Unpacking the New Copilot Studio Workflows Designer
The new workflow designer in Copilot Studio offers a comprehensive toolkit. Multiple powerful blocks are available for users. These blocks allow for unprecedented automation capabilities. An agent block can invoke a full Copilot Studio agent. These agents come with their own tools and knowledge. A prompt block performs a single-focused LLM call. Classification routes AI-powered interactions into user-defined categories. M365 Copilot grounds AI processes directly in your Microsoft 365 data. This includes SharePoint, Teams, and OneDrive. Human review blocks pause workflows. They ensure a human is involved exactly when needed. Traditional building blocks also remain. Connectors, functions, variables, and loops are still essential. Each block serves a specific problem. They allow the creation of robust, intelligent **Copilot Studio workflows**.Beyond Manual Triage: Automating Shared Mailbox Emails
A common business challenge involves managing shared mailboxes. These mailboxes often receive a flood of diverse inquiries. Billing complaints, technical issues, and sales questions arrive daily. Traditionally, someone must manually open each email. They read it, figure out its purpose, and respond. This process is highly inefficient. It is also prone to human error. The new **Copilot Studio workflows** can intelligently automate this entire process. Human oversight is still integrated where it matters most. Imagine a “customer care” shared mailbox. Every new email triggers an automation. This eliminates manual sorting. The workflow begins with a trigger. By default, workflows start manually. However, this can be changed. Workflows can run on a schedule. They can also be triggered by an external service. This opens up the entire Power Platform connector ecosystem. For agents built in Copilot Studio, a dedicated trigger exists. It activates “When an agent calls the flow.” In our scenario, a shared mailbox trigger is chosen. This automation starts when a new email arrives.Intelligent Sentiment Analysis with AI Prompts
The first crucial step in processing an incoming email is understanding its sentiment. This information is vital for subsequent classification paths. An AI prompt is used for this purpose. The prompt settings are highly configurable. Various language models can be selected. Instructions are provided to the model. For instance, “Analyze the sentiment of the customer email below.” It must “Return ONLY a JSON object, nothing else.” Dynamic content feeds the prompt. The email subject and body are passed from the trigger action. The prompt’s output is structured. A property called “sentiment” is defined. This is an enum type. Its values can be “positive,” “neutral,” “negative,” or “urgent.” Another property, “sentiment reason,” is a string. This provides context for the detected sentiment. Testing the prompt is possible directly within the designer. Sample text can be provided. A “Love your product” subject yields a “positive” sentiment. A “Still waiting for a response” subject with frustration details results in “negative.” This instant feedback streamlines prompt refinement.Dynamic Email Classification and Routing
Following sentiment analysis, the email needs to be classified. The Classify block efficiently handles this. Users simply drag and drop this block. It requires setup within a full-screen experience. Both the subject and body of the email are fed into the classifier. This captures all main information for categorization. Different inquiry types flood mailboxes. Specific categories are created to manage this. For example, “Billing” covers invoices, payment issues, and refund requests. “Technical” includes bug reports, error messages, and login issues. “Sales” handles purchase questions and pricing inquiries. An “Other” category serves as a catch-all. It captures emails not matching predefined categories. Examples significantly improve classification accuracy. Users provide sample texts for each category. This trains the AI model. Testing verifies the classification. A sample inquiry might correctly classify as “Sales.” This intelligent routing streamlines the entire process.Harnessing Agents and Human Review in Copilot Studio Workflows
Once an email is classified, specific actions are taken. For a “Billing” classification, an agent can be called. A Copilot Studio agent, specifically a billing agent, is designed to handle these inquiries. These agents are often grounded in knowledge bases. For instance, a billing refund policy can inform the agent. The agent is called autonomously. A message is passed to it. This message instructs the agent to “process this customer billing email.” It also specifies information to return. A draft reply response is expected. The agent also indicates if a refund amount requires human approval. Multiple pieces of information are passed to the agent. Customer sentiment from the prompt action is included. The sentiment context or reason is also passed. The email subject and body complete the input. After the agent processes the request, a conditional check is performed. An If/Else action is added. It checks if approval is needed. The agent’s response indicates this. If “requires approval” is true, human review is initiated.The Power of Human-in-the-Loop Approvals
Human review is critical for sensitive tasks. It ensures compliance and accuracy. For refund approvals, a human is brought into the loop. A title like “Refund approval needed” is displayed. The message informs the approver: “Customer is requesting a refund that requires your approval.” The draft reply from the agent is presented. The potential refund amount is also shown. The approver is specified. This could be an individual user or a team. The approval channel is chosen. Outlook is a common choice for this scenario. An input is added for the approver’s decision. A dropdown provides “Yes” or “No” options. Notes can also be allowed. These allow the approver to add context. All updates are saved regularly. The workflow can then be published. This makes the changes live. The flow starts listening for new emails. Version history is maintained for the workflow. Users can compare, preview, and restore versions. This provides flexibility and control.Optimizing Layout and Responding to Customers
The workflow layout can be customized. Workflows often grow horizontally. Options at the bottom of the canvas are provided. Users can zoom in or out. The view can be fitted to the current screen. A “tidy up” function arranges blocks neatly. The layout can also be switched. It can be viewed vertically or horizontally. This visual flexibility helps users manage complex **Copilot Studio workflows**. After human review, another If/Else condition is used. It checks the approval decision. If “Approved” is “Yes,” an email response is sent. A shared mailbox connector sends this email. The recipient is the original sender. The original subject is included. The email body contains the agent’s draft reply. Copilot can generate the expression for this. It plugs the draft response directly into the email. This final step completes the automated, intelligent response cycle.Advanced AI Automation: M365 Copilot and Dynamic Q&A
Beyond basic classification, **Copilot Studio workflows** can leverage M365 Copilot. This block grounds AI interactions in your actual Microsoft 365 data. This includes information from SharePoint, Teams, and OneDrive. Imagine a “Technical” email. An M365 Copilot block is called. It uses this vast corporate knowledge. For a “Can’t log in this morning” issue, the AI can provide immediate troubleshooting steps. These are pulled from the company’s internal documentation. This ensures accurate and context-aware responses. For “Sales” inquiries, more dynamic processing might be needed. A prompt can extract multiple distinct questions. It reads the customer email. It then returns a JSON array of strings. Each string represents a single question. This structured output is crucial. A loop is then added. It iterates through each extracted question. Inside the loop, another prompt is used. This prompt acts as a knowledgeable sales representative. It answers each individual question concisely. Product information is provided to the prompt. This could be hardcoded for a demo. In a real scenario, it would come from a knowledge base.Building Dynamic Responses and Finalizing the Workflow
Within the loop, a compose action concatenates the question and its answer. This creates an HTML-style Q&A format. This composite content needs storage. A variable, initialized as an empty string, is used. Inside the loop, the variable is appended with the compose action’s output. This populates the variable with all questions and answers. After the loop completes, an email is sent. This uses the shared mailbox connector. The body of this email contains the populated variable. This delivers a comprehensive Q&A response. This demonstrates how easily **Copilot Studio workflows** can be built. They combine AI automation with traditional blocks. Connectors, loops, and variables are seamlessly integrated. This creates powerful, intelligent automation solutions.AI Automation Reimagined: Your Copilot Studio Workflow Questions Answered
What are Copilot Studio workflows?
Copilot Studio workflows are tools that allow you to create automated processes by combining artificial intelligence (AI) actions with traditional automation steps. They help build intelligent solutions for various tasks.
What is special about the new Copilot Studio workflows designer?
The new designer offers a unified environment where you can use AI-native actions directly alongside traditional automation blocks on a redesigned canvas. This makes it easier to build intelligent, automated solutions.
How can Copilot Studio workflows help with managing shared mailboxes?
Copilot Studio workflows can automate the entire process of managing shared mailbox emails, from categorizing inquiries and analyzing sentiment to drafting and sending responses, making the process much more efficient.
What is a ‘Human review’ block in Copilot Studio workflows?
‘Human review’ blocks are used to pause an automated workflow and involve a person to review, approve, or provide input on critical steps. This ensures accuracy and compliance for sensitive tasks.