Extract data from documents with Microsoft Flow

[Jay-Encodian]

Many of us over time will have worked on projects/solutions where there is a requirement to extract data from documents. A common scenario could be processing a scanned document or processing documents sent from an external source, commonplace in 'Invoice Processing' scenarios.

This step by step guide details how to configure a Microsoft Flow to extract data from a document and add to the document as metadata.

For the purposes of this guide we'll use this simple scenario;

The finance department generates invoices using a third party application which uploads the documents to a SharePoint library for storage. To enable invoice reporting, tracking and related activities we have a requirement to extract data from each invoice and add as metadata to the document.

The SharePoint library is configured as follows:

Lets create the Flow;

1. Create a new Flow using the 'Automated -- from blank' option

2. Enter a name for the Flow, select the SharePoint 'When a file is created in a folder' trigger, click 'Create'

3. Configure the 'When a file is created in a folder ' trigger action setting the 'Site Address' and 'Folder Id' fields to the location where documents will be added.

NOTE: For this demo; documents will already be in PDF format. However, should there be a need to extract data from a Word document, PowerPoint file, CAD drawing etc. simply convert to PDF first using the Encodian 'Convert to PDF' action

4. Add the Encodian 'Extract Text Regions' action

4.b. Filename: Select the 'File name' property from the ' When a file is created in a folder' action

4.c. File Content: Select the 'File Content' property from the ' When a file is created in a folder' action

To progress the configuration of the 'Extract Text Regions' action we need to provide co-ordinates of the data on the source document, i.e. Zonal extraction.

So how do we get the coordinates? Easy! simply use the 'Text Region Generator' utility found in the Encodian administration portal.

4.d. Upload a sample PDF document

4.e. Drag and move the area selector to the target area of the document

4.f. Define a name for the region and then click 'Add to JSON'

4.g. Repeat this process for all target regions of the document.

4.h. Copy the generated JSON data into your clipboard

4.i. Go back to Microsoft Flow; On the 'Extract Text Regions' action, click the 'Switch to input entire array' icon

4.j. Copy and past the JSON data obtained in step 4.h. into the 'Text Regions' field

5. We now need to obtain a sample of the generated JSON data which will enable us to add additional actions to parse and use the returned JSON data.

5.a. Test the Flow using your preferred method, click 'Save & Test'

5.b. For this example I selected 'I'll perform the trigger action' which I invoked by manually uploading a PDF invoice document to the SharePoint library aligned to the configuration of the trigger action (step 3)

5.c. Once the Flow has executed open the 'Extract Text Regions' action, copy the 'Simple Text Region Results'JSON returned.

NOTE: If you have submitted a large file Flow may display the outputs differently prompting you to manually download the output. See the example below:

Should this occur you'll need to manually download the payload, locate the 'Simple Text Region Results' variable. You'll also need to manually remove any escape characters '\' using either a text/code editor or an online service such as https://www.freeformatter.com/json-escape.html

6. Add a 'Parse JSON' action

6.a. Content: Select the 'Simple Text Region Results' property from the ' Extract Text Regions ' action

6.b. Click 'Use sample payload to generate schema'

6.c. Paste the 'Simple Text Region Results' obtained in step 5.c into the text-area control, click 'Done'

7. Add a 'Get file metadata using path' action

7.a. Site Address: Set as per step 3.

7.b. File Path: Select the 'File path' property from the ' When a file is created in a folder' action.

8. Add an 'Update File Properties' action

8.a. Site Address: Set as per step 3.

8.b. Library Name: Set as per the library name contained within the 'Folder Id' property of step 3.

8.c. Id: Select the 'ItemId' property from the 'Get file metadata using path' action

8.d. Map data from the 'Parse JSON' action to the relevant fields

9. Test the Flow by using data from the previous run

10. Validate the flow run has successfully executed

11. Validate data has been extracted and added as document metadata correctly

While this example has focused on how to extract document data before setting SharePoint document metadata, once the data has been extracted you can literally do anything with the data using the power of Microsoft Flow!

Thanks for reading!

[Anonymous]

Hi Jay,

Thanks for this helpful post. However, in step 4, I cannot find File Name or File Content in the dynamic content drop-down. Has those names changed since your post?

Thanks

Hadi

[Jay-Encodian]

Hi Hadi,

No, the dynamic data in Step 4 is obtained from the SharePoint trigger action.

Can you either create a post or contact Encodian support providing full details of your Flow so I can help?

Thanks Jay

[teylyn1]

@Jay-Encodian  You don't mention with a single word in your opening post that the method you describe requires a paid third party tool. Which your company builds. 

So, this is advertising.

Not useful. Rather annoying.

[Jay-Encodian]

@teylyn1 The Encodian connector is available under a free plan, so you can do this without incurring any cost unless you have a higher throughput requirement which would require a paid plan. 

Encodian have thousands of M365 tenants utilizing the free plan so it must be useful for some

Fair call, that this could / should be highlighted at the start of the post

[takolota]

If anyone wants to extract data from a PDF or image without training a model for select documents, try this new GPT data extraction method: https://powerusers.microsoft.com/t5/Power-Automate-Cookbook/Extract-Data-From-PDFs-and-Images-With-G...

It doesn’t require specifying certain document areas, wordings, styles, etc. It just OCRs the file, converts it to a replica text (txt), and passes it to a GPT prompt where you can ask GPT to do whatever you want with the document data.