas. Does what it says on the tin..." This personalization helps in creating a more human-like interaction, enhancing customer experience.

Topics also define the conversation flow and how the agent responds to user input. The agent looks for matching phrases within topics to determine the appropriate response. If no specific topic is found, it can fall back to using its knowledge sources. For instance, you might create a topic for handling inquiries about delivery times, ensuring the agent responds with accurate information.

Building Custom Topics for Pizza Ordering

Creating custom topics for specific tasks, like pizza ordering, allows for a structured conversation to gather necessary details. This involves defining "trigger phrases" that initiate the topic, such as "pizza order." The conversation flow is built using nodes, which include asking questions with different response types like multiple choice or entities.

For example, the first key question could be whether the order is for collection or delivery, with subsequent questions branching based on the user's choice. You might ask, "Great, we make delicious pizzas. Can I ask if this order is for collection or delivery?" Depending on the response, the agent can then inquire about collection time or delivery address.

Another example could involve asking about the type of pizza the customer wants. By defining trigger phrases like "order pizza" or "I'd like a pizza," the agent can seamlessly guide the customer through the ordering process.

Utilizing Entities for Data Extraction

When asking open-ended questions, such as collection time or delivery address, Copilot Studio uses "entities" to intelligently extract relevant information from the user's response. Built-in entities, like "time," "street address," or "person's name," can be used to capture specific data points.

For instance, if a customer provides a delivery address, the "street address" entity can extract this information accurately. Similarly, if you ask for a collection time, the "time" entity can interpret and record the customer's response.

Another practical application is using the "user's entire response" option for capturing detailed information, such as special instructions for the order. This ensures that no critical information is missed during the interaction.

Leveraging Adaptive Cards for Structured Input

For questions with multiple options, such as pizza base, cheese, and size, Adaptive Cards provide a user-friendly, structured interface. While Adaptive Cards use JSON-based code, Copilot can assist in generating this code based on a description of the required questions and options.

For example, you can use Copilot chat to create an Adaptive Card that asks about the pizza base, cheese, and size options. By describing the questions and choices, Copilot generates the JSON code needed for the card. This empowers users without coding knowledge to implement complex input methods.

Another example could involve creating an Adaptive Card for drink options, allowing customers to choose from a list of beverages to accompany their pizza order. This structured approach simplifies the ordering process for customers.

Order Confirmation and Error Handling

Confirming the order with the customer is crucial to avoid mistakes. This can be achieved by repeating the order details and asking for confirmation. Variables are used to store the customer's choices throughout the conversation, which can then be used to construct the confirmation message.

For instance, you might say, "So to confirm, you've ordered a base of {variable:pizza base}, cheese {variable:cheese}, size {variable:size}. Is this correct?" Conditions can be implemented to handle different responses, such as "yes" to confirm or "no" to go back and correct the order.

Another example of error handling involves using "topic management" to redirect the conversation flow back to previous steps if the customer indicates an error. This ensures the accuracy of the order and enhances customer satisfaction.

Integrating with Power Automate for Order Processing (Email Notification)

To automate the process of notifying the pizza shop of new orders, Copilot Studio can be integrated with Power Automate. This involves creating a Power Automate flow triggered by the Copilot agent. Inputs are defined in the flow to receive the order details from the Copilot agent's variables.

For example, an action can be added to send an email using Office 365 Outlook, with the order details included in the email body. The flow is then linked to the Copilot agent, mapping the agent's variables to the flow's inputs. This enables real-time notification of new orders without manual intervention.

Another practical application could involve integrating with a customer relationship management (CRM) system to log customer interactions and order history, providing valuable insights for future marketing efforts.

Publishing the Co-pilot Agent

Once the agent is built and tested, it can be published to various channels. Publishing options include Teams, demo websites, custom websites, and more. Before publishing, security and authentication settings need to be configured appropriately.

For instance, for the pizza ordering agent, "no authentication" might be set as it's assumed to be a public-facing service. This ensures that customers can easily access the agent without needing to log in.

Another example could involve embedding the agent on the business's website, allowing customers to interact with it directly from the homepage. This accessibility enhances customer engagement and streamlines the ordering process.

Conclusion

Congratulations! You have now completed the course on building an AI agent in Copilot Studio from scratch. By following this guide, you have learned how to initiate an agent with AI assistance, add knowledge sources, customize conversation flows, and integrate with other tools like Power Automate. These skills empower you to create AI agents that enhance business operations, improve customer interactions, and drive efficiency.

Remember, the thoughtful application of these skills is key. As you implement AI agents in your business, consider how they can best serve your customers and align with your brand's voice. With Copilot Studio, building AI agents is not just accessible but also a powerful way to transform your business.

Podcast

There'll soon be a podcast available for this course.

