A Retrieval-Augmented Generative AI Framework for Intelligent Enterprise Workflow Design and Process Optimization

Abstract

Designing and improving enterprise workflows is slow, expert-driven work. Analysts interview staff, study event logs, consult
policies, draw process models, and estimate the effect of changes, often with limited evidence about how a redesigned process
will actually perform. Large language models (LLMs) can now read requirements, draft process models, and explain design
choices, and business process management (BPM) research has begun to test them on process tasks. Used alone, however,
they may produce designs that ignore organizational policies, repeat known mistakes, or perform poorly under real workloads.
This article proposes a retrieval-augmented generative AI framework for workflow design and process optimization. The
framework grounds LLM-generated designs in a process knowledge base containing reference models, past designs, policies,
and event logs. It verifies designs for correctness and compliance and evaluates alternatives through simulation models
discovered from event logs, combined with intelligent resource scheduling. Approved designs are deployed to a configurable
workflow engine, and performance feeds back into the knowledge base. Using a design-oriented approach grounded in ten
studies published between 2019 and 2024, the article maps the evidence base and presents the framework and its design-todeployment
loop. It sets out the generative tasks and their checks, and the optimization levers the framework can evaluate.
It illustrates the framework with a telecom company redesigning its broadband installation process. It argues that generative
AI accelerates workflow design most safely when every draft is grounded in organizational knowledge and tested against
realistic simulation before it reaches production.