How to Get Tables from Documents in Power Automate

Table data isn’t straightforward to extract from a document. Depending on the document type and format, table rows, columns, headers, totals, and line items aren’t always structured the same way.

AI table extraction makes dealing with inconsistent document types and table structures far easier to manage. A smart document AI model can intelligently identify the tables in a document and return the contents as an organized object containing all rows, cells, headers, and values.

Implementing Table Extraction in Power Automate

Below, we’ll walk through table extraction in Power Automate using the Document AI connector. The Document AI “Extract Tables from a Document” action locates and retrieves data from all tables in a document, and it doesn’t require any extra model training: it supports a wide range of different document types (PDF, DOCX, XLSX, PPTX, JPG, PNG, etc.) by default.

We’ll begin by creating an Instant Cloud Flow with a manual trigger.

Next, we’ll search for the Document AI connector in the Power Automate connector library.

From the actions list, we’ll select the action “Extract Tables of Data from a Document using AI”. We’ll then create our connection with a Cloudmersive API key (we can get one by creating a free account on the Cloudmersive website).

To configure the “Extract Tables” action, we’ll simply provide our file bytes and file name in the bottom two parameters. We’ll leave Recognition mode empty to it defaults to “Advanced” recognition.

Next, we’ll (optionaly) add a Compose action to capture the “Extract Tables” output. This simply makes the output of our test run cleaner to look at.

Now we’ll test our flow with an example document. In this walkthrough, we’ll use an Invoice document for testing. Invoices usually have tables of data in unpredictable formats, and this connector action makes retrieving that data easy.

When our flow finishes running, we’ll find all the tables in our invoice neatly presented in the Compose action output.

Here’s what the invoice looked like:

That’s all there is to it: we can now add one quick action to our flow and accurately extract all the table data into a JSON array.