PDF - Extract text from PDF - Alternative Options

[BI_RPA_Andy]

Hey everyone!  First post here, just starting to play with Power Automate Desktop.  In an effort to move off of a larger more expensive RPA platform, I am putting together some POC's and doing some feasibility planning with Power Automate to show that it will be sufficient to replace our existing platform.  One of our larger existing solutions on our current platform involves parsing a lot of data from insurance applications that we receive as PDF files.  I am running into an issue with the built in PDF - Extract text from PDF function in that it is returning the text in a different way than expected, to the point that we would not reasonably be able to consume it.

Here is a mocked up version of one page of an application...

When using the PDF - Extract text from PDF action in PAD, this is how the content is returned...

This makes it nearly impossible to confidently ascertain which data should belong to which fields.

If I open the PDF file in Adobe Reader DC and use the built in 'Export PDF' tool and export it to a .txt file, the content gets rendered like this...

The field values come over adjacent to the field names making it consumable.  This also happens to be the way the content is returned in our existing platform with its built-in 'Extract Text From PDF' command.

Conclusion:  I need to find an alternative method that will extract the text from these files in a format that will be consumable.  Worst case scenario is I would just automate the interaction with Adobe Reader DC but I was hoping there might be a better alternative.  Thanks!

[Pstork1]

I can think of two different approaches to this.

1) If the format of the PDF is predictable you might want to look into Power Apps AI Builder.  You can use that to map the fields in the PDF and extract the values directly.  Use AI Builder in Power Automate - Power Automate | Microsoft Docs

2) If you can't use AI Builder because the fields will change position then you could use Power Automate Desktop (RPA) to convert the PDF to text using Adobe Acrobat.

[MichaelAnnis]

Might sound stupid, but how about:

Send Keys, Control A

Send Keys, Control C

Open New Excel

Send Keys, Paste clipboard content to A1

Use Excel UI to Set Text to Columns delimited by ":"

This will result in your row titles in column A and your populations in column B

[BI_RPA_Andy]

Good idea.  Tried it manually and unfortunately it produces the same undesired output.

[BI_RPA_Andy]

Thanks for the suggestions!

1. Played around with this already and unfortunately not all of the pages on the PDF file are standardized.  The sample I showed you was just the first page, but they can be upwards of 70 pages with additional documentation.  Some of them are standard forms but it's not consistent enough.  I would have to basically pull apart every page and examine them individually which seemed like not the easiest solution.

2.  This is the route I am probably going to end up going.

[BI_RPA_Andy]

I have also attempted to use the Power Automate Adobe PDF Services connector which has the Export PDF function available, but I'm struggling to figure out how to use it.  I have the connection set up, it's the Actions I am struggling with.

Here is the documentation for it .... https://docs.microsoft.com/en-us/connectors/adobepdftools/ ...

Here is the section for the Export PDF Action (https://docs.microsoft.com/en-us/connectors/adobepdftools/#export-pdf-(preview)) 

What I am confused about is the required input parameter 'File Content' which is described as 'The content of the source file'.  This seems particularly odd given the purpose of exporting the PDF to another file type is so that I can get the content.  So I'm not sure where to go from there.

[vanderben]

File Content is which file you are wanting to convert. You might store this in Microsoft SharePoint, OneDrive, etc. and access it from one of those connectors. One of the examples would be this template: https://us.flow.microsoft.com/en-us/galleries/public/templates/15eb05ba02924398a22edb0091896add/conv...

[Geezer]

Hi @BI_RPA_Andy  I tried this early on and found absolutely no workaround to extracting from PDF. The various connectors will be able to create Excel files, which will show the same indiscriminate splitting/combining of info into adjacent cells.  Even the AI Builder, I was recently told, can require 1000+ training documents to get 99% accuracy, but will not guarantee 100% results. As a consumer of insurance products, I would find that unacceptable from an insurance provider. 

What is beginning to take shape for me, at least, is to bypass the PDF step altogether and extract the data from upstream--the process which would generate the PDF.  In our case, this is to use attended RPA from PAD ( "web recorder") to extract the data from the web site screen itself. The HTML grouping and formatting is much more predictable than the PDF extraction.

There is a MySQL connector as well, which I'm told could be used to query the web site's database directly. But I'm personally too lazy to learn yet another programming language at this point 🙂

[BI_RPA_Andy]

Thanks everyone.

I went ahead and just created a PAD flow to open the file directly in Adobe Acrobat Pro DC and used UI control/image recognition to do all the clicks necessary to use the 'Export PDF' tool to convert the file to .txt, read the text file and return the contents as an output variable.  It's not as slick as I was hoping for but it achieves what I need it to for this specific process.

[Ricardo111]

useful, pdf text exactration via browser pdf viewers does not work on browser automation.

[takolota]

You can now also try this template for extracting data from PDFs or Images with GPT in a cloud flow: https://powerusers.microsoft.com/t5/Power-Automate-Cookbook/Extract-Data-From-PDFs-and-Images-With-G...