An intelligent automation framework for optimizing purchase invoice processing

Abstract

The vast majority of processing of purchase invoices still occurs manually; on average, it costs $9.87 per invoice and can take 12–18 days, as well as having an estimated 12–15 percent error rate. Organizations lack systematic guidance for implementing intelligent automation solutions. This study employed a design science research (DSR) approach to design an integrated, technology-independent framework that incorporates Artificial Intelligence (AI) technologies–Intelligent Document Processing (IDP), Large Language Models (LLMs), and Machine Learning (ML) with traditional automation–Optical Character Recognition (OCR) and Robotic Process Automation (RPA). The integrated framework includes five loosely coupled layers: invoice ingestion, document understanding and extraction, validation and business logic, human-in-the-loop decision-making, and Enterprise Resource Planning (ERP) system integration. It also includes technology selection matrices that enable organizations to identify optimal methods for extracting data from their suppliers based on characteristics of their supplier portfolios and value-stratified confidence thresholds to operationalize the degree of risk-appropriate human oversight. Evaluation of the framework using scenarios in both manufacturing and non-profit health care settings demonstrated a potential 50-56% reduction in total cost, a 40-56% improvement in cycle times, and a 70-80% reduction in error rates with a 12-18 month payback period. The framework extends technology acceptance model (TAM) through trust calibration constructs, operationalizes hybrid intelligence theory via confidence-based task allocation, and demonstrates viability for resource-constrained organizations through LLM zero-shot learning and file-based legacy ERP integration.