AI Builder, a powerful feature in Microsoft Power Platform, allows you to introduce machine learning-based automation into your business processes. In a Power Pages scenario, you can use AI Builder to auto-categorize records (e.g., tickets, inquiries, feedback) based on content submitted via portal forms — saving time and increasing consistency.
Real-World Example
You run a customer support portal. When users submit a case or feedback, you want the system to auto-categorize it as:
- Product Inquiry
- Billing Issue
- Technical Support
- General Feedback
Instead of relying on users to choose the category, let AI Builder infer it intelligently based on the text they write.
Step-by-Step Guide
Step 1: Prepare Your Training Data
Before using AI Builder, you need a dataset that includes:
- Text descriptions or user input (e.g., case description)
- The correct categories (your “labels”)
Format:
| Description | Category |
|---|---|
| “I can’t access my invoice.” | Billing Issue |
| “How do I upgrade the software?” | Product Inquiry |
| “The app keeps crashing.” | Technical Support |
Use Dataverse table or Excel/CSV in OneDrive or SharePoint for training.
Step 2: Create a Category Classification Model
- Go to Power Apps → AI Builder → Explore
- Select Category Classification
- Upload or connect to your labeled data
- Define your text and category columns
- Train and publish the model
The model learns how to associate text input with correct categories.
Step 3: Create a Power Automate Flow
Once your AI model is trained and published:
Trigger:
When a row is added(Dataverse)- Choose your portal submission table
Actions:
- Use AI Builder: Select Predict action
- Choose your published model
- Input: User’s description or submitted text
- Extract Prediction Output:
- Result will include Top Category, Confidence Score
- Update Record in Dataverse:
- Set
Categoryfield to the predicted value
- Set
- (Optional) If confidence < 70%, assign to manual review.
Step 4: Display Category on Portal
- Add the
Categoryfield to views/forms - Allow users to see the categorized result
- Use conditional formatting (e.g., tags, colors) for clarity
Advanced Enhancements
- Train separate models for different languages or domains
- Combine with Sentiment Analysis for smarter routing
- Trigger approval if category = “Complaint”
- Add manual override option in portal
Security Tips
- Restrict access to model configurations
- Use environment variables to store model IDs
- Monitor flow usage and AI model performance over time
Use Case Ideas
- Classifying support tickets
- Routing feedback to correct teams
- Categorizing knowledge base content
- Auto-tagging documents or forms
- Segmenting customer issues for analytics