Create an AI tool easily with AI Builder from Power Platform

Hardly a day goes by in the IT world without hearing about artificial intelligence (AI). In recent months, the buzz around Open AI's Chat GPT has been intense. The use of AI-based tools has thus become accessible to everyone, the use cases more concrete by the day, and their added value undeniable. Even so, the development of customized AI tools seems to remain in the hands of a handful of specialists. Microsoft's low-code Power Platform suite now includes an AI Builder module that lets you create applications or processes integrating AI-based functionalities without a single line of code. The opportunity was too good to pass up.

Before going any further, I'd like to make it clear that this article is not sponsored by Microsoft or any other publisher.

AI is no longer the domain of a few specialists.
AI is no longer the domain of a few specialists. (Image source: Pixabay)

What is AI Builder?

AI Builder is a set of predefined AI models that can be reused directly in Power Apps or Power Automate workflows:

  • Some models can be used as is (language detection and translation, for example);
  • Others are customizable and trainable (I'll come back to this later), such as extracting data from a document.

In practical terms, this means using an AI Builder connector in workflows or applications, which technically functions in the same way as those in SharePoint or Outlook, for example.

This connector, like the others, takes parameters as input and returns other information as output.

The simplest example is the language detection action. It takes a text as parameter and returns the detected language with an associated probability (number between 0 and 1).

AI Builder models are divided into 4 categories:

  • Documents: extract text from a document ;
  • Text: sentiment analysis, classification, translation ;
  • Structured data: result prediction from history ;
  • Images: object detection.
AI transforms incoming data into a different type.
AI transforms incoming data (an image, for example) into another type of data (a detected object and a probability). (Image source: Pixabay)

A few examples

OK, technically it works, that's great. But what would be really nice is to find cases where it brings value. Here are three use cases I've tested.

Inbox e-mail translation

This can be useful for people who don't speak English, but occasionally receive e-mails in English. Here I've created a Power Automate flow which, when a new e-mail arrives in the inbox, checks whether the text is in English. If it is, it will automatically translate it and send it in a second e-mail.

Translation is automatic, so there's no need to "train" the AI in this case.

Completion time: 30 minutes.

Object detection in a captured image

This can be useful for field employees who work with their smartphone and need to identify a particular object in a photo they're about to take. For example, to identify damage to company equipment on the public highway. In this case, I've integrated the connector into a Power App. All we have to do is provide an image (camera or reader) and the component will give us the object(s) identified in the image, with the associated probabilities. In this case, it's necessary to train the AI.

A minimum of 15 images is required to define an object to be identified. This is clearly the most tedious action for this use case. All the more so as we rarely limit ourselves to a single object.

Completion time: 2 hours without special Power App formatting (with 3 separate objects, i.e. 45 training images).

Extracting information from an invoice

Here, we use AI Builder's OCR (Optical Character Recognition) functions to identify different values in the fields of an invoice in PDF or image format, such as the supplier, the amount before tax, the date, the invoice number and so on. In my test, I added the images to a specific directory on a SharePoint site. This addition triggers a Power Automate flow automatically. The flow will use the OCR template I trained beforehand (a minimum of 5 template documents are required), and then feed a SharePoint list with the fields I'm interested in, taken from the invoice.

NB: I'm dealing here with an invoice, but this can be extended to any type of document.

Completion time: 1 hour.

Some AI models require training to be effective
Some AI models require training to be effective. (Image source: Pixabay)


You can test all these use cases (and even your own that better match your business) with an AI Builder evaluation version for 30 days.

After 30 days, you must purchase AI Builder capacity (add-on). In simple terms, you buy a quantity of AI Builder uses, the price of which depends on your consumption. You can estimate this cost using a calculator provided by Microsoft(AI Builder Calculator | Microsoft Power Apps).

To go further

AI Builder is easy enough to learn if you're already familiar with the Power Platform.

In my opinion, this simplicity is its strength, but also its weakness, in the sense that the possibilities are rather limited for slightly more sophisticated use cases. For more complex projects, it's better to turn to Azure, where you'll have a much more technical ecosystem, but also a much richer one. On the other hand, it's hard to imagine implementing these more sophisticated use cases in just a few hours. As always, you need to keep things as simple as possible by default, estimate the possible return on investment, and make the necessary developments and investments.

You can find out more about AI Builder and the possibilities of AI with Microsoft solutions by contacting SmartView.

Damien CELLE

Damien CELLE

Office 365 Consultant


An article by

Damien CELLE

Damien CELLE

Office 365 Consultant

Want to go further?