AI Builder doesn't recognize certain PDFs

[JFlipot]

Hello All, I have searched the forum without finding a proper solution.

Our system creates muti-page PDF invoices.

I cant' either submit sample nor explain how they are created, unfortunately.

My Power Automates Flow detects that I receive an email and triggers different actions.

One of these being to launch my AI Builder to collect datas from the PDF file. Works 95% of the time properly (besides that it collects all the characters in the location, including ":" when I don't want it => Corrected with another script to erase, which is fine)

Sometime the flow doesn't recognize the file at all. I tried also to used the "failed" documents to set another collection, but it immediately gives an error and won't even store the documents. Somehow I have the feeling from searching in the forum this is linked with a "64 byte vs Binary" thing, but this is far beyond my knowledge.

I'm posting a print screens of entry and exit where the step fails. Any idea how to solve this with the existing document?

Bonus question: Can I set a step that says "if the AI builder fails, do this no matter what"?

thanks for the support !

[JoeF-MSFT]

Hi @JFlipot - thanks for the question.

For the files that are not recognized. Are you able to open those files from your email client? Are those files that are failing to be processed protected in some way (password protected, access restricted...)?

Another test we can try - before the AI Builder action, if you add an action to store the file to OneDrive for business. Do the files that are failing to be processed by AI Builder are created correctly in OneDrive?

On the bonus question 🙂 you can use the "Run After" settings in Power Automate: Using "Run After" as a Condition in Power Automate / Flow (flyingpolymath.com)

[JFlipot]

Hello, files do not seem to have any difference. They are not protected. 

Yes I can open them from mail and the flow can also create the pdf in OneDrive with no problem. 
This is actually a part of my flow just before adding a line in excel with AI extracted datas. 

thanks for the run after idea, will try today. 

[JoeF-MSFT]

Hi @JFlipot - here's another couple of things we can try in our journey to discover what might be happening. 🧐 

• If you take the documents that fail to process on the flow, and go back to your model details page and try to do a Quick test with those documents. Do you also get an error?

• If you send an email with all the documents that have failed to be processed to retrigger the flow, do all documents keep failing in the new flow run? 

[JFlipot]

I tried again a quick test - it fails.

leads me to a standard page on "reqs and limitation for a document processing model"

My pdfs are not password protected and don't seem to have any difference between the ones that work and the other ones.

I've tried to send the failing ones in one mail. the copying on the onedrive works, only the ai builder fails.

But guess what, I've just had the idea to fake a modification through iLovePDF on the web and the generated file didn't fail... just added a blank circle on a blank zone...

not satisfactory as I'll have to do manual processing, but for now this will be better than nothing...!

Still interesting in understanding the rootcause / finding a fix .

it's good night for me ! 😴

[JoeF-MSFT]

Thanks for all these updates @JFlipot. Nice tool name iLovePDF 🙂

Let's try one last thing (thanks for your patience while we go through these), if you go to the AI Builder homepage (https://aka.ms/tryaibuilder) You select Text recognition Extract all the text in photos and PDF documents (OCR) and upload one of the original PDFs that failed with the document processing model, are you also getting an error?

[JFlipot]

Hello,

Just tried and the same mistake message on format shows up.

I'll have the IT checking the PDF creation mechanism...