Choosing the Right AI Builder Model

1. Introduction to AI Builder

AI Builder is a Microsoft Power Platform capability that allows you to add intelligence to your business applications and flows with a low-code interface. It democratizes AI by providing a set of pre-trained models for common tasks and allowing you to build, train, and publish your own custom models without needing deep data science expertise.

You can think of AI Builder models in two main categories:

• Prebuilt Models: These are ready-to-use models for common business scenarios like scanning business cards, processing receipts, or analyzing text sentiment. They are analogous to using a well-defined library or API in C#—you provide the input, and you get a structured output without worrying about the internal implementation.
• Custom Models: These are models that you train with your own data. This is necessary when your business problem is unique, such as processing a custom-designed form, identifying company-specific products in images, or predicting outcomes based on your historical business data. This is conceptually similar to training your own machine learning model, but with the heavy lifting of framework selection and infrastructure management handled by the platform.

Navigating the available models is your first task. Let's see what the landscape looks like.

2. A Tour of AI Builder Model Types

The best way to start selecting models is to understand what's available. The official Microsoft documentation provides an excellent overview.

As you've just read, the models are organized logically based on the type of data they operate on. Let's briefly summarize the key capabilities in each category:

• Documents: This category is focused on extracting information from files like PDFs and images.

  • Document processing (Custom): The workhorse for extracting data from structured or semi-structured documents like invoices, purchase orders, or proprietary forms. You train it by showing it examples of your documents.
  • Prebuilt Models: For common scenarios like Invoice processing, Receipt processing, Business card reader, and ID reader. Always check if a prebuilt model meets your needs before building a custom one.

• Text: These models understand and process natural language.

  • Category classification (Custom/Prebuilt): Assigns predefined tags or categories to text. Perfect for routing support tickets or classifying customer feedback.
  • Entity extraction (Custom/Prebuilt): Pulls out specific pieces of information from a block of text, like names, dates, amounts, or custom-defined entities (e.g., policy numbers, product codes).
  • Other Prebuilt Models: Sentiment analysis, Key phrase extraction, Language detection, and Text translation provide powerful, ready-to-use text analytics.
  • Text generation (Prebuilt): This uses GPT to perform tasks like summarizing text, drafting replies, or extracting information in a conversational way.

• Structured Data: This is for classic machine learning on tabular data, like you would find in a Dataverse table or SQL database.

  • Prediction (Custom): Predicts a binary (yes/no) outcome based on historical data. For example, "Will this customer churn?" or "Is this transaction likely fraudulent?". This is a powerful tool for proactive business decisions.

• Images: These models analyze the content of images.

  • Object detection (Custom): Identify, locate, and count specific items within an image. You could use this to count products on a shelf or identify damaged parts in a photo.
  • Text recognition (Prebuilt): A straightforward Optical Character Recognition (OCR) model to extract printed or handwritten text from images.

3. The Decision Framework: Choosing the Right AI Tool

Knowing the models is one thing; choosing the right one for a specific business need is another. Furthermore, you need to know when AI Builder is the right tool versus when to use a different technology, like a generative AI model (GPT) or a full-scale Azure Machine Learning solution.

Let's start with a resource that excellently contrasts AI Builder with Generative AI. Given your background, understanding this strategic distinction is key.

The key takeaway is that AI Builder excels at producing reliable, structured, and auditable outputs. It's designed for predictable automation. In contrast, GPT-based models excel at understanding and generating unstructured language, but their outputs are probabilistic and less suited for driving critical, deterministic business logic.

With that distinction in mind, here is a simple framework for selecting the right AI Builder model:

1. Define the Business Problem: Clearly state what you want to achieve. Use the business scenarios from the Microsoft Learn article as a guide.
2. Identify the Input Data: What data will the model process? Is it a PDF invoice, a customer email, a row in a Dataverse table, or a photo from a phone? This immediately points you to one of the four main categories (Documents, Text, Structured Data, Images).
3. Check for a Prebuilt Solution: For your data type and problem, is there a prebuilt model available? (e.g., processing a standard receipt). If a prebuilt model fits, it is almost always the fastest and most cost-effective option.
4. Evaluate the Need for a Custom Model: If no prebuilt model exists, or if your data is unique to your business (e.g., custom inspection forms, company-specific terminology), you will need a custom model.
5. Confirm AI Builder is the Right Platform: For most in-platform automation, AI Builder is perfect. However, if you need to train a complex neural network with terabytes of data or require an extremely high-throughput, low-latency API for an external application, you might consider a dedicated Azure Cognitive Service or Azure Machine Learning model. For your PL-400 goals, we'll focus on scenarios where AI Builder is the intended solution.

Applying the Framework: Examples

Let's walk through a few scenarios:

• Scenario A: Automating Expense Reports

  • Problem: Employees submit photos of receipts, and an admin manually enters the vendor, date, and total into the finance system. The goal is to automate this data entry.
  • Data: Images of receipts. This points to the Documents category.
  • Prebuilt? Yes, there is a prebuilt Receipt processing model designed for exactly this.
  • Conclusion: Use the prebuilt Receipt processing model in a Power Automate flow.

• Scenario B: Predicting Customer Churn

  • Problem: A subscription-based service wants to proactively identify customers who are at high risk of cancelling their subscription in the next quarter.
  • Data: Historical customer data in Dataverse, including subscription start date, usage logs, support ticket history, and payment status. This is Structured Data.
  • Prebuilt? No, there's no prebuilt model for predicting churn for your specific business.
  • Custom? Yes. The problem is a yes/no question ("Will they churn?") based on historical data. This is a perfect fit for a custom Prediction model.

• Scenario C: Triaging Incoming Support Requests
  • Problem: Support requests arriving via email need to be automatically routed to the correct support queue (e.g., "Technical Issue," "Billing Inquiry," "Feature Request").
  • Data: The body of an email. This is Text.
  • Prebuilt? There is a prebuilt category classification model, but the categories ("Technical Issue," "Billing Inquiry") are specific to your business.
  • Custom? Yes. You will need to train a custom Category classification model using examples of emails you have already categorized.

Conclusion

In this lesson, you've taken the first and most important step in using AI Builder: learning how to map a business problem to the correct AI model. This strategic skill is crucial for designing effective and efficient solutions on the Power Platform.

Key Takeaways:

• AI Builder offers prebuilt models for common scenarios and custom models for business-specific needs.
• Models are categorized by the data they process: Documents, Text, Structured Data, and Images.
• The selection process involves defining the problem, identifying the data, and then choosing between a prebuilt or custom model.
• AI Builder is ideal for structured, reliable, and auditable AI to drive business process automation, whereas generative AI (GPT) is better for tasks involving natural language generation and summarization.

What's Next?

Now that you know how to select a model, our next lesson will move into practical application. You will learn how to create and train a custom AI Builder form processing model to extract data from documents. We will take one of the most common business problems—automating data entry from custom forms—and build a solution from start to finish.