Blueprints for AI-Driven Automation: Mining a SaaS Dependency Network for Agent Toolkits and Workflow Scenarios

While artificial intelligence (AI)-driven automation is rapidly advancing, design processes remain largely ad-hoc due to a lack of empirical analysis on real-world software-as-a-service (SaaS) workflows. To address this gap, this study proposes a data-driven framework that transforms SaaS usage patterns into actionable design resources. We analyzed 5,602 automation templates consisting of several SaaS applications from Zapier to construct a SaaS dependency network capturing cross-app connectivity. Based on this topology, the framework yields two core solutions for system configuration. First, for AI agents, the Leiden algorithm identifies compatible tool clusters, defined as “Toolkits,” to optimize tool selection boundaries. Second, for intelligent workflows, a subgraph pattern miner (SPMiner) extracts sequential motifs, which are subsequently translated into executable natural language scenarios using a large language model (LLM). The results demonstrate that empirical usage patterns can be systematically converted into ready-to-use automation templates with minimal engineering effort. This research contributes to the field by shifting the focus of business automation from adoption determinants to proactive engineering design. Practically, the proposed framework provides a scalable methodology for configuring AI agents and workflows, