How to Extract Data From Invoices in Power Automate

This guide shows how to extract structured information from invoices in Power Automate using an AI model. You’ll end up with JSON output and variables that you can use to for example write the extracted data to Excel, Dataverse, SQL, or SharePoint. We’ll also set up a human-in-the-loop for uncertain predictions.

Before we start

You'll need:

• A Microsoft account with a Power Automate license

• A Cradl AI account

Step 1: Setup your AI agent

In Cradl AI, choose Invoices in the onboarding dialog. This simply gives you a starting point with a set of pre-defined fields. You can customize the fields to fit your exact requirements at any time. For now, we’ll stick with the auto-generated ones to keep things simple.

Configuring your AI agent.

Next, go to the Workflow tab and add both a Power Automate trigger and a Power Automate export. When you’re done, your workflow should look something like this:

Screenshot of Cradl AI workflow.

Step 2: Create a Connection

Next, head over to Power Automate and either open an existing flow or create a new one.

In this example, we’ll assume invoices arrive by email and that we want to write the extracted data to an Excel sheet. In a real-world setup, you’d likely send the data to an ERP system or store it in something like a Dataverse table, but we’ll use Excel here to keep things simple.

To connect Cradl AI to your Power Automate flow, add a new action and search for “Extract data from document.” Select it, then create a new connection to your Cradl AI account:

Copying Client Credentials from Cradl AI to Power Automate.

In Cradl AI, go to Workflows, click Export to Power Automate, and copy the Client Credentials. Then paste those credentials into the Client Credentials field in Power Automate.

Next, configure the Extract data from document action. Select your newly created Invoices agent from the dropdown. For the Document field, choose the Content Bytes variable from the Download the email attachment step.

Configuring the Power Automate action.

Step 3: Build the Power Automate flow

Now let’s put everything together by building a Power Automate flow that:

• Retrieves emails from Outlook

• Downloads any PDF attachments

• Sends those attachments to Cradl AI for processing

• Writes the extracted data to an Excel sheet

When we’re done, your flow will look like the diagram below. If you’d rather skip ahead, you can download the complete Power Automate solution here and start using it right away.

The Power Automate workflow.

Here’s how to configure each action:

When a New Email Arrives (V3): In the trigger, enable Include Attachments.
You can also enable Only with Attachments to ensure the flow only runs for emails that actually contain attachments.

Get Attachment (V2): Use the Message ID and Attachment ID from the previous step. Because Attachments is an array, Power Automate will automatically create an Apply to each loop, so the flow runs once for every attachment on the email.

Condition: Next, add a Condition to avoid processing unnecessary files. In this example, we only want to process PDFs, so configure the condition to continue only when the Content-Type equals application/pdf.

Extract Data from Document: Select the agent you created earlier and map the downloaded attachment from the previous step as the document input.

Add a Row into a Table: Finally, append the extracted data to your Excel table. Select the appropriate workbook and table, then map each extracted field to the corresponding column.

Step 4: Take it for a spin!

Now you’re ready to test the flow. Because we set it up to process invoices received by email, send a sample invoice to the email address you configured and make sure everything runs as expected:

Human-in-the-loop.

Since this is the first document being processed, the AI will be conservative about what it approves automatically. In this case, you might be asked to confirm that the invoice date was interpreted correctly. After you’ve validated a few invoices, the AI will become more confident over time. You can also adjust the automation thresholds manually if you want more, or less, automatic approval.

Wrapping up

That’s it, that’s all you need to build a basic invoice processing pipeline in Power Automate using Cradl AI.

The next step is to add more advanced processing and validation logic. For example, you can match invoices against purchase orders to automatically verify amounts, line items, or vendors. This helps you build a more robust and reliable accounts payable process, with fewer manual checks and fewer surprises downstream.