AI-Based Software Product for Fully Automated Invoice Processing

About Our Client

The Client is a US technology company which offers web and mobile solutions for procurement, expense, and accounts payable (A/P) automation powered up by robust automation, optical character recognition (OCR), AI and ML capabilities.

Challenge

The Client wanted to enhance their Microsoft Azure-based A/P management software product with a module for intelligent paperless invoice processing and was looking for an IT vendor with proven expertise in OCR and AI technologies.

Solution

ScienceSoft’s team covered the end-to-end development of an AI-powered invoice processing module, from introducing a Proof-of-Concept (PoC) to integrating the module with the Client’s core product. In a month after the initial request from the Client, the team analyzed the Client’s invoice processing needs and came up with an optimal concept of the invoice processing module.

The module was designed to facilitate and speed up invoice processing and ensure its high accuracy. Key features included:

• Automated upload of invoices in various formats (PDF, XML, digital images, etc.) from emails and connected systems (e.g., a supplier portal).
• OCR-enabled capture and extraction of numeric and textual invoice content.
• AI-powered validation of invoice data at the line-item level (invoice number, date, quantity, price, etc.) and identification of invoice exceptions, such as missed, mismatched, duplicate or inaccurate data.
• Automated routing of valid invoices for multi-department approval.
• Automated submission of approved invoices for payment.
• Centralized cloud storage for digital invoices.

To guarantee high availability, scalability, and easy evolution of the new invoice processing module, ScienceSoft’s team opted for the microservices approach to its architecture design. The team used Azure Service Fabric to streamline the development of microservices and employed Azure DevOps to automate and speed up their deployment. With the Infrastructure as Code approach, the team ensured quick and error-free configuration of the cloud infrastructure for the invoice processing module.

For security reasons, all microservices were hosted in a private network with a single public secure API Gateway as an entry point. Protection of data stored in the invoice processing module was achieved with data encryption, multi-factor user authentication and permission-based access control.

Instead of building a custom OCR engine for the module from scratch, ScienceSoft’s experts relied on Azure Cognitive Services Form Recognizer. This helped the Client optimize the project’s cost and cut the solution’s time-to-market while getting robust OCR.

Within the development service, ScienceSoft’s team integrated the invoice processing module with the Client’s existing A/P management product.

To assure smooth performance of the new module and its compatibility with the A/P product, ScienceSoft’s team conducted a series of frontend and backend unit tests with 95%+ coverage for particular pieces of code, load testing, and integration testing.

Results

The Client got a fully functional AI-based invoice processing module that ensures reduction in time and efforts for invoice processing while improving its accuracy and performs stably even under heavy load. The new module has extended the capabilities of the A/P management software product and made it more relevant to user needs.

Technologies and Tools

.NET Core, React, Redux, TypeScript, Microsoft Azure Service Fabric, Microsoft Azure DevOps, Microsoft Azure Cognitive Services / Form Recognizer.