Azure Document Intelligence and Content Understanding

Hi omer,

Thank you for reaching out on the Microsoft Q&A.

Azure Document Intelligence provides advanced capabilities for extracting and understanding content from documents using AI models. If you’re working with Content Understanding, here are key points and steps to ensure success:

It’s part of Azure AI Document Intelligence, designed to analyze structured and unstructured documents, extract text, tables, and semantic meaning, and enable downstream workflows like classification or summarization.

Common Scenarios

• Processing invoices, receipts, contracts, or forms.
• Extracting entities and relationships for automation.
• Applying custom models for domain-specific documents.

How to Use It

1. Choose the Right Model
   • Prebuilt models for invoices, receipts, ID documents.
   • Custom models for specialized layouts or language.
2. Set Up the Endpoint
   • Use the documentIntelligence endpoint in your Azure resource.
   • Ensure correct region and authentication (API key or Microsoft Entra ID).
3. Send the Request
   • Use REST API or SDK (Python, C#, Java).
   • Include the document file or URL in the request body.
   • Specify the model ID (e.g., prebuilt-invoice or your custom model).
4. Handle the Response
   • Parse JSON output for extracted fields, tables, and confidence scores.
   • Apply post-processing logic for validation.

Troubleshooting Tips

• If results are incomplete or confidence is low:
  • Check document quality (resolution, clarity).
  • Validate model training data for custom models.
  • Confirm you’re using the latest SDK version.
• For errors like “no content detected”:
  • Ensure file format is supported (PDF, JPEG, PNG).
  • Verify the request payload and headers.

References

• https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/overview
• Troubleshooting Document Extraction Issues with Azure Document Intelligence
• Understanding Document Size Limits in Azure Document Intelligence
• Choosing the Right Foundry Tool for Document Processing
• Azure AI Services Overview

Please let me know if there are any remaining questions or additional details, I can help with, I’ll be glad to provide further clarification or guidance.

Thankyou!

Thank you, Anshika.

Do we need to build a separate custom model for each template type?

When a business user uploads a new document (PDF or Excel), how do we determine which custom model should be applied? Should we attempt to run multiple models and select the one that yields the highest extraction confidence or the most complete set of fields?

From a technology perspective, which approach would you recommend? Is Azure Document Intelligence sufficient on its own, or should we consider using both Document Intelligence and Azure Content Understanding together?

Hi Ömer Faruk Özsakarya,

Yes, if your templates differ significantly in layout or field structure, creating a separate custom model for each template type. Azure Document Intelligence models are optimized for consistent layouts mixing very different templates in one model can reduce accuracy.

Determining Which Model to Apply for New Documents: Common strategies include:

• Classification First: Use a lightweight classifier (e.g., based on document metadata or a simple layout analysis) to route the document to the right model.
• Confidence-Based Selection: If classification isn’t feasible, you can run multiple models and pick the one with the highest confidence or most complete extraction. This is more resource-intensive but works when templates are similar.

Technology Recommendation

• Azure Document Intelligence Alone: For most structured extraction scenarios, Document Intelligence is sufficient. It provides prebuilt models and custom training for PDFs, images, and Office files.
• Combining with Azure Content Understanding: If you need semantic enrichment (e.g., categorization, summarization, or contextual insights beyond field extraction), pairing both services can add value. Document Intelligence handles structured data extraction, while Content Understanding adds meaning and context.

For more details, check the official docs:

https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/overview?view=doc-intel-4.0.0

https://learn.microsoft.com/en-us/azure/ai-services/content-understanding/

I Hope this helps. Do let me know if you have any further queries.

Thankyou!

Thank you very much for the detailed expression. I do not understand the role of Content Understanding. For instance I have the below sample templates

• template 1: column names are written. some cells contain multiple values.
• some column names are not written. For instance IS 660, Panel are name of the laboratory tests but it is not written in excel sheet.

I can create two custom models in Azure Document Intelligence to extract product, test, result, and batch values from Excel sheets and load them into a normalized table in Databricks.

However, I would like to clarify the following points:

1. Multi-value cell handling Can a Document Intelligence custom model interpret multiple values within a single cell and output them as multiple records? For example, if a cell contains MgCl2·6H2O, can the model correctly decompose this and produce three logical records? (first screenshot)
2. Domain-specific semantic understanding Can a Document Intelligence custom model recognize that values such as IS = 660, AOCS Td 1a-64, ISO 6685, ISO 3960, Panel, and GC-MS represent laboratory test names, rather than treating them as free-text strings? (second screenshot)
3. Role of Azure Content Understanding How does Azure Content Understanding add value in this scenario? Specifically, how does it complement Document Intelligence?

Thanks

Thanks for sharing the detailed context and examples! Let me address each point:

1. Multi-value cell handling

Azure Document Intelligence custom models are designed to extract structured fields, but they do not automatically split multiple values within a single cell into separate records. If a cell contains something like MgCl2·6H2O, the model will treat it as one string. To achieve decomposition into multiple logical records, you would need post-processing logic in your pipeline (e.g., using Python or Databricks transformations) after extraction.

2. Domain-specific semantic understanding

Custom models in Document Intelligence primarily learn patterns from labeled training data. They do not inherently “understand” domain semantics like recognizing IS 660 or GC-MS as lab tests unless you include these examples in your labeled dataset. By providing sufficient annotated samples, the model can learn to classify these as test names rather than generic text. For deeper semantic interpretation, you might consider integrating Azure AI Language or custom NLP models alongside Document Intelligence.

3. Role of Azure Content Understanding

Content Understanding complements Document Intelligence by adding higher-level semantic enrichment and classification. While Document Intelligence focuses on extracting structured data from documents, Content Understanding can:

• Classify documents by type or purpose.
• Apply domain-specific taxonomies.
• Enable search and retrieval based on meaning rather than raw text.

In your scenario, Content Understanding could help identify that a document relates to “Lab Test Reports” and tag entities like “Test Name” or “Panel,” improving downstream analytics and search.

In Short:

• Document Intelligence extracts fields but won’t split multi-value cells automatically use post-processing for that.
• Domain-specific recognition depends on your training data; include examples for better accuracy.
• Content Understanding adds semantic classification and context, making your solution more intelligent when combined with extraction.

Here are the official Microsoft documentation links:

Custom Model Training in Azure Document Intelligence

Azure Content Understanding Overview

Please let me know if the issue persists after these checks. If you have any remaining questions or need additional details, I’ll be glad to provide further clarification or guidance. If the above steps resolve your issue, kindly confirm.

Thankyou!