Frequently Asked Questions

Welcome to the FAQ section for the 'Video Course: How to Build an AI Agent in Copilot Studio from Scratch.' This resource is designed to answer your most pressing questions about building AI agents, from basic concepts to advanced techniques. Whether you're a beginner or a seasoned professional, you'll find valuable insights to help you make the most of Copilot Studio.

What is a co-pilot agent, and how can it benefit a small business like a pizza shop?

A co-pilot agent is an AI-powered virtual assistant that can interact with customers to automate tasks and improve efficiency. For a small business like a pizza shop, a co-pilot agent built with Copilot Studio can handle customer orders, answer basic inquiries about opening hours, delivery areas, and payment options, and relay order details directly to the kitchen. This frees up staff to focus on other critical tasks, reduces wait times for customers, and ensures orders are taken accurately, even when the shop is busy or short-staffed.

Is building an AI agent with Copilot Studio a complex and expensive process, especially for someone with limited technical knowledge?

Copilot Studio is designed to be user-friendly and accessible, even for individuals without extensive coding or AI expertise. The platform offers both a guided approach where you can describe the agent you want to build and have Copilot assist with the initial setup, as well as a more manual configuration option. While there might be licensing costs associated with Microsoft 365 and Copilot, the video demonstrates that a basic agent can be created relatively quickly and easily using a standard user account with a Copilot license. The visual, flowchart-based interface for designing conversation flows also helps to simplify the process.

How does a co-pilot agent learn about a business, such as a pizza shop's menu, delivery policies, and opening hours?

Copilot Studio uses a feature called "knowledge" to provide the AI agent with information about a business. This knowledge can be sourced from various locations, including public websites, SharePoint sites, and uploaded documents. In the example, a simple Word document containing details about the pizza shop's address, opening hours, delivery information, and payment methods is uploaded. The co-pilot agent then uses this information to answer customer queries accurately. It's also possible to switch off the agent's general AI knowledge to ensure it relies solely on the provided business-specific data.

Can the conversational tone of a co-pilot agent be customised to match a business's brand and customer service style?

Yes, the greetings and responses of a co-pilot agent are customisable through "topics" in Copilot Studio. System topics, such as the initial greeting ("conversation start") and acknowledgements ("thank you"), can be easily edited to reflect a more friendly or specific tone. By modifying the text within these topics, businesses can ensure the AI agent communicates with customers in a way that aligns with their brand voice, moving away from generic, robotic-sounding interactions.

How are complex interactions, such as taking a full pizza order with different bases, cheeses, toppings, and sizes, handled by a co-pilot agent?

For more intricate interactions, Copilot Studio utilises "adaptive cards." These are interactive elements that can be embedded within the conversation flow to present customers with clear choices and gather specific information. While the underlying structure of adaptive cards involves code (JSON), Copilot can assist in generating this code based on a description of the questions and options required. This allows businesses to create structured order forms directly within the chat interface, making it easy for customers to customise their orders.

How can a business ensure the accuracy of customer orders taken by a co-pilot agent and allow for corrections?

To mitigate errors, the conversation flow can be designed to include a confirmation step. After the customer has specified their order, the co-pilot agent can summarise the details (base, cheese, size, toppings) and ask for confirmation. This is achieved by using "variables" to store the customer's choices and then displaying these variables back to them. If the customer indicates an error, "conditions" and "topic management" features can be used to loop back to the relevant question, allowing them to correct their selection.

How are the orders placed through the co-pilot agent actually received and processed by the pizza shop?

Copilot Studio can be integrated with other Microsoft Power Platform services, such as Power Automate, to automate actions based on customer interactions. In the example, a Power Automate flow is created to automatically send an email containing the order details to the chef. This flow is triggered at the end of the pizza ordering conversation and uses the variables captured during the interaction (base, cheese, size, toppings, customer name) to populate the email content. This ensures that new orders are promptly communicated to the kitchen without manual intervention.

Once a co-pilot agent is built, how can it be made available to customers?

Copilot Studio offers various publishing options to make the AI agent accessible to customers. These include embedding the agent on a website for testing or as a custom integration, integrating it with Microsoft Teams, or even creating a dedicated demo website. The choice of publishing method will depend on where the business wants to interact with its customers. Before publishing, it's important to configure security and authentication settings appropriately for the intended audience.

What are the potential benefits and challenges of implementing an AI co-pilot agent for a small business like Slice Slice Baby?

Implementing an AI co-pilot agent can significantly enhance customer service and operational efficiency. Benefits include 24/7 availability, reduced waiting times, and freeing up staff for other tasks. However, challenges may include initial setup costs, the need for ongoing maintenance, and ensuring the AI agent accurately reflects the business's brand and service style. Additionally, businesses must handle customer data responsibly and comply with privacy regulations.

How do "knowledge" and custom "topics" work together to create a comprehensive conversational experience?

