Power Platform Community Forum Thread Details

I'm reaching out to ask for your support in setting up a document classification process. The goal is to automatically route different types of documents into specific SharePoint folders based on their content.
For some customers, this is straightforward: they use distinct PDF filenames that clearly indicate the type of credit note. These filenames can easily be used as conditions in Power Automate.
However, for other customers, the document type is not identifiable from the filename alone.
To address this, I initially implemented AI Builder in Power Automate to automatically detect keywords within the document and classify it accordingly. This worked well at first, but I’m now encountering an issue with the AI Builder action “Text in image or document.” The flow fails with the following error:

InvalidPredictionInput – The base64 binary content is missing for input single
InvalidImage – The file submitted couldn't be parsed (supported formats: JPEG, PNG, BMP, PDF, TIFF)

The input is a PDF file attached to an email. It opens normally and is not password-protected. This process previously worked without issues. I have an active AI Builder trial, and the environment has Dataverse enabled.
To troubleshoot, I added a step to convert the file into PDF again, but this didn’t resolve the issue.
After some research, it seems that PDFs automatically generated by SAP or similar systems may not be readable by AI Builder in Power Automate.
Would it be possible for Nanonets to create a classification model that can help us distinguish between different types of credit notes and route them to the correct folders? Or do you have any other suggestions or alternative approaches?

What I really do not understand is, why suddenly it doesn't work anymore.