For a company building AI tools for accounting

The Problem

Our client wanted to use AI to automatically extract data from invoices and receipts. The extracted data can then be programmatically added to an accounting software such as Quickbooks. This eliminates the need for manual data entry - which is significant for larger companies that may need to process hundreds of invoices and receipts on a recurring basis.

Our Solution

We used a multi-stage AI-based approach for this. In the first stage an AI model detected the layout of the invoice, including identifying header and footer fields and item tables. Each section was independently passed to an OCR model. For the item table, we used an OCR model specifically trained to understand and preserve the structure of the table (i.e., its rows and columns). Key features of our solution included:

- Automatically extract information such as billing address, due date, invoice date and line items

- Provide the location of each extracted information (e.g., where the billing address is written) to allow users to confirm that the information is correct. This helps users avoid the need of scanning through the entire invoice to find the relevant information.

Results

Our solution achieved:

- Near-perfect accuracy for most invoices and receipts, including scanned ones

- End-to-end automation of the entire workflow of extracting information from invoices and receipts and adding them to an accounting software