Views:

Applies to Product- Power Automate

What’s happening?
Customers are experiencing difficulties with AI Builder models for data extraction, specifically when attempting to extract information from multiple pages of PDF documents or when the model provides incorrect data.


Reason:
The underlying cause for this includes: - The AI Builder model may not be properly configured to handle data extraction from multiple pages. - The model may not have been trained adequately with sufficient examples or the correct tagging for tables that span across multiple pages.


Resolution:

  • For further guidance, refer to the documentation when tagging documents with multipage tables
  • For further guidance, refer to the documentation best practices when building a custom document processing model
    • For template documents
      • One (1) layout per collection.
      • The layout must be identical within the same collection.
      • A minimum of five (5) and a maximum of 20 documents per collection (don't exceed this limit).
    • For general documents
      • Create a single model for all variations of a document type including both structured and unstructured documents
      • Label fields relevant to their values for better extraction accuracy.
      • For example, use "supplier_id" for a supplier ID. Field names should match the document's language.
      • A minimum of five (5) and a maximum of 20 documents per collection (don't exceed this limit).