A process-orchestration reference architecture for end-to-end visibility across AI agents, RPA, and enterprise systems

Abstract

Enterprises increasingly combine robotic process automation (RPA), intelligent document processing (IDP), artificial intelligence agents, application programming interfaces (APIs), human tasks, and software-as-a-service applications to automate business operations. Although these technologies can improve individual activities, their isolated deployment frequently produces fragmented execution, disconnected monitoring, and limited visibility into end-to-end business outcomes. This paper proposes a processorchestration reference architecture that treats the business process as the primary unit of execution, governance, and observability. The architecture provides a common control layer for coordinating heterogeneous automation components while maintaining process state, correlating events, managing exceptions, enforcing human oversight, and preserving execution evidence. Its applicability is demonstrated through two enterprise scenarios: procure-to-pay invoice processing and order-to-cash payment collection. The scenarios combine IDP, RPA, AI-assisted exception analysis, APIs, human review, controlled retries, and enterprise-system integration. The architecture is also mapped to representative capabilities available in Camunda and UiPath Maestro. A scenario-based evaluation examines the architecture in terms of process visibility, cross-component traceability, governance, resilience, and adaptability. The proposed model provides technology leaders with a structured approach for progressing from disconnected automation toward governed and observable end-to-end business processes.