AI Builder is the name of a new Power Platform capability that offers turnkey AI solutions based on Microsoft AI Technologies. It is a great point-and-click wrapper that enables non-developers to access the power of integrating some custom AI models on their Power Apps.

The AI Builder is part of the new 2019 release wave 2 set of features. It is now available on Preview on your Power Apps environments with an expect general availability of Oct’19. To enable the new AI Builder capabilities on an environment, you have to go to the Power Platform Admin Center https://admin.powerplatform.com and open your environment settings. There you select Features, and you will see a new option to enable AI Builder on your environment like in this screen shot. Full details available on the documentation.

Enabling AI Builder in the Power Platform Admin Centre

AI Builder offers four types of AI Models to use in your Power Apps. To create a new model for your Power Apps, you just need to choose the model type and follow the steps to configure it. The process is fairly easy, following the wizard you will select the settings for your AI Model, the data to use, and train it. There different AI Model types are:

  • Binary Classification: This model tries to predict a binary – e.g. yes/no – business outcome. You can choose a binary field as the target of the prediction and then use the rest of the data fields on an entity to try to predict the value. You can use it to predict customer churn, late payments, etc. Details on Binary Classification model.
  • Form Processing: This model wraps a Computer Vision model that allows you to extract key-value pairs and table data from documents. After you create the model, you can use it in your Canvas Power Apps with the Form Processor control, or with Flow. Details on Form Processing model.
  • Object Detection: Like the Form Processing model, this type helps you to build a model that can identify objects in images. It can be used to automatically identify products in retail scenarios, or part in field service applications. Once you train your model, you can use it in your Canvas Apps with the Object Detector component, or in Flow. Details on Object Detection Model.
  • Text Classification: The Text Classification model allows the tagging of text. You provide examples of text and labels, and the model will try to predict the labels. It can be a very helpful model to classify or route significant volumes of emails, surveys, or any other text. You use an entity in your CDS with the text and sample tags as the source; the model will generate a new entity to store the precited tags. You can also use Flow to classify text using a published model. Details on Text Classification Model.
AI Builder Model Types

Once you choose the model type and go through the wizard to build and train it, the model will be stored in your environment under the AI Builder models section. There you can see the training report showing its performance and quality, and you have options to modify or retraing the model. You can also compare your latest trained version with the published one, and publish your model.

Manage your Model

Obviously, to create a model you need data to train it. Classification examples, tagged text, images, etc. For example, to obtain the best results in the Binary Classification type you would need at least 1000 records. Fortunately, Microsoft is also providing some data samples so we can play with this.

AI Builder comes as a complement to other packaged AI models that we had already available like the Sentiment Analysis or Topic Extraction we have with Azure Cognitive Service, which is already available in Flow and PowerBI. It wraps some of the more sophisticated capabilities of Cognitive Services, like Computer Vision and makes it easily consumable within Power Apps.

This new AI Builder capability of the Power Platform is a fantastic example of the effort that Microsoft is doing to democratise the access to AI technologies for all the business. It allows all types of users to create AI models for their business applications without requiring complex tools or processes, and help to deliver increased value to their business.