ACME Construction

Streamlining Invoice Processing with AI-Powered Automation

Executive Summary:

A mid-sized construction company with approximately 250+ employees, faced significant inefficiencies in processing invoices from multiple subcontractors and vendors owing to the volume and varied format of these invoices. The manual invoice entry process was time-consuming, error-prone, and costly. To address this, a custom document intelligence solution was developed using Azure Document Intelligence, Azure Open AI, Azure Logic Apps, Azure Functions, and SQL Database. The solution automated invoice detection, invoice data extraction, and seamless integration with the ERP system, reducing processing time as well as associated operations costs. This transformation allowed the construction company to reallocate staff to higher-value tasks, enhance operational efficiency, and maintain scalability for fluctuating invoice volumes.

Challenges Faced:

The transition to the new ERP system severed the integration between the company’s invoicing software and its ERP system. Without automation, the accounts payable team had to manually upload and input data from PDF invoices into the ERP, a process that was labor-intensive and prone to errors. Additionally, invoices arrived in varied formats, and some emails contained multiple invoices or non-invoice documents, complicating processing.

Business Impact:

Time Inefficiency: Manual processing consumed multiple hours per week for the employees, delaying payment cycles and straining vendor relationships.

High Costs: The labor cost for manual processing diverted resources from strategic initiatives.

Scalability Issues: Variable invoice volumes and multi-invoice attachments overwhelmed the accounts payable team especially during peak periods.

Objectives:

The objective was to develop an AI enabled automated invoice processing solution that:

• Eliminated manual data entry.
• Accurately extracted invoice data from diverse formats.
• Seamlessly integrates with the new ERP system.
• Scaled to handle variable invoice volumes, including multi-invoice attachments.
• Reduces operational costs and errors.

The Solution:

Project Overview

A cloud-based document intelligence solution was designed to monitor the company’s accounts payable mailboxes, detect and process invoices, and integrate data into the new ERP for further audit and approval. The solution handled diverse invoice formats, multi-invoice attachments, and other non-standard payment request forms, ensuring accuracy, scalability and reliability.

Key Features

Inbox Monitoring: Automated detection of incoming emails with attachments across multiple inboxes, mapped to specific company codes.

• Invoice Identification: Logic to distinguish invoices from non-invoice documents and identify multi-invoice attachments.
• Data Extraction: Extraction of key data points (e.g., invoice number, vendor code, subtotal) from varied invoice formats.
• Vendor Code Resolution: Fuzzy matching of vendor identification number using Azure Open AI to align with ERP vendor codes.
• ERP Integration: Automated API-based data posting to new ERP with error handling and audit trails.
• Audit Trail: Storage of raw data and processing metadata in a SQL database for tracking and reporting.

Technology Stack:

The solution included the following technology components -

• Azure Logic Apps: Monitored inboxes and orchestrated workflows.
• Azure Document Intelligence: Extracted key-value pairs from PDFs with a pre-trained invoice model.
• Azure Functions: Processed logic for invoice validation, splitting multi-invoice files, and ERP integration.
• Azure Open AI: Enhanced vendor code identification for ambiguous or inconsistent data.
• Azure Blob Storage: Stored processed invoice files.
• SQL Database: Logged processing metadata and outcomes.

Lessons Learned

• Adaptability is Critical: Diverse invoice formats required flexible logic, with Open AI bridging gaps in Document Intelligence’s pre-trained model.
• Edge Cases Matter: Non-standard documents (e.g., other payment forms) and multi-invoice attachments necessitated specialized workflows.
• Beta Testing is Essential: Initial testing with a limited dataset revealed format variability, prompting enhancements before full deployment.

Business Outcomes

• Time Savings: Considerable reduction in invoice processing time by freeing staff for strategic tasks.
• Cost Reduction: Critical savings achieved in operational costs, saving multiple thousands annually in labor expenses.
• Enhanced Scalability: Handled variable invoice volumes and multi-invoice attachments without performance degradation.
• Faster Payment Cycles: Shortened invoice approval time leading to improved sub-contractor & vendor relationships