Use a document processing model in Power Automate

1. Sign in to Power Automate.

2. Select My flows > New flow > Instant cloud flow.

3. Enter a name for your flow.

4. Under Choose how to trigger this flow, select Manually trigger a flow, and then select Create.

5. Expand Manually trigger a flow, and then select +Add an input > File as the input type.

6. Select +New step > AI Builder, and then select Extract information from documents in the list of actions.

7. Select the document processing model you want to use, and then select the document type.

8. In the Form field, add File Content from the trigger.

   

9. In the successive actions, you can use any of the fields and tables extracted by the AI Builder model. For example, let's say that our model is trained to extract the Lot number, the Net weight, and the Gross weight values. We also want to post these to a Microsoft Teams channel after AI Builder has extracted them from the document. Just add the Post a message action from the Microsoft Teams connector, and then select your fields from the list of tokens.

   Note

   • To retrieve the value for a field, select <field_name> value . For example, for the Lot number field, select Lot number value.
   • To retrieve the value for a checkbox, select <checkbox_name> value. For example, for a checkbox named Priority shipping, select Priority shipping value. The return value is of type Boolean: true if the checkbox is marked as selected in the document, false if it’s not.
   • To retrieve the confidence score for an extracted item, select <field_name> confidence score. For example, for the Lot number field, select Lot number confidence score.

   

Congratulations! You've created a flow that uses an AI Builder document processing model. Select Save on the top right, and then select Test to try out your flow.

Page range

For documents that have multiple pages, it's possible to specify the page range to process.

1. On the Extract information from documents card, select Show advanced options. This option changes to Hide advanced options.

2. In the Pages parameter, enter a page value or page range. Example: 1 or 3-5.

   

Note

If you have a large document with only one form, we strongly recommend you use the Pages parameter. Doing this can reduce the cost of model prediction, which can increase performance. However, the page range should contain a unique form for the action to return correct data.

Example: A document contains a first form in page 2 and a second form that spans over pages 3 and 4:

• If you enter page range 2, it will return the data of the first form.
• If you enter page range 3-4, it will only return the data of the second form.
• If you enter page range 2-4, it will return partial data of first and second form (should be avoided).

Parameters

Input

Name	Required	Type	Description	Values
AI Model	Yes	model	Document processing model to use for analysis	Trained and published document processing models
Document type	Yes	list	The file type of the form to analyze	PDF Document (.pdf), JPEG Image (.jpeg), PNG Image (.png)
Form	Yes	file	Form to process
Pages	No	string	Page range to process

Output

Name	Type	Description	Values
{field} value	string	The value extracted by the AI model
{field} confidence score	float	How confident the model is in its prediction	Value in the range of 0 to 1. Values close to 1 indicate greater confidence that the extracted value is accurate
{table}{column} value	string	The value extracted by the AI model for a cell in a table
{table}{column} confidence score	float	How confident the model is in its prediction	Value in the range of 0 to 1. Values close to 1 indicate greater confidence that the extracted cell value is accurate

Note

• More output parameters may be proposed such as field coordinates, polygons, bounding boxes and page numbers. These aren't listed on purpose as they're mainly intended for advanced use.

• Coordinates are represented as percentages of the document's height and width, originating from the top-left corner. For instance, if coordinates X = 0.10 and Y = 0.20 are given, this signifies a location at 10% of the document's width along the X-axis and 20% of its height along the Y-axis, both measured from the top-left corner.

Common use cases

Iterate a document processing table output in Power Automate

To illustrate this procedure, we use the following example where we've trained a document processing model to extract a table that we've named Items with three columns: Quantity, Description and Total. We wish to store each line item from the table into an Excel file.

1. Select the field you wish to write the cell for a table. The dynamic content panel will open showing everything that the document processing model knows how to extract. Search for {your table name} {your column name} value. Our example uses Items Quantity value.

   

2. Once you add this value, the action where you added it is automatically inserted into an Apply to each control. This way, every row in the table will be processed when the flow is run.

3. Keep adding columns you want to iterate.

Note

Tables extracted by document processing currently don't return a confidence score.

Process outputs of checkboxes in Power Automate

Checkbox values are of type Boolean: true means the checkbox is marked as selected in the document, and false means it’s not.

One way you can check its value is with a Condition action. If the checkbox value is equal to true, then execute one action. If the value is false, execute a different action. The following illustration shows an example.

Another option is to map the true/false output of the checkbox to other values of your choice by using the if expression. For example, you might have a column in an Excel file where you want to write ‘Priority’ if one of the checkboxes in the document is selected, or ‘Non-priority’ if not selected. To do this, you can use the following expression: if(<document processing output>, 'Priority', 'Non-priority'). The following animation shows an example.

Remove currency symbols (€, $,…) in a document processing output in Power Automate

To illustrate, the Total value extracted by the document processing model might have a currency symbol, for example, $54. To remove the $ sign, or any other symbols you want to omit, use the replace expression to remove it. Here's how:

replace(<document processing output>, '$', '')

Convert a document processing output string to a number in Power Automate

AI Builder document processing returns all extracted values as strings. If the destination where you want to save a value extracted by AI Builder document processing requires a number, you can convert a value to number using the int or float expression. Use int if the number has no decimals. Use float if the number does have decimals. Here's how to do it:

float('<document processing output>')

Remove blank spaces in a document processing output in Power Automate

To remove blank spaces from output values, use the replace function:

replace(<document processing output>, ' ', '')

Convert a document processing output string to a date in Power Automate

AI Builder document processing returns all outputs as strings. If the destination where you want to save a value extracted by document processing is required to be in date format, you can convert a value that contains a date into date format. Do this by using the formatDateTime expression. Here's how to do it:

formatDateTime(<document processing output>)

Filter email signature from a flow so that it's not processed by the document processing model (Microsoft 365 Outlook)

For incoming emails from the Microsoft 365 Outlook connector, email signatures are picked up by Power Automate as attachments. To keep these from being processed by the document processing model, add a condition to your flow that checks if the output from the Microsoft 365 Outlook connector named Attachments is Inline is equal to false. In the If yes branch of the condition, add the document processing action. With this, only email attachments that aren't inline signatures will be processed.

See also

• Overview of the document processing model
• Training: Process custom documents with AI Builder (module)