Azure AI Document Intelligence and Power Platform AI Builder: When To Use What

Introduction   

Microsoft offers two powerful tools for automating document processing and extracting valuable information from unstructured data: Azure AI Document Intelligence and AI Builder which is a Power Platform capability. Both tools use advanced machine learning algorithms to analyze documents, identify relevant content, and extract it for further processing. However, they cater to different user types and offer varying levels of customization and integration.

In this article, we’ll:

• Compare the pricing and features of Azure AI Document Intelligence and AI Builder,
• Consider their backend models, ease of use, and scale requirements.

What are Azure AI Document Intelligence and AI Builder

Azure AI Document Intelligence is a “pro-code”, cloud-based service that leverages flexible and customizable artificial intelligence tools to analyze and process documents. A collection of pre-built models are available for use along with custom models that can be trained to solve multiple use cases.

AI  Builder is a “no-code” AI solution integrated with Microsoft Power Apps and Power Automate. AI Builder document processing is built on top of Azure Form Recognizer. It enables users to build custom AI models to solve specific business problems, such as document processing, without writing any code. AI Builder offers pre-built models for common tasks, such as form processing, object detection, and text classification. It is designed for “citizen developers” and projects that require minimal customization.

Feature Comparison

Data Storage (Input):

• Document Intelligence: Connect to Azure storage accounts, URL’s, and various other sources.
• AI Builder: Train models by uploading directly to an AI Builder instance. Can connect to Azure storage accounts, Dataverse, SharePoint and various other sources for input documents.

Data Storage (Output):

• Document Intelligence: Connect, write to pretty much anywhere you want it to. JSON output.
• AI Builder: Default is Dataverse. Any other location requires other tools to pass data (i.e. Power Automate).

File Size:

• Document Intelligence: Up to 500 MB (S0 Tier) or 4 MB (F0 Tier).
• AI Builder: Up to 20 MB.

File Dimensions:

• Both tools have the same requirements:
  • Between 50 x 50 pixels and 10,000 pixels x 10,000 pixels.

File Formats:

• Document Intelligence: PDF, JPG, PNG, TIFF.
• AI Builder: PDF, JPG, PNG, TIFF (Extraction only. Not suitable for training).

Developer Experience

• Document Intelligence: Browser based low-code, point & click, SDK pro-code.
• AI Builder: No-code, browser-based point & click.

Infrastructure Requirements

• Document Intelligence: Azure resource will need complimentary infrastructure around it.
• AI Builder: No infrastructure required.

Custom Model Training Requirements

• Document Intelligence: Requires a minimum of 5 documents to train.
• AI Builder: Requires a minimum of 5 documents to train.

Pre-Trained Model Capabilities

• Both tools come with a standard set of models including:
  • General Document (Key-Value pairs, document structure)
  • Read OCR (handwritten text, detect languages, locations)
  • Invoices
  • Receipts
  • Identity Documents (ID’s)
  • Business Cards
• Document Intelligence extends to:
  • Health Insurance Card
  • US Tax Document
  • Contracts

Accuracy Metrics

• Both tools provide model accuracy and confidence score per field extracted.

Pricing

Pricing for Document Intelligence is based on a per page basis. Pricing for AI Builder is based on ‘capacity’ basis in the form of ‘Service Credits’. Service Credits can be used for a variety of AI builder services within Power Platform. One add-on unit provides 1 million service credits. Pre-built and custom models cost the same in AI Builder. Think of service credits as a bucket of coins that you can use wherever you want, once you run out, you buy more.

Document Intelligence offers a Free Tier (F0) where 0 – 500 pages per month is free.

For example, we’re going to look at 3 scenarios and a monthly cost. These scenarios represent using the general document models.

1. Small Scale: 1,000 pages per month
2. Medium Scale: 10,000 pages per month
3. Large Scale: 20,000+ pages per month

AI Builder add-on volume/capacity pack pricing is $500 per 10,000 pages when purchasing 1-9 units (Tier 1), $375 when purchasing 10-49 units (Tier 2), $250 when purchasing more than 50 unites (Tier 3).

Small Scale:

• Document Intelligence:
  • S0 Tier (Pre-built models): $10 per 1,000 pages
  • S0 Tier (Custom models): $50 per 1,000 pages
• AI Builder:
  • Min 1 Unit of AI Builder: $500 per 10,000 pages

Medium Scale:

• Document Intelligence:
  • S0 Tier (Pre-built models): $100 per 10,000 pages
  • S0 Tier (Custom models): $500 per 10,000 pages
• AI Builder:
  • 1 Unit of AI Builder: $500 per 10,000 pages

Large Scale:

• Document Intelligence:
  • S0 Tier (Pre-built models): $200 per 20,000 pages
  • S0 Tier (Custom models): $1,000 per 20,000 pages
  • Commitment Tiers:
    • Pre-built (Azure source): $190 per 20,000 pages
    • Custom (Azure source): $800 per 20,000 pages
• AI Builder:
  • 2 Units of AI Builder: $1,000 per 20,000 pages

TLDR Pricing Comparison:

In most cases, Document Intelligence is going to be cheaper through the pre-built models or commitment tiers. There might be edge cases where the bucket of service credits is about to be empty, and document intelligence is more expensive, but these would not be the norm and would require a very known number of documents per month to be processed.

Summary

Both tools provide an excellent way to extract text from documents using both pre-trained and custom models. AI Builder is heavily leveraged towards ‘citizen’ developers who prefer a truly no/low-code, lightweight experience that offers little customization, but high ease of use. Document Intelligence is geared towards pro-code, highly customizable workloads where scaling, control, and integration is key.