What is AI Builder for Microsoft Power Platform?

Microsoft AI Builder Preview

Now available in preview, Microsoft AI Builder is a new Power Platform capability to speed up processes, predict outcomes and improve performance through AI insights.

Immediate thoughts about AI often include concerns about the level of complexity combined with uncertainty about how AI technology can be beneficially used within an organisation.

Through its Power Platform and Common Data Service, Microsoft is helping business connect their data but in many instances intelligence is limited to what analysts are able to pull together in creating forecasts and insights.

Relatively few organisations have data scientists or pro-developers on hand so this new solution aims to place AI solutions within the reach of business users with the same no-code / low-code capabilities that has driven widespread adoption of PowerApps. These can be built by a Microsoft Power Platform Partner or are now easier than ever to create in house.

Build in PowerApps

In the same way that PowerApps is helping organisations to build connected apps and automate tasks, AI Builder makes it possible to do machine learning and Azure-based cognitive services without the need to write code. 

Accessible from the navigation pane within the PowerApps Studio or the Microsoft Flow website, AI Builder integrates with 250+ PowerApps connectors and writes data outputs back to the Common Data Service. This enables AI to be added to flows that will provide predictions for Dynamics 365 and other connected apps across the Power Platform.

AI Solutions Types

For this preview, Microsoft has released four AI models:

Object Detection

This example has featured strongly in Microsoft's presentations demonstrating how AI models can be built to recognise physical objects. Early retail adopters have developed object detection to support inventory management where pictures can be taken to recognise and count stock. In a field service use case, this could speed up repair processes by bringing up a manual or knowledge resources where any product including machinery can be instantly identified through an object detection model via a PowerApp on a mobile device.

To train an object detection model to recognise items sufficient images will need to be uploaded to increase the accuracy of prediction results. At least 50 images per product is recommended but the more the better. These model-training images would need to be representative of what would be submitted by end users and so will be most effective when shown in context of 'real' and varied backgrounds rather than neutral backdrops.

Other factors to consider when uploading images to effectively train an object detection models will be to capture different parts of an object, provide varying sizes, using different lighting and alternative camera angles.

Binary Classification

This AI model will predict yes / no type outcomes by drawing on historical data and associating historic data patterns with previous outcomes. Once learned patterns are detected in new data the model will be able to classify fields with predicted future outcomes.

For instance, this model could be deployed to set predictions that include, 'will this customer churn?', 'is this a key account?' 'will an applicant be eligible for membership?' or 'will more inventory be needed?'.

This requires data to be stored in the Common Data Service and for the best results, Microsoft recommends training this model with at least 1000 records. 

This model will support any binary outcome including true / false and pass / fail. 

Form Processing

Advanced machine learning will extract structured data from PDF, digital paper and forms by selecting the fields that matter in each use scenario.

For example, a form processing AI model could be trained to automatically recognise and extract the key detail from invoice documents that are received as PDF email attachments to include the from address, invoice number and the main table for product items, quantity and rate. 

By automatically recognising and extracting relevant data, and putting it in the Common Data Service, this AI model type will help to reduce the reliance on manual data input to shortcut processing time.

To train this model type, a minimum of 5 form document samples would need to be uploaded enabling machine learning to understand what information needs to be extracted. This will analysis the form structure so that the important form fields can be selected. 

In the preview, text based PDFs are the recommended input while JPG and PNG formats are also supported. 

Once the model is trained and published, the form processor control can be used within the context of a canvas PowerApp or through Microsoft Flow.

Text Classification

With organisations handling ever-increasing volumes of data, it presents a challenge to identify actionable insights, increase accuracy and improve processing efficiency that will lead to better products and services being provided. 

Across large volumes of data stored in the CDS, the text classification model will seek to automatically classify, categorise and group text. For example, this could include training an AI model to recognise sentiment from survey responses and using these categorisations to drive workflows.  

Also, by automatically tagging or labelling text entries using this model, classifications could be used to route workflows, detect spam and trigger notifications in combination with Microsoft Flow and PowerApps. 

These outcomes could also be used as an input for other AI capabilities including predictive analysis by learning from previously labelled text items.

For an effective predictive text classification model, 1000 or more data records is recommended with a focus only on the key fields that will influence the prediction. 

AI Builder Availability

The AI Builder Preview feature is enabled for any eligible environment in the Power Platform Admin Center. This environment must have the Common Data Service and its region must be supported. Currently, the AI Builder is only supported in environments created in the United States or Europe. As a result, this is available for customers in the EMEA Geo (CRM4) but it is not presently available for UK specific geo's (CRM11).

More Microsoft AI Examples

AI capabilities aren't confined to this new builder solution as Microsoft has been steadily infusing Dynamics 365 and the Power Platform with AI features in recent years. This includes relationship insights action cards, resource scheduling optimisation and recommended knowledge posts.

Its recently announced business card reader is another example. This represents a pre-trained AI model that is directly available within PowerApps to extract data from a scanned business card and store this in the CDS. As well as being a component in the canvas app builder, the business card reader will be available via the Dynamics 365 mobile app to analyse cards, extract relevant detail and instantly create new contact records.

First party AI apps are also available. These turnkey solutions have been designed and trained by Microsoft so are available out of the box to provide insights based on pre-defined AI models. These include:

Sales Insights promoting personalised engagement and proactive decision-making through predictive lead scoring, notes analysis, opportunity health analytics and interactive reports with more AI capabilities announced for the next release wave in October 2019 including mobile sales assistant...READ MORE

Customer Service Insights now included with Customer Service Enterprise licences to help service leaders improve customer satisfaction and increase efficiency through AI-driven insights. This includes using natural language understanding to automatically group cases by topics so high impact and growing issues can be understood at an early stage...READ MORE

Customer Insights, currently available in preview, unifies data into comprehensive customer profiles by mapping records into the Common Data Model with AI applied to match and merge data. Contextual customer insights and tailored customer profile cards can then be infused into business applications to help sales, service and marketing teams drive personalised engagements across channels. This is designed to enable business analysts to create customer segments based on a combination of common attributes and interactions associated with unified profiles...READ MORE

Market Insights, also available in preview, serves up daily or weekly emails that leaders and employees with a summary of the most important news related topics of interest in a quick and easy way and discover emerging market trends...READ MORE 

What Next?

With enterprise-grade AI capabilities increasingly becoming generally accessible, small and mid-sized firms are able to apply AI within the context of their business apps and processes to drive better business outcomes. To learn more and explore the possibilities offered by the AI Builder and other AI capabilities featured in this post please get in touch.

RELATED: 


Receive CRM and Microsoft Dynamics 365 Updates

If you enjoyed this post, why not join our mailing list to receive emails about the latest updates and events from Preact?