AI builder is a new Power Platform tool that allows users with no programming knowledge or data science skills to create powerful AI models that can help run your business. Using its easy “point-and-click” experience, business users can benefit from rich data analysis and predictions. In addition to this, tedious business processes such as analysing customer feedback or digitalisation of paper forms can be autotomised.
There are currently 4 model types available, which we will take you through in more detail in this blog:
All 4 of these model types are extremely powerful and use modern machine learning algorithms to analyse and transform data in intelligent ways. Each time a model is trained, it is given a performance score (out of 100%) to indicate its usefulness. The model can then be published and used in CDS, Canvas apps, Microsoft Flow, Dynamics 365 and more.
For some models, a training report is produced, which can help users identify errors and warnings within their data, such as fields with low relevancy or no data. Other models allow you to quickly test them by providing some sample data, great for knowing if your model will work.
Binary Classification is an AI model that takes a Two Option field (yes/no) and uses historical data within the entity to make predictions. These predictions are stored in new fields on the entity that can be leveraged by CDS. These new fields can then used on 365 views and forms, Canvas Apps, Flow and the Dynamics 365 SDK
1 - The new fields created by the builder
Required Data to Train |
CDS Entity with at least 50 records |
Input |
Selected or all (relevant) fields on the Entity |
Output |
3 new fields on the entity:
|
Can be used in |
Microsoft Flow |
For example, this model could be used to predict the likelihood of a customer being approved for a loan, based on patterns in their information and from within the AI model..
2 - Loan application results displayed in a Canvas app
Form Processing is an extremely powerful model that can extract data from structured documents and store them within CDS, this could be extremely useful for the automation of invoicing or the processing of tax forms
3 - A structured invoice that could be processed
Required Data to Train |
At least 5 documents (.JPG, .PNG, .PDF) |
Input |
A document that is of similar structure to the trained ones |
Output |
Text values of the data found from the input document |
Can be used in |
Microsoft Flow |
For example, using Flow, an invoice could be retrieved from a system (provided there is a connector) and passed to the published model. The model would then process the form and store the data in CDS without the need for manual intervention by employees.
Similar to Forms Processing, Object Detection is used to identify items within an image or scene. Once the model is trained, it can be used to detect these items and provide a count of how many it finds and a percentage of how accurate it thinks the prediction is.
Required Data to Train |
Object names to detect (from CDS) At least 15 images which contain the objects More than 50 recommended |
Input |
Any image |
Output |
Count of object names found Predicted accuracy of the results (0-100) |
Can be used in |
Microsoft Flow PowerApps (Canvas) |
As an example, it could be used to detect products within images and then assign those images to the actual product within CDS. A user could take pictures or upload pictures to a device (using a Canvas app), and have the model automatically assign them. No need for searching and finding the correct product, which reduces time and user error.
Text Classification can be used to identify key information within pieces of text. Each piece of text is tagged, and these tags can be utilised to perform further analysis on the data. It currently supports up to languages (English, German, French, Italian, Portuguese, and Spanish)
Required Data to Train |
CDS entity with text and tags fields |
Input |
Any piece of text (social media, documents emails, etc) |
Output |
The relevant tags for the piece of text |
Can be used in |
Microsoft Flow |
Some good use cases for this would be automating customer survey processing. For example, reading responses from Microsoft Forms Pro, using Flow to match against certain tags that the model has come up with, and then contacting the relevant people within a company for those tags (e.g. a manager or a specialist).
Alternatively, it could be used for sentiment analysis (e.g. from social media) or detection of spam (e.g. from emails or social media)
4 - An example of the tags that may be matched to text
Microsoft is also providing some pre-built models that don’t need to be trained. These tools can be leveraged by Flow and PowerApps to automate processes. Currently, Microsoft are providing a business card scanner “App”, which will scan a business card and store the data in CDS.
The app will allow you to review and edit the data before storing it in your contacts .
They plan on adding more of these in the future.
5 - Canvas App builder dropdown for AI models
AI builder is an awesome and powerful tool that will really help bring high-quality data analysis into the hands of many. The fact that it already has so many uses and great integration into existing Power Platform apps means it can only get better. Business analysts, power users and even developers will love it.
Contact us today for more information on how AI Builder can accelerate your business.