Slashing Invoice Processing Costs with BigQuery, DocAI, and GenAI

Our impact

• Deployed an invoice processing solution capable of reducing invoice costs for each invoice by 70%
• Increased efficiency with automation and provided visibility to LSS’ Accounting Operations team on the reasons why some invoices cannot be processed automatically, enabling them to correct upstream processes
• Provided an easy-to-use solution, simplifying the process for LSS’ suppliers with new capabilities, such as accepting any format of invoice

 

The challenge

LSS is responsible for processing an estimated 350,000 supplier invoices per year on behalf of its customers. 

Invoices are sent from suppliers to LSS’ customers in various different formats. But there are several time-consuming steps to process them; the document structure varies between suppliers, contents of the fields for each invoice need to be parsed correctly, and validation of the values of each field is complicated. 

This is compounded with additional challenges in implementing the logic to correctly identify the purchase order corresponding to the invoice. Complexities such as multiple purchase orders, line items and various tax rates within one invoice can trigger errors in the process, requiring manual effort to fix any issues.

LSS wanted to revolutionize the way its invoices are processed through standardized, automated & repeatable solutions, developed in-house so that LSS can enhance the solution over time. This would also enable them to unlock new revenue by providing invoice matching technology to other businesses via Google Cloud Marketplace.

 

Our solution

Seeking to accelerate its time-to-market, Liberty Shared Services collaborated with Datatonic to develop a cutting-edge solution on Google Cloud using BigQuery, DocAI and Generative AI.

The solution uses an automated pipeline to ingest invoices from a mailbox, parse the relevant information using DocAI and GenAI, before matching invoice data to PO data and submitting back to Oracle as ready for payment. 

For invoices that do not match, a user-friendly UI has been developed to handle the exceptions, speeding up time to resolve these invoices by 95%.

The technical solution features:

• Developing Google Cloud Build pipelines to deploy code automatically in Liberty Global’s Google Cloud environment
• Using Terraform to define all Google Cloud Infrastructure as code to enable cost and consumption reporting 
• Implementing a micro-batch ingestion of emails with invoice attachments using Cloud Functions
• Triggering Cloud Run to sent batch prediction loads to DocAI for entity extraction
• Supplementing the DocAI extraction by using GenAI to remove the need for routing and custom models, reduce the amount of complex logic (RegEx), and minimize the maintenance overhead
• Creating a custom User Interface running a React on Cloud Run to view failed validations, make amendments with a few clicks, reducing time spent validating by 95%
• Using BigQuery to store information on platform performance
• Creating performance dashboards in Looker to monitor platform performance, increase visibility, and help to drive process improvements

With the solution in place, Liberty Shared Services is reducing invoice processing costs by 70%.

Find out more about our Generative AI Back Office Agent to optimize your operations.