From Words to Workflows: Automating Business Processes

TL;DR

This work introduces Text2Workflow, a generalized approach to convert natural language business requests into executable, JSON-based workflows. Central to the method is a Master+Expert prompt architecture that decomposes the task into a skeleton, detailed step definitions, and a human-in-the-loop for refinement, supported by a new Process2JSON dataset of 60 requests. Empirical results show Text2Workflow achieves strong JSON accuracy, particularly on hard tasks, while incurring higher token and time costs due to multiple prompts; ablations demonstrate the value of user feedback in improving correctness. The study highlights both the potential of LLM-driven intelligent automation for broad process automation and the practical challenges in formatting, parameterization, and security, outlining concrete directions for robust prompt design and evaluation improvement.

Abstract

As businesses increasingly rely on automation to streamline operations, the limitations of Robotic Process Automation (RPA) have become apparent, particularly its dependence on expert knowledge and inability to handle complex decision-making tasks. Recent advancements in Artificial Intelligence (AI), particularly Generative AI (GenAI) and Large Language Models (LLMs), have paved the way for Intelligent Automation (IA), which integrates cognitive capabilities to overcome the shortcomings of RPA. This paper introduces Text2Workflow, a novel method that automatically generates workflows from natural language user requests. Unlike traditional automation approaches, Text2Workflow offers a generalized solution for automating any business process, translating user inputs into a sequence of executable steps represented in JavaScript Object Notation (JSON) format. Leveraging the decision-making and instruction-following capabilities of LLMs, this method provides a scalable, adaptable framework that enables users to visualize and execute workflows with minimal manual intervention. This research outlines the Text2Workflow methodology and its broader implications for automating complex business processes.

Figures (62)

• Figure 1: General Process JSON.
• Figure 2: The process of breaking down a natural language user request into a workflow skeleton.
• Figure 3: The user feedback loop mechanism, enabling the user to validate and/or modify the workflow skeleton, if needs be.
• Figure 4: Complete workflow generation, including verifying and creating natural sounding questions for missing parameters.
• Figure 5: Manually completing the workflow by allowing the user to fill in the missing parameters.