From Manual to Automated: Tackling Document AI Workflows with SharePoint, Power Automate, and Claude

This weekend I helped a client "upgrade" a SharePoint document library using Autofill columns to leverage a new AI model and align their automated document processing to manual pipelines they had already established using Claude 4. This involved Power Automate and the Claude API as there's no direct way to just connect Autofill to an alternative model.

The fact is that user-facing AI tools like Copilot, ChatGPT, or Claude, make it super easy to process a whole bunch of different types of files alongside your prompts; but as soon as you jump from the user space to the maker space things get hard really quickly.

It feels like it should be equally simple to turn to Power Automate or Copilot Studio to make this happen, but the reality is that while plugging in AI-based text processing capabilities is now as simple as can be, turning a variety of file types into formats that AI capabilities can consume is much harder. Every AI model gets indigestion very fast when you try to feed them DOCX files.

Here, PDFs and PDF conversion capabilities are your friend, and certainly within the context of Microsoft's inbuilt AI Builder prompt action capabilities provide a foundation for ensuring your AI need can see your content.

There are other options like Azure Document Intelligence or even micro-services you can set up as standalone Azure apps that can do this file ingestion and text or image output for you. However, a quick alternative that can get you to POC fast is just to use a Prompt with a cheap model like GPT-4.1-mini to do the content conversion from PDF to text for you. You're then free to use this output with any model you like.

❓ What approaches have you taken to move manual AI processes using documents into automated workflows? Where have you run into barriers?

First posted on Linkedin on 09/11/2025 -> View Linkedin post here