Lewis Intelligence

Book a discovery session

Turn AI Builder into Your Business’s Secret Weapon

,

—

8–12 minutes

How to Use AI Builder in Power Automate to Create Intelligent Workflows

Nathan Lewis – Founder of Lewis Intelligence

In today’s fast-paced digital landscape, organizations are constantly seeking ways to streamline processes, reduce manual effort, and deliver faster, more accurate results. One of the most powerful tools for achieving these goals is Microsoft Power Automate. By integrating AI Builder with Power Automate, you can transform ordinary automation into truly intelligent workflows. In this blog post, we will explore what AI Builder is, how it works within Power Automate, and why it essentially inserts intelligence into your processes enabling “agents” to execute complex tasks for you.

What Is AI Builder in Power Automate?

AI Builder is a no-code, low-code artificial intelligence platform from Microsoft that allows you to add AI capabilities directly into your Power Platform solutions, including Power Automate. Instead of hiring data scientists or writing custom machine learning code, you can leverage prebuilt AI models or create custom AI models with just a few clicks. AI Builder offers several types of models:

  • Form Processing: Automatically extract key fields from forms or documents (invoices, purchase orders, surveys).
  • Object Detection: Identify and count specific objects in images (products in a warehouse, items on a shelf).
  • Prediction: Forecast outcomes by analyzing historical data (customer churn, loan defaults, sales forecasts).
  • Text Classification: Categorize text into predefined categories (customer feedback sentiment, support ticket types).
  • Entity Extraction: Detect and pull out specific information from unstructured text (names, addresses, invoice numbers).
  • Business Card Reader: Automatically capture contact details from business cards.

By embedding these AI models into Power Automate, you enable every flow (workflow) to become “smarter.” In essence, AI Builder acts as a digital agent, making decisions and processing data on your behalf.

Why Add Intelligence to Your Workflows?

Automated Decision Making

With AI Builder, you can shift from rule-based automation to AI-driven decision making. Traditional flows rely on static conditions – if a document has a specific keyword, route it to a folder. However, AI Builder can analyze the content of a document, interpret context, extract key-value pairs, and make routing decisions without relying on rigid rules. This delivers a more flexible, adaptive workflow that can handle variations and exceptions.

Increased Accuracy

Manual data entry and simple rule-based logic are prone to errors. By integrating AI Builder models such as Form Processing or Text Classification, you significantly reduce the risk of human error. The AI model learns from examples and becomes more accurate over time. This improved accuracy can lead to better compliance, fewer rework cycles, and greater overall efficiency.

Scalability

When you build intelligent workflows, they can scale seamlessly as data volumes grow. For instance, if you receive hundreds of invoices per day, a Form Processing model can extract line items, totals, and vendor information at scale, something that would be impossible for a team of humans to match in speed and consistency.

Enhanced Customer Experience

AI-driven automation not only benefits internal operations but also enhances the end-user or customer experience. Chatbots powered by AI Builder can classify incoming support requests and route them to appropriate service agents. Automatic sentiment analysis can flag negative feedback in real time and trigger proactive outreach, improving customer satisfaction and retention.

Creating “Agents” for Your Processes

One of the most compelling aspects of AI Builder is that it turns your flows into digital agents. These agents can:

  1. Analyze: Read and interpret unstructured data (emails, documents, images).
  2. Decide: Use predictive models to forecast an outcome or classify content.
  3. Act: Trigger downstream actions (send notifications, update databases, create tasks).

In other words, your flows behave like intelligent virtual workers that operate 24/7, making decisions and executing tasks without human intervention.

Step-by-Step Guide: Adding AI Builder to Power Automate

Follow these steps to embed AI Builder models into your Power Automate workflows and create intelligent, agent-like automations.

1. Identify the Use Case

Before you begin, ask yourself:

  • What problem am I trying to solve?
  • Do I want to extract data from documents, classify text, detect objects, or make predictions?
  • Where in my current process does manual intervention slow things down or introduce errors?

For example, you may want to automate invoice processing, classify customer feedback, or predict sales leads that are most likely to convert.

2. Access AI Builder

  1. Sign in to Microsoft Power Automate (flow.microsoft.com) using your Microsoft 365 credentials.
  2. In the left-hand navigation pane, select AI Builder.
  3. Click Explore, and then choose either Prebuilt models or Build a custom model.

If you want to start quickly, select a prebuilt model such as Form Processing or Text Classification. If you have unique data and need a custom model, choose Build, pick the AI model type you need, and follow the guided steps to upload training data, tag fields, and train the model.

3. Create or Configure Your AI Model

  • Prebuilt Model (Form Processing example)
    1. Select Form Processing from the AI Builder menu.
    2. Provide sample documents (e.g., invoices, receipts) in PDF or JPEG format.
    3. Tag the fields you want to extract—invoice number, date, total amount, vendor name.
    4. Train the model. After training, AI Builder will give you a unique model ID.
  • Custom Prediction Model
    1. Choose Prediction from the AI Builder menu.
    2. Connect to a data source (Excel spreadsheet, Dataverse table, Azure SQL).
    3. Select the column representing the outcome you want to predict (e.g., “Sale Closed” = yes/no).
    4. Choose input columns (customer interactions, lead score, region).
    5. Train and evaluate the model. AI Builder will provide accuracy metrics and a model ID.

4. Build the Power Automate Flow

With your AI model ready, create a new flow and insert the AI “action” into it.

