How do you train a custom model in AI Builder?

The training process for a custom AI Builder model involves selecting a model type, providing a labeled dataset, defining evaluation metrics, and publishing the model after validation. The graphical interface allows this without coding.

How can AI Builder models be integrated with Power Automate or Power Apps?

AI Builder models can be integrated directly into Power Automate flows through native actions, or in canvas and model-driven apps via controls and connectors. The Maker Portal allows managing integration and lifecycle of these models.

AI Builder: Custom Models

Q: What types of custom models does AI Builder offer?

AI Builder offers custom models such as prediction, category classification, image classification, entity extraction, object detection, and document processing. These models can be trained with organization-specific datasets to tailor AI capabilities.

Training, datasets, evaluation, and limits to create tailored AI models integrated into Power Platform

Introduction to AI Builder Custom Models

AI Builder is a core component of the Microsoft Power Platform that enables the creation of artificial intelligence solutions in a no-code environment. Custom models allow training AI on organization-specific data, turning complex tasks into intelligent automated processes.

The goal of custom models is to enable predictive or classification scenarios tailored to specific business needs, without requiring machine learning expertise. The graphical interface integrated into the Power Apps Maker Portal allows configuring, training, and publishing models intuitively.

Available Custom Model Types

According to official documentation and the reference “Microsoft Power Platform Enterprise Architecture”, AI Builder allows the creation of the following custom models:

Prediction: to forecast future results based on historical data.
Category Classification: to classify textual or document content into defined categories.
Image Classification: to recognize and categorize images based on trained labels.
Entity Extraction: to identify key entities within text or documents.
Object Detection: to detect specific objects in images or videos.
Document Processing: to automate structured information extraction from documents.

These models can be customized through user-provided training datasets, improving accuracy and adaptability compared to prebuilt models.

Creating and Training Models

The process begins in the Power Apps Maker Portal. The user selects the model type, defines the dataset, and starts the training phase. This phase is crucial to optimize the predictive performance of the model.

During training, AI Builder uses Microsoft-managed machine learning algorithms to analyze patterns and build predictive models. Users can monitor the training status and view metrics like accuracy, precision, and recall.

Once completed, the model can be tested and, if it meets quality criteria, published for use in Power Automate, Power Apps, or Power Pages.

Flow diagram of AI Builder model creation and training process.

Datasets and Data Quality

The quality of the dataset directly affects the model’s performance. Accurate and balanced labeled data is required to avoid bias. AI Builder supports importing data from Microsoft Dataverse, Excel, or other supported sources.

The system provides preview and validation tools to identify anomalies or missing values. It’s a best practice to maintain a separate validation dataset not used in training to test the model’s generalization ability.

For more on data management and dataset preparation, visit Microsoft’s AI Builder documentation.

Evaluation and Publication

Once trained, the model must be evaluated using quantitative metrics. AI Builder provides an evaluation panel showing overall accuracy and error distribution.

Only after achieving satisfactory performance should the model be published. Publication makes the model available as a service that can be integrated into other Power Platform solutions. This can be done directly in the Maker Portal with a few clicks.

A published model can be reused across multiple applications, ensuring consistency and centralized AI capabilities within the organization.

Integration with Power Platform

Published AI Builder models can be integrated into:

Power Automate: via AI Builder actions to automate recognition or prediction-based flows.
Power Apps (canvas or model-driven): embedding AI controls directly into the UI.
Power Pages: to analyze user-uploaded content on public portals.
Dataverse: as part of solutions combining automation and centralized data.

These integrations enrich business processes with predictive and analytical capabilities, without requiring dedicated AI infrastructure.

Limits and Considerations

Despite being a powerful no-code platform, AI Builder has technical limits to consider:

Maximum dataset size and training duration.
Maximum number of active models per Dataverse environment.
Dependence on Microsoft cloud infrastructure for processing.
Licensing restrictions based on Power Platform or Dynamics 365 plan.

Monitoring usage via the Power Platform Admin Center is recommended, and capacity upgrades may be needed for large-scale AI projects.

Frequently Asked Questions about AI Builder Custom Models

What types of custom models does AI Builder offer?

AI Builder supports custom models for prediction, classification, object detection, entity extraction, and document analysis. Each model can be adapted to specific business data.

How is a model trained?

Training is done directly in the Maker Portal by uploading data and starting the training process. Metrics and results can be reviewed before publication.

Are there usage limits for AI Builder models?

Yes, Microsoft imposes limits on dataset size, number of models, and compute capacity based on your Power Platform license.

Start Building Your Custom AI Model

Access the Power Apps Maker Portal and discover how AI Builder can turn your data into predictive value. Leverage the power of integrated artificial intelligence within Power Platform.

Explore PL‑400 Course All Power Platform Paths