How can we automatically convert unstructured SharePoint documents into structured, reusable knowledge records?

Question

We are exploring a Microsoft 365-based approach to build a structured knowledge repository from a large volume of documents stored in SharePoint.

The broad idea is:

• Users upload documents (PDFs, reports, presentations, web content, etc.) into a SharePoint document library.
• The system automatically extracts relevant information from those documents.
• Extracted information is classified against a predefined taxonomy.
• Structured records are stored in Microsoft Lists for search, reuse, reporting, and downstream content generation.
• Human reviewers validate the generated records before publication.

A few questions for those who have implemented similar solutions:

1. Is it better to maintain a single standardized List structure for all extracted records, or separate structures for different content types?
2. Has anyone successfully used SharePoint Agents, Copilot, AI Builder, or Power Automate to extract information from uploaded documents and populate SharePoint Lists?
3. What is the recommended architecture for applying controlled taxonomy and metadata during extraction?
4. Should extracted records first be written to a review/approval queue before being published into the main repository?
5. What limitations, scalability considerations, or governance challenges should we be aware of when implementing this pattern using native Microsoft 365 capabilities?

Any lessons learned, reference architectures, or implementation experiences would be greatly appreciated.

Answer 1

Hi Paul N,

Thank you for sharing your proposed approach. This is a thoughtful architecture and aligns well with several Microsoft 365 capabilities for document processing, metadata enrichment, and knowledge management.

A possible approach is to use SharePoint document libraries as the authoritative content source, leverage AI-powered extraction and classification capabilities, and then store validated metadata in Microsoft Lists or Dataverse for reporting and downstream applications. AI Builder document processing models can extract information from uploaded documents and populate metadata fields automatically, while Power Automate can orchestrate review and approval workflows.

Regarding your specific questions:

• List structure: A hybrid approach is often the most manageable. Organizations typically maintain a common set of metadata fields across all record types while allowing additional fields for content-specific scenarios.
• AI extraction: AI Builder, SharePoint document processing, and Power Automate have been successfully used to classify documents, extract metadata, and populate SharePoint columns and Lists automatically.
• Taxonomy nanagement: Using a centralized SharePoint Term Store with managed metadata is generally recommended. AI can suggest classifications, but controlled taxonomy helps maintain consistency and improves search quality.
• Review and approval: Introducing a review stage before publication is a best practice. A validation queue allows human reviewers to verify extracted data and taxonomy assignments before records become available to a broader audience.
• Governance and scalability: Key considerations include taxonomy governance, metadata quality, library growth, AI confidence thresholds, lifecycle management, and licensing requirements for AI-driven capabilities.

You can consider an architecture looks like this:

SharePoint Library > AI Extraction > Taxonomy Mapping > Review Queue > Approved Repository > Search, Reporting, and Copilot Consumption

For more information, please consult:

AI Builder in SharePoint with document processing overview - AI Builder | Microsoft Learn

Overview of taxonomy tagging | Microsoft Learn

However, for a live session guidance, the most effective next step is using your admin credentials to submit a support request through the Microsoft 365 Admin Center.

Microsoft also provides customer service phone numbers based on your region. You can find the appropriate contact number here: Customer service phone numbers - Microsoft Support 

This approach allows a Microsoft Support engineer to conduct a remote session, investigate backend configurations, run advanced diagnostics, and recommend or implement configuration changes based on your organization's specific requirements. These backend-level tools and resources are essential for resolving issues that cannot be diagnosed or addressed through the user interface alone.

As community moderators, we’re here to guide you, but due to privacy and security limitations, we don’t have access to the backend tools required for a full resolution. For this reason, contacting Microsoft Support via the Admin Center is the most secure and efficient way forward. 

I hope this helps you resolve the issue quickly. I’m glad to assist and truly hope the information provided has been useful. Please feel free to reach out anytime if you need further assistance.  

If you find my post helpful, kindly consider marking it as the accepted answer. Doing so can assist others in the community who may have similar questions in finding solutions more quickly.  

Thank you for your kindness and contributions to the forum. 

---

Note: Please follow the steps in our documentation to enable e-mail notifications if you want to receive the related email notification for this thread.