What is the problem?
Hello,
As a beginner, I’d like to create the simplest possible workflow using as few modules as possible.
The goal is to retrieve a PDF of an invoice from an email, extract all its content, and then extract the various elements (product name, price, etc.).
The best solution for me is OCR to extract everything correctly, but I can’t find an OCR tool that works and returns the text correctly.
IlovePDF gives me data but it’s encoded, 0codekit doesn’t work at all, and since I’m just getting started, I don’t know how to do all this correctly and cleanly.
If anyone knows how, I’d be happy to receive advice and suggestions.
Thanks in advance !
Screenshots: scenario setup, module configuration, errors
HI there,
You have a couple of options. When selecting a module try navigating to AI by Make or search for extract text from image.
Make has a couple of modules that can extract from documents or images. Also see pre-built templates here, there is an invoice extraction template that may give you some ideas. (https://www.make.com/en/ai-agents-library)
Hi @HurtemWorkshop,
Welcome to the community! 
You can skip the OCR step here, and just pass the PDF directly to the Gemini module (Generate a response) in Make (or other AI modules available in Make like Open AI etc).
It can read the PDF and extract the fields you need (product name, price, etc.) all in one step.
Hope this helps!
Best Regards,
@Mohamed_Jahar
Yes, that would however require that you have an API key and credit available on those 3rd party platforms, as opposed to make credits
Hello,
The simplest approach here is to actually use Claude or OpenAI’s vision/document API directly. They would be able to handle PDFs natively returning clean and structured text in one step.
Here’s the flow:
-A Gmail/Outlook module which watches for emails with attachments
-Download the PDF attachment
-Send it to Claude or GPT-4o using an HTTP and have the prompt be something like “Extract the invoice line items, product names, and prices as JSON”
-Parse the JSON response and map it where you need it.
This cuts out reliance on a third party app entirely. The AI can read the PDFs reliably without any encoding issues. You will also be able to get structured JSON with clean data.
Okay.
Thanks for your replies. I’ll see what’s the simplest option and requires the fewest credits or subscriptions possible.
That would be the built in Make functions that Neill recommended.
