AI Builder in Power Apps

Analogy categoryAnalogies

What AI Builder Adds to Power Apps

AI Builder brings machine learning capabilities directly into Power Apps and Power Automate without requiring a data science background, offering prebuilt models for common tasks like reading forms, detecting objects, and analyzing sentiment. Makers add an AI Builder-specific control or call a model from Power Fx exactly the way they'd use any other data source.

🏏

Cricket analogy: AI Builder is like giving a club team access to a national selector's statistical model without needing to hire their own analytics department — sophisticated prediction made available with no data science staff.

Prebuilt Models vs Custom Models

Prebuilt models such as Business Card Reader, Text Recognition, Object Detection, and Sentiment Analysis work immediately with no training, since Microsoft trained them on broad, general-purpose datasets. Custom models — form processing, object detection, prediction, and category classification — are trained inside AI Builder studio on your own labeled sample data, tailoring accuracy to your specific documents, objects, or historical outcomes.

🏏

Cricket analogy: A prebuilt model like Business Card Reader is like using a franchise's standard fielding drill straight out of the manual, while a custom model is like designing a bespoke net session tailored to your bowler's specific weakness.

powerfx

// Power Fx: call an AI Builder prediction model bound to a form's submit
Set(
    varPrediction,
    'Loan Default Predictor'.Predict(
        {
            Income: Value(TextInput_Income.Text),
            LoanAmount: Value(TextInput_LoanAmount.Text),
            CreditScore: Value(TextInput_CreditScore.Text)
        }
    )
);
UpdateContext({ locRiskLevel: varPrediction.Predictions."Risk Level" })

Consuming AI Builder Results in Power Fx

AI Builder controls, like the Business Card Reader or Object Detector, can bind directly to a device camera or an uploaded attachment, extracting structured fields such as VendorName and TotalAmount that the app reads as properties. Custom prediction and classification models are instead called with ModelName.Predict(...), which returns structured output including a confidence score the app can use to flag uncertain results for manual review.

🏏

Cricket analogy: Binding a Business Card Reader control to a camera in the app is like a scorer photographing the scoreboard at the end of an innings — the app extracts structured data automatically instead of manual data entry.

AI Builder usage is metered by consumption credits: prebuilt model calls and custom model predictions each consume credits per document or transaction, drawn from capacity purchased or included with licensing for the environment. Monitor credit usage in the Power Platform admin center to avoid unexpectedly hitting the environment's cap in production.

Common Use Cases and Model Training

Typical use cases include invoice and receipt processing, ID/document verification, object counting on a shop floor, and predicting outcomes like loan approval or customer churn from historical Dataverse records. Training a custom prediction or classification model requires a sufficient number of labeled historical rows — records tagged with the actual known outcome — before AI Builder can calculate a trustworthy accuracy score.

🏏

Cricket analogy: Training a custom prediction model to flag likely no-shows for nets sessions is like a franchise analyst feeding years of attendance data into a model until it reliably forecasts who's likely to skip training.

Model accuracy depends heavily on the quality, quantity, and diversity of labeled training data — sparse or biased samples produce unreliable predictions that can quietly skew business decisions. Always review each model's quick-test accuracy score before publishing, and account for AI Builder credit consumption limits when planning production rollout.

AI Builder brings prebuilt models — Business Card Reader, Text Recognition, Object Detection, Sentiment Analysis — into apps with no training required.
Custom models (form processing, object detection, prediction, category classification) train on your own labeled sample data inside AI Builder studio.
Prediction models need a minimum number of labeled historical records with known outcomes before they can be trained.
Models are called via ModelName.Predict(...) in Power Fx or bound directly to AI Builder-specific controls.
Usage is metered by consumption credits per prediction/document, requiring provisioned capacity in the environment.
AI Builder controls can bind straight to a camera or uploaded file to extract structured data automatically.
Model accuracy depends heavily on the quality, quantity, and diversity of labeled training data.