AI Builder in Power Platform: Practical Use Cases

AI Builder in Power Platform: Practical Use Cases

Introduction

[Explain democratization of AI via low-code; AI Builder models accelerate intelligent automation without data science expertise.]

Prerequisites

• Power Platform environment
• AI Builder capacity (trial or paid)
• Sample data for training custom models

AI Builder Model Types

Model Type	Use Case	Training Required
Form Processing	Extract data from invoices, receipts	Yes (custom)
Object Detection	Identify objects in images	Yes (custom)
Text Recognition (OCR)	Read text from images	No (prebuilt)
Sentiment Analysis	Analyze text sentiment	No (prebuilt)
Business Card Reader	Extract contact info	No (prebuilt)
Prediction	Forecast outcomes	Yes (custom)

Step-by-Step Guide

Step 1: Invoice Processing with Form Processing

Train Model:

1. Navigate to AI Builder → Create → Form processing
2. Upload 5+ sample invoices
3. Tag fields (Vendor, Total, Date, Invoice Number)
4. Train model

Use in Power Automate:

Trigger: When a file is created (SharePoint)
Action: Process and save information from forms
  - AI model: InvoiceProcessor
  - Form: File Content (from trigger)
Action: Create item (SharePoint List)
  - Vendor: outputs('AI_model')?['body/fields/Vendor']
  - Total: outputs('AI_model')?['body/fields/Total']

Step 2: Sentiment Analysis in Canvas App

Power Fx Expression:

Set(
    SentimentResult,
    'AI Builder Sentiment Analysis'.Predict(TextInput1.Text)
);

If(
    SentimentResult.Label = "Positive",
    Green,
    If(
        SentimentResult.Label = "Negative",
        Red,
        Gray
    )
)

Step 3: Custom Prediction Model

Scenario: Predict customer churn

Steps:

1. Prepare dataset (CustomerID, TenureMonths, SupportTickets, Churned)
2. AI Builder → Prediction model → Select Churned column
3. Train model
4. Publish model

Integration in Canvas App:

Set(
    ChurnPrediction,
    'Churn Prediction Model'.Predict(
        {
            TenureMonths: Value(TenureInput.Text),
            SupportTickets: Value(TicketsInput.Text)
        }
    )
);

Label1.Text = "Churn Risk: " & Text(ChurnPrediction.Score * 100, "0.00") & "%"

Step 4: Object Detection for Quality Control

Train Model:

1. Upload images of products (100+ images)
2. Tag defects (scratch, dent, crack)
3. Train and publish

Power Automate Flow:

Trigger: When an image is uploaded (OneDrive)
Action: Detect objects in images
  - AI model: QualityControlModel
  - Image: File Content
Condition: If object detection found "defect"
  - Yes: Send notification to quality team

Step 5: Receipt Processing

Prebuilt Model in Power Automate:

Action: Extract information from receipts
  - Receipt: File Content
Action: Create expense record (Dataverse)
  - MerchantName: outputs('Receipt')?['body/merchantName']
  - Total: outputs('Receipt')?['body/total']
  - Date: outputs('Receipt')?['body/transactionDate']

Step 6: Business Card Scanner in Canvas App

OnScan Property:

Set(
    ContactInfo,
    'AI Builder Business Card Reader'.Predict(CameraControl1.Photo)
);

Set(varName, ContactInfo.FullName);
Set(varEmail, ContactInfo.Email);
Set(varPhone, ContactInfo.MobilePhone);

Integration Patterns

Pattern 1: Document Approval Workflow

[Upload document → AI extracts data → Populate approval form → Manager reviews → Auto-create records]

Pattern 2: Customer Feedback Analysis

[Collect feedback via Forms → Sentiment analysis → Route negative feedback to support team → Dashboard aggregation]

Pattern 3: Predictive Maintenance

[IoT sensor data → Prediction model → Forecast equipment failure → Schedule maintenance work order]

Performance & Limits

• Form processing: 100 pages/month (trial), higher with capacity add-ons
• Prediction models: Retrain monthly or when accuracy degrades
• API throttling: Monitor AI Builder usage via admin center

Troubleshooting

Issue: Low model accuracy
Solution: Add more training data (50+ samples); ensure data variety

Issue: Form processing fails on new layout
Solution: Retrain model with new document samples

Issue: Prediction model confidence low
Solution: Review feature importance; add relevant input fields

Best Practices

• Start with prebuilt models before custom training
• Use at least 50 training samples for custom models
• Validate model outputs before automated actions
• Monitor model performance metrics over time

Key Takeaways

• AI Builder democratizes ML for citizen developers.
• Prebuilt models enable instant value (OCR, sentiment, receipts).
• Custom models require training data but unlock domain-specific automation.
• Integration with Power Automate and Canvas Apps is seamless.

Next Steps

• Implement document approval automation
• Build predictive analytics dashboard
• Explore GPT integration via AI Builder prompts (preview)

Additional Resources

• AI Builder Overview
• Form Processing
• Prediction Models

---

Which manual process will you automate with AI first?