AI Innovation in Utility Invoice Processing
A Conversation with Rod deVillers, VP Product, Delivery and Production Operations at TEOCO
TEOCO’s Invoice Processing Automation service is trusted by many of the largest U.S. operators to collect, process, analyze, and pay their wholesale telecom invoices. Now, TEOCO is bringing the same expertise to client utility invoices. Jacob Howell, Executive Director of Business Services, sat down with Rod deVillers, VP Product, Delivery and Production Operations, to discuss how TEOCO is leveraging AI-powered automation to transform utility invoice processing. Below are highlights of the conversation edited for length.
Interview Highlights
How is TEOCO leveraging large language models for utility invoice automation?
TEOCO uses Microsoft Azure AI Document Intelligence, a commercially available large language model (LLM), to automate utility invoice processing. While Azure is a powerful tool, it was originally designed for generic invoices and doesn’t fully address the complexities of telecom and utility invoices, which can vary widely by state and vendor. To overcome these challenges, TEOCO has invested heavily in training and customizing the model, especially when onboarding new vendors.
Currently, TEOCO processes about 18,000 paper utility invoices per month, automating the capture of key fields like account numbers and bill dates. This automation has reduced manual data entry by up to 50%, and ongoing improvements aim to increase that number even further.
How does your team leverage automation while tackling the complexity of invoices requiring manual review?
Despite these advances, many paper invoices still require manual review. TEOCO’s process ensures that duplicate invoices—such as both paper and electronic versions—are identified and handled appropriately. Paper invoices are sent to the invoice automation team for manual entry, a labor-intensive process that TEOCO is working to further automate.
To address this, TEOCO is developing a Python-based AI solution that leverages LLMs to digitize every piece of text from paper invoices, including important vendor messages about outages or upgrades. This comprehensive data capture provides valuable insights for downstream systems.
How do you address all the different types of invoice formats?
Invoice formats can differ dramatically, even among vendors using the same billing system, due to state regulations and local requirements. Sometimes information is neatly organized in tables; other times, key details are scattered across the page. TEOCO’s AI agents are being trained to handle these variances, with the goal of creating a core model that fits most vendors and supports business scalability.
With greater automation, TEOCO can easily add new clients, whether regional, nationwide, or from different industries. The framework being built will allow for faster, more accurate invoice recognition and processing, focusing on essential data points like account number, vendor name, address, and amount due.
Is your team able to tackle these different invoice formats without delays or disruptions?
TEOCO’s model is designed to be flexible, supporting variations in field names (e.g., “bill date” vs. “invoice date”) and adapting to new prompts for less common fields. While the model looks for keywords, it also accounts for nuances in vendor formats, inferring information when necessary.
Processed invoices are output in JSON format, enabling transformation and loading of data, which minimizes errors and rework. Users can add validation rules directly, reducing engineering time and allowing for immediate updates.
The Road Ahead
TEOCO’s solution is powerful and future-proof, supporting ongoing improvements and scalability. By combining deep industry expertise with advanced AI, TEOCO is setting a new standard for utility invoice automation.
Microsoft, Azure, and Document Intelligence are trademarks of the Microsoft group of companies.
For more information about TEOCO’s suite of services, visit: https://aircomnext.com/utility-expense-management/