"Knowledge" provides the AI agent with specific information about the business, while "custom topics" define the conversation paths the agent can handle. These elements work together to ensure the agent can accurately answer customer queries and guide interactions. For instance, knowledge might include detailed product information, while custom topics handle specific tasks like order processing, ensuring a seamless and informative customer experience.

How does Copilot Studio empower non-technical users, and what limitations might still exist?

Copilot Studio's user-friendly interface and guided setup process make it accessible to non-technical users. The platform's visual tools allow users to design conversation flows without coding, and its integration with Microsoft services simplifies deployment. However, some limitations may include the learning curve associated with understanding AI concepts, potential customization constraints, and the need for technical support for more complex integrations or troubleshooting.

What are the key steps involved in designing a conversational flow for a specific task within Copilot Studio?

Designing a conversational flow involves several key steps: identifying trigger phrases that initiate the conversation, defining the conversation path using nodes (questions, conditions, and actions), and incorporating variables to store user input. The flow should include confirmation steps to ensure accuracy and a final action, such as sending order details to the kitchen. Testing and refining the flow is crucial to ensure a smooth user experience.

How does integration with Microsoft services like Power Automate enhance the capabilities of the AI agent?

Integration with Microsoft services like Power Automate allows the AI agent to perform automated actions based on customer interactions. This enhances capabilities by enabling seamless workflows, such as automatically sending order details to the kitchen or updating inventory systems. Additional integrations, like with Microsoft Teams or Dynamics 365, can further extend the agent's functionality, providing a comprehensive solution for business operations.

What are "entities," and how are they used in building a co-pilot agent?

Entities are data types that help the AI agent understand and extract specific pieces of information from user input. They play a crucial role in identifying information like dates, times, or product names, enabling the agent to respond accurately to customer queries. For example, in a pizza ordering scenario, entities might be used to capture details about the order, such as size, toppings, or delivery time, ensuring precise order processing.

How are "adaptive cards" created and used in Copilot Studio?

Adaptive cards are interactive UI elements that present information or gather input in a structured way. In Copilot Studio, they are created using JSON code, which can be generated with the help of Copilot itself. These cards enhance the user experience by providing clear options and collecting detailed information, such as customer preferences in a pizza order, directly within the chat interface.

How are "conditions" and "variables" used in the conversation flow to ensure accuracy and flexibility?

Conditions create branching logic based on specific criteria, allowing the conversation to adapt to user responses. Variables store user input or generated data, enabling the agent to recall and use this information later in the conversation. Together, conditions and variables ensure that the conversation flow remains accurate and flexible, accommodating different user needs and scenarios.

How does Power Automate send orders to the kitchen via email?

Power Automate is used to create an automated workflow that sends order details to the kitchen via email. The process involves defining inputs such as order specifics and customer information, adding an action to generate an email with these details, and setting the workflow to trigger upon completion of the order conversation. This ensures prompt communication of new orders without manual intervention.

How can security and authentication be managed when deploying a co-pilot agent?

Security and authentication are critical when deploying a co-pilot agent to ensure that only authorized users can access the service. Copilot Studio offers tools to configure authentication settings, such as requiring user logins or integrating with existing authentication systems. Businesses should also consider data privacy regulations and implement measures to protect customer information, such as encrypting data and regularly updating security protocols.

What are the common challenges in implementing an AI agent, and how can they be overcome?

Common challenges include ensuring the AI agent accurately represents the business, managing customer data responsibly, and integrating with existing systems. To overcome these challenges, businesses should invest in thorough testing and refinement of the agent, provide training for staff, and work closely with technical experts to ensure smooth integration and operation. Regular updates and feedback loops can also help improve the agent's performance over time.

What are the future trends in AI agent technology?

Future trends in AI agent technology include increased personalization, improved natural language processing, and greater integration with IoT devices. These advancements will enable AI agents to provide more tailored and context-aware interactions, enhancing customer experience and operational efficiency. Additionally, the growing focus on ethical AI and data privacy will shape the development and deployment of AI agents, ensuring they align with societal values and legal requirements.

How can businesses measure the success of an AI agent?

Businesses can measure the success of an AI agent through key performance indicators (KPIs) such as customer satisfaction, response time, and task completion rates. Regular analysis of these metrics can provide insights into the agent's performance and identify areas for improvement. Additionally, gathering customer feedback and monitoring engagement levels can help businesses understand the agent's impact on customer experience and operational efficiency.

How can AI agents enhance customer service?

AI agents can significantly enhance customer service by providing instant responses, handling routine inquiries, and offering 24/7 availability. They free up human staff to focus on more complex or personalized interactions, improving overall service quality. AI agents can also gather and analyze customer data to offer personalized recommendations, creating a more engaging and satisfying customer experience.