AI Builder for Beginners￼

AI Builder Fundamentals: Introduction

Artificial Intelligence (AI) meets Power Platform within AI Builder. AI Builder was built with the intent to eliminate data science and advanced coding skills for the Power Platform. This tool gives Makers the ability to train models to quickly transcribe and grab data to be readily available within the Power Platform. AI Builder is comprised of five different models to quickly consume, integrate, and manipulate your data within your systems. This is crucial to data development, and enhancing customer service through a reduction in data processing times and deprecating paper processes.

Prebuilt Models vs. Custom Models

Prebuilt models are predefined models that Microsoft has provided for Makers within AI Builder. This enables citizen developers and pro developers to begin using AI to immediately read, define, and transform data. You can begin designing, building, and training your data model immediately and easily through Power Automate or Power Apps within AI Builder. The table below provides prebuilt data models that a citizen developer might be immediately intrigued by.

Model type	Availability
Business card reader	Power Automate and Power Apps
Category classification	Power Automate and Power Apps
Entity extraction	Power Automate
ID reader	Power Automate
Key phrase extraction	Power Automate
Language detection	Power Automate
Receipt processing	Power Automate and Power Apps
Sentiment analysis	Power Automate
Text recognition	Power Automate and Power Apps
Text translation	Power Automate

*From Microsoft*

Custom Models are available for all developers but will allow Makers to further read and transform data and automation processes with advanced modeling skills. We see five key components of custom models with AI Builder:

Category Classifications

Category classification sits at the core of Natural Language Processing (NLP) constraints. With Microsoft’s Category Classification, we see an increase in the ability to read text entries for simple data reading like sentiment analysis, customer requests, spam detection, and overall data classification. This allows for better identification of customer inquiries, potential sales opportunities, and strengthens your Customer Relationship Management (CRM) data.

Text Extraction

Text extraction allows your AI Builder model to recognize specific data and texts within structured, semi-structured, and unstructured documents. Makers can create pre-defined data extraction requirements that allow the model to process and retrieve information, facts, and answer customer questions. Using Optimal Character Recognition (OCR) AI builders can manage semi-structured and un-structured handwritten data and transform them into readily available data for Apps and Flows.

Form Processing

Form processing allows AI Builder to digitally read and write structured documents like invoices or tax documents easily without the need to manually extract information. This is a common AI Builder function to get you started. By training your data in a structured document you’re able to remove manual paper processes and reduce human errors and day-to-day processes of data collection.

Prediction

AI Builder analyzes patterns and data to predict outcomes within AI Builder. This ability allows for end-users to receive questions about your data using binary options like yes/no, and give the appropriate answer in the form of a number. For example, a binary prediction might be – “Is this a valid customer for a Marketing campaign?” AI Builder will run through the trained logic and prediction to giver you a simple binary answer based on set calculations.

Image Classification

Utilizing the third-party application Lobe, AI Builder allows Makers to import and label images and files to automate the data information and intake process.

What's New With AI Builder?

Learn more about new features in AI Builder in Microsoft Power Platform's 2022 Wave 1 Release.

READ THE BLOG POST

Build a Prebuilt AI Model in AI Builder

Step 1: Select and build an AI model

As a beginner to AI Builder, the first thing that you are going to do is choose a prebuilt AI model. We recommend the invoice processing AI Builder model as it is among some of the most common scenarios and relevant in the majority of the customer, client-facing businesses. You have the option to choose a Prebuilt or Custom model. Custom models are generally going to be geared towards pro developers with added coding functionality.

Step 2: Train your AI model

Once your prebuilt model is defined you are going to begin training your AI Builder data. This is done through collections. Within AI Builder you are going to have a series of document collections that you use to train the data AI to grab the most accurate information. This can be monitored through AI Builder’s accuracy score. It gives the Maker an idea of how accurately the data is being interpreted and if you should possibly add new data collection and training to the process for data accuracy.

Step 3: Manage your AI model

When your AI Builder model is created you are now given the ability to manage your AI Builder model. You want to ensure that your data model is being continuously evaluated and that the performance score is performing high. There are two common issues within AI Builder which is underfitting and overfitting models. Underfitting models is when your data is under-performing and it typically tells you that you did not train your data accurately or there is a lack of training data within your model. This could also be an indicator of the types of fields you’re using to collect that data. The other common issue is overfitting your data models. This means that you have over-trained your data for too precise of outcomes. Your historical data is not running cohesively with new relevant data at the same time and your predictions are working against each other.

Step 4: Publish your AI model

Once your model has trained your AI Builder model is now ready to be shared across your environment. When you publish your model you now have the ability to share this across users in your Power Apps or Power Automate environments. A Maker would want to publish this if it’s being broadly used across a company for Application Lifecycle Management (ALM) best practices and to ensure updates are taking properly across production environments.

Step 5: Use Your AI model

Once your model is published and in the proper production environment, you can now enable your end-users to optimize business processes. They can upload models directly into Power Automate or Canvas Apps models to start digitally transforming your business data intake.

Closing

Data processing and integration is getting easier by the day. Microsoft has taken steps to reduce the amount of data science and coding necessary to work within these business applications. As a citizen developer we continue to see the gaps between all forms of developers close with these ready-to-use products. The Power Platform just enhances business process automation on every end. Schedule your call with Avantiico today to learn how we can help close the gap of your outdated processes and transform them into the digital world.