Multi-Agent Reinforcement Learning for Autonomous Procure-to-Pay (P2P) Optimization

Procure-to-pay process is a very valuable business process having broad-based organizational performance and financial effects. Traditional P2P applications could not simultaneously optimize mutually competing objectives such as reducing costs, improving supplier performance, managing risks and compliance; etc. They introduce a new multi-agent reinforcement learning (MARL) system in this research work optimizing P2P autonomously to remove complexity due to interdependencies between procurement selection, relationship between suppliers, and organizational constraints. It outperforms current paradigms of optimization such as 96.0% classification capacity on classification of procurement decisions using ensemble methods on a full spend analytics dataset having 71 procurement features. It identifies key research gaps in existing literature about autonomous procurement systems and presents a full architecture importing intelligent agents for purchasing requisitions processing, selection of a supplier among procurement strategies, contract negotiations and payment optimizing process. Experimental results achieve 15% saved costs, 23% improved collaborative decision outputs, and 98.5% on-time shipment rates considering synchronized multiagent conduct.