Dixstone automates invoice processing with generative AI

Context

Dixstone's Services branch aims to automate its invoice validation workflow, which currently requires four full-time employees.

The diversity of client invoice templates and the different names for the same product made automation extremely difficult. Generative AI addresses this challenge by having a more nuanced understanding of human language. It is more flexible and thus more robust to slight variations, unlike other IT systems. Additionally, it does not require training data to perform the task successfully. This frees up time for teams to focus on higher-value tasks.

The Challenge

Dixstone is a global industrial player operating with over 40,000 suppliers. Its financial teams currently spend significant time on repetitive, low-value tasks. Automating this process using deep learning is complex due to the need to generate a dataset and generalize it to new suppliers.

They needed quick initial results to demonstrate feasibility and gain internal support.

Solution

We have developed an email processing tool for invoices:

• It retrieves the PDF(s) from the email.
• It classifies the pages of each document.
• It extracts the values from each document to ensure compliance with the purchase order specifications.
• It sends a detailed error report to the user.

Tech Stack