{"id":8212,"date":"2020-02-11T10:53:19","date_gmt":"2020-02-11T18:53:19","guid":{"rendered":""},"modified":"2025-06-11T08:00:18","modified_gmt":"2025-06-11T15:00:18","slug":"ai-builder-prediction-now-supports-related-entities-and-filtering","status":"publish","type":"post","link":"https:\/\/cm-edgetun.pages.dev\/en-us\/power-platform\/blog\/power-apps\/ai-builder-prediction-now-supports-related-entities-and-filtering\/","title":{"rendered":"AI Builder prediction now supports related entities and filtering"},"content":{"rendered":"

We\u2019ve updated AI Builder prediction to include support for related entities and filtering. These capabilities give users even more flexibility in model creation. In this post, we demonstrate how these features work by building an end to end scenario for how AI Builder can help to optimize delivery performance for an e-commerce company.<\/p>\n

Related entities<\/h2>\n

We want AI Builder to work better for the data that you already have in Common Data Service. We know that all of the data that makes your business run probably isn\u2019t in a single entity, and that there are many entities and relationships all of which may help to improve the accuracy of your prediction model.<\/p>\n

We\u2019ve added support for adding related entities to be used in training, similar to the way you can add related entities in a view. Currently we only support many-to-one relationships, but we are looking to add more relationship types to support later on.<\/p>\n

Filtering<\/h2>\n

What happens when you know that there are irrelevant data within the same entity you are using to train a model? For example, if you don\u2019t want to train on outdated records, or if you want to build region-specific AI models<\/p>\n

We\u2019ve added support for filtering, which is applied to both training and prediction. As an example, if I apply a filter to only look at the U.S. region, then the model will train itself on records where the outcome is known only for U.S. region. Once this model is trained, it will only make a prediction for records where the outcome is not known only for U.S. region.<\/p>\n

More information on how to use these features can be found in our documentation<\/a>.<\/p>\n

Example<\/h2>\n

Let\u2019s say I\u2019m a maker in a typical e-commerce company in Brazil. My CDS database contains typical entities like Order, Product, Customer, etc. I\u2019m noticing that there have been a lot of complaints that deliveries have been delayed specifically for Sao Paulo.<\/p>\n

Without fundamentally changing the system, since this is an isolated issue, I want to see if I can train a model to predict whether a delivery to Sao Paulo is likely to be delayed, and automatically upgrade the shipping for the customer from regular to express.<\/p>\n

Step 1: create output field<\/h3>\n

First, I want to create an output field to denote what constitutes a delayed delivery. In my Order entity, I have the estimated delivery date and the actual delivery date. Every record where actual delivery date is a week or more later than estimated delivery date is considered a late delivery. I will create two fields:<\/p>\n

Delivery Delta<\/strong> \u2013 this is the difference in days between Delivered Date<\/strong> and Estimated Delivery Date<\/strong>.<\/p>\n

\"\"
\nDelivery Timeliness<\/strong> \u2013 this field categorizes the Delivery Delta<\/strong> column into Early<\/strong>, On time<\/strong>, and Late<\/strong>.<\/p>\n

\"\"<\/p>\n

If an order has not been delivered, then the Delivery Timeliness<\/strong> value will be null, and those are the records which the model will make a prediction on, once the model is trained and published.<\/p>\n

Step 2: create prediction model<\/h3>\n

Next, I create a new prediction model from AI Builder section of Power Apps.<\/p>\n

\"\"<\/p>\n

I use my BC Order<\/strong> entity and the Delivery Timeliness<\/strong> field, which we created earlier. In addition, I want to predict\u00a0 whether the delivery will be late, so here, I\u00a0 tell the model that Late is my Yes<\/strong> outcome, and everything else is No<\/strong>.<\/p>\n

\"\"<\/p>\n

Next, I select my related entities, Customer<\/strong> and Product<\/strong>.<\/p>\n

\"\"<\/p>\n

Also, make sure that Delivery Date and Delivery Delta are not selected, because the model should not have information that is only known after the product has been delivered. If you do keep it selected, then the model will show that it has 100% accuracy, but it will not be able to properly predict new orders. This is known in machine learning as target leak, and you can learn more about it here<\/a>.<\/p>\n

\"\"<\/p>\n

Next, I add a filter, so the model only looks at Sao Paulo orders. The filtering experience is the same as the PowerApps view filtering experience that you may be familiar with. We want to create a filter condition on the Customer<\/strong> entity, since the state information is in that entity and not orders. Set the filter condition as shown here:<\/p>\n

\"\"<\/p>\n

Once all that is done, it is time to train the model. For this model, we have an accuracy score of 91%.<\/p>\n

\"\"<\/p>\n

Step 3: publish and use in app<\/h3>\n

Publish your model. As part of publishing, we actually create three outputs:<\/p>\n