How to Streamline Invoice Processing with AI

May 7, 2025

Toni Lopez

Software Engineering at Stack AI

Processing invoices manually makes up 62% of accounts payable tasks, with each costing between $12 to $35 in labor. With potential human errors and higher risk of fraud in the mix, AI automation is poised to be a safer, faster, and more efficient option.

Despite being a big problem, the solution is easy to build with Stack AI, a generative AI workflow automation platform. In this guide, we’ll walk you through how to upload an invoice, process it with AI, and store it in database software for future reference.

Streamline invoice processing with AI

File upload input
Process the invoice with GPT-4o
Save the invoice details to Airtable
Clean the output for better user experience
Export as a form interface

File upload input

For this guide, we’ll be using the Workflow Builder. Select the New Project to start from scratch. This option creates a project with 3 nodes as a base:

A text input
A LLM node, usually set to an OpenAI model
An output node to display the results

For this guide, we want the prompt to be an invoice file. Delete the text input by clicking the cross icon.

On the left menu, expand the Inputs section. Then, drag and drop the Files node onto the canvas.

Notice the Files node comes with the Expose as input setting activated by default. This will let your team upload files to the tool once you export it. Drag the Files output handle onto the OpenAI’s LLM input handle.

Process the invoice with GPT-4o

To process the invoice, we’ll use the GPT-4o model. Due to its multimodality and vision capabilities, it can accurately understand images, read documents, and extract text.

The OpenAI LLM node is already on the canvas. Make sure the GPT-4o model is selected from the dropdown.

LLM system instructions

The model needs system instructions to understand how to process the uploaded files. We’ve prepared a template you can adapt to your needs.

You are an expert document parser specialized in processing invoices. Your task is to extract key invoice details from the raw text or OCR output of an invoice. Be accurate, consistent, and return the data so it can be written to Airtable (follow the Airtable API guidelines).

Extract the following elements. If you can't find any of these, just write "null":

invoice number,

invoice date,

vendor name,

vendor email,

vendor phone,

bill to,

total amount,

currrency,

bank name,

bank account number,

description,

invoice total

Please note that this system prompt assumes you’ll be using Airtable as data storage and the most common elements to extract from an invoice. Edit these instructions according to your use case.

LLM prompt

The prompt will include a simple instruction to start running the action along with the document.

Write the following in the prompt input field:

Process this invoice:

Then, type a forward slash, bringing up all the available data inputs. Select the Files input node from the dropdown. This will pull the document into the prompt, making it available to GPT-4o for processing.

Save the invoice details to Airtable

For this tutorial, we’ll be storing the invoice details as a row in an Airtable base. Expand the Apps section in the left menu, select the Airtable app, and drag the Write to Airtable action node onto the canvas, to the right of the LLM node.

Click the Write to Airtable node to open its settings. Click the New Connection button to begin connecting Airtable to Stack AI.

Follow the instructions in the pop up windows to successfully connect both platforms. Once complete, click the Test button to check connection health.

Connect the OpenAI LLM node to Write to Airtable.

This node needs to know what to write to Airtable. In the Inputs > Data section, type a forward slash in the input field and select the LLM node.

Next, we need to target the Airtable base and table to store the information. Open the Airtable base where you want to save processed invoice data. Fill the Base ID with the text right after the Airtable domain in the URL input of your browser.

Fill out the table name where the data should be saved. This is visible in Airtable’s user interface. Type the exact name.

Stack AI and Airtable are now properly connected and ready to transfer data.

Clean the output for a better user experience

At this point, you’ve built the core functionality of the tool. However, when running, your users will see a snippet of JSON (JavaScript Object Notation, a notation system for safe and accurate data transmission between machines). Let’s improve user experience by providing a message indicating whether the action completed successfully or not.

Expand the LLMs section. Drag and drop another OpenAI node onto the canvas, to the right of the Write to Airtable node.

Connect the Write to Airtable action to the new OpenAI LLM.

The GPT-4o Mini model is a good match for this task, as it doesn’t require a high level of intelligence. Use these system instructions as a template to interpret the output of the Write to Airtable action.

You're a message formatter.

If the user message has any content, format it in a list format to show that an invoice was successfully saved.

If the user message has no content, state that there was an error while saving the invoice.

Provide a prompt containing the data from the Write to Airtable action. Write the following in the prompt input field:

User Message:

Then, type a forward slash to open the data sources dropdown. Add the Write to Airtable (Record ID). Repeat the process to add the Write to Airtable (Fields) data.

Connect this LLM to the output. Its response will be exposed to the user, letting them know the status of the processing task.

Export as a form interface

Click the Publish button at the top right to save your project.

Click the Export tab at the top left of the screen.

This screen will let you configure how the tool looks like to your users. Start by changing the app name and the description.

Rename the input field’s alias to Upload invoice. This will update the label on the interface.

Scroll all the way down the settings to find the Security controls. You can set up password protection, SSO and access control to protect this interface from unauthorized use. Edit these settings as needed.

Click the Save Interface button at the top right to save changes. These will now be available on the tool’s interface.

Share and improve

Share the tool with your team by copying the Published Interface link and pasting in in an email or message.

Explore usage in the Analytics tab. Here you’ll be able to see total runs, number of users, errors, tokens consumed, as well as a list of all runs. You can generate an AI report of all data and download logs for further analysis.

Reduce invoice processing time

Invoice processing is a time-consuming task, a waste of your finance team’s brainpower. Automating it with AI reduces costs, improves accuracy, and reclaims your team’s mental space so they can use their skills elsewhere. Sign up for a free Stack AI account and discover other use cases you can automate: