Azure AI Document Intelligence - Multi-page table rows

Anis Elosta 20 Reputation points
2026-07-07T12:58:17.5766667+00:00

Hello All

I have a custom model extracting an operations table where each row represents an event with From/To times and details. The output is mapped to a SharePoint list.

When an event description continues onto the next page, the model extracts it as two separate rows because table rows cannot span pages during labeling/training.

How are others handling this scenario? Is there a recommended approach to automatically merge cross-page rows before writing to SharePoint?

**Example attached. The highlighted row at the bottom is not a new event. It is the continuation of the 10:00–12:00 event highlighted above. Because the event description spans a page break, the model extracts the continuation as a separate record, which then creates an additional SharePoint list item. Ideally, both fragments should be combined into a single 10:00–12:00 event record.

User's image

Azure Document Intelligence in Foundry Tools
0 comments

Answer accepted by question author

Jose Benjamin Solis Nolasco 9,451 Reputation points Volunteer Moderator
2026-07-07T13:48:59.0766667+00:00

Welcome to Microsoft Q&A

Hello @Anis Elosta , I hope you are doing well,

This is a known limitation with Azure AI Document Intelligence. Currently, table rows cannot span page boundaries during labeling or extraction, so when a logical row continues onto the next page, it is typically returned as two separate rows.

The recommended approach is to merge the rows during post-processing before writing the data to SharePoint. For example, you can:

  • Detect rows where the continuation on the next page has empty or missing key fields (such as From and To times) while only the description column contains data.
  • Merge that description with the previous row instead of creating a new record.
  • Use page number, row order, and the presence (or absence) of identifying fields as heuristics to determine whether a row is a continuation.

In your example, the second fragment has no new From/To values and only continues the event description. This makes it a good candidate to append to the previous 10:00–12:00 event before creating the SharePoint list item. If your documents follow a consistent format, this type of post-processing is generally more reliable than attempting to solve it during model training.

  • User's image

https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/prebuilt/layout?country=us&culture=en-us&view=doc-intel-3.1.0&tabs=rest%2Csample-code

If this answer helped you, please consider marking it as the accepted answer. This helps others in the community find the solution more quickly.

Was this answer helpful?

1 person found this answer helpful.

0 additional answers

Sort by: Most helpful
Most helpful Newest Oldest

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.