  1. In Power Automate, select Create > Automated cloud flow (trigger: when a new email arrives, when a file is created, or any other trigger).
  2. Add the appropriate trigger, for example:
    • When a file is created (OneDrive, SharePoint)
    • When an email arrives (Outlook)
  3. Click New step, search for AI Builder, and select the model you configured (e.g., Predict, Extract information from forms).
  4. Configure the AI action:
    • For Form Processing, choose your custom model from the dropdown, and set the “Document” field to the output of your trigger (e.g., file content or attachment).
    • For Prediction, select your model, set the “Record URL” or “Input data” using dynamic content from a previous step.
  5. After the AI action, add subsequent actions based on the AI output:
    • Condition: If the AI-predicted outcome is “Yes,” route the flow down one path; otherwise, route it to another.
    • Create item: Write extracted form fields into a SharePoint list or Dataverse table.
    • Send email: Notify stakeholders when an invoice is processed or a sentiment is negative.
    • Start an approval: If a document requires human validation, trigger an approval flow.

5. Test and Refine

  • Test with Real Data: Upload sample invoices or send test emails to ensure that the AI Builder action is extracting the correct fields or making accurate predictions.
  • Monitor Accuracy: For custom models, review model performance metrics in AI Builder periodically. Update training data or retrain the model as needed.
  • Adjust Logic: As you see AI outputs in your flow runs, you may need to tweak conditions, add error handling, or build additional steps for edge cases.

Real-World Examples of AI Builder in Power Automate

Automate Invoice Processing

  1. Trigger: When a new invoice PDF is uploaded to a SharePoint folder.
  2. AI Builder Action: Use Form Processing to extract invoice number, date, line items, and total amount.
  3. Conditional Logic: If the total amount exceeds a certain threshold, route to finance manager for approval; otherwise, post directly to accounting system.
  4. Output: Create a record in Dataverse or a SharePoint list that logs invoice details. Send a confirmation email to the vendor automatically.

Classify Customer Feedback

  1. Trigger: When an email with feedback arrives in the support mailbox.
  2. AI Builder Action: Use Text Classification to label feedback as “Positive,” “Negative,” or “Neutral.”
  3. Conditional Logic: If the sentiment is negative, create a high-priority support ticket in ServiceNow; if positive, send a “Thank You” email.
  4. Output: Logged feedback in Dataverse for analytics and trend reporting.

Predictive Lead Scoring

  1. Trigger: When a new lead is created in Dynamics 365.
  2. AI Builder Action: Use a Prediction model trained on historical sales data to assign a “Lead Score.”
  3. Conditional Logic: If lead score is above 80 percent, assign to Sales Representative A; if below, send to nurture campaign.
  4. Output: Create a follow-up task in Planner or Microsoft To Do, based on predicted outcome.

Best Practices for Intelligent Workflows

  1. Start Simple: If you are new to AI Builder, begin with prebuilt models like Form Processing or Business Card Reader. These quick wins help you understand how AI actions fit into flows.
  2. Gather Quality Training Data: For custom models, the quality of your training data directly impacts accuracy. Ensure documents or records are clearly labeled and representative of real-world scenarios.
  3. Use Dynamic Content Wisely: When mapping AI outputs into downstream actions, use dynamic content to avoid errors. For example, map extracted invoice fields precisely into corresponding columns in SharePoint or Dataverse.
  4. Monitor and Retrain: AI models can drift over time if your data patterns change. Schedule quarterly reviews of model performance and retrain as needed.
  5. Implement Error Handling: Not every document or email will be perfect. Build fallback paths in your flow—send an “Unprocessed” email alert to a mailbox, or route to a human reviewer for ambiguous cases.
  6. Document Your Flow: Add descriptive names and comments to each step in Power Automate so that others can understand the logic behind every AI action and subsequent step.

Benefits of AI-Driven Workflows in Power Automate

  • Time Savings: Tasks that once took hours of manual work can be completed in seconds. As the saying goes, “Time is money”—AI Builder helps you save both.
  • Reduced Errors: Human data entry errors are minimized when AI extracts and processes data directly from documents or text.
  • Better Compliance: Automated, standardized processes ensure every document is handled consistently, meeting audit and compliance requirements.
  • Scalable Operations: AI Builder models can handle large volumes of data and documents without fatigue—unlike human workers.
  • Improved Employee Satisfaction: By offloading repetitive tasks to AI, your team can focus on higher-value strategic work, boosting morale and productivity.
  • Competitive Advantage: Businesses that adopt AI-driven automation gain a competitive edge by becoming more agile, responsive, and data-driven.

Common Challenges and How to Overcome Them

Challenge: Selecting the Right AI Model

Solution: Review available prebuilt models first. If you need custom logic (e.g., a proprietary form layout), invest time in creating and training a custom model. Always validate with test data.

Challenge: Ensuring Data Privacy

Solution: Use secure connectors (Dataverse, Azure Blob Storage, SharePoint). Ensure sensitive documents are stored in trusted repositories. Leverage Azure AD security groups to control who can view AI outputs.

Challenge: Handling Exceptions

Solution: Build exception branches within your flow. For example, if the AI Builder action returns a low confidence score or fails to extract required data, route that record to a “needs review” queue. This prevents data from being lost or misrouted.

Challenge: Maintaining Model Accuracy Over Time

Solution: Plan a regular cadence for model retraining. Collect new examples of documents or labeled data, retrain your AI model, and deploy the updated version. Document version control helps you track improvements.

Leave a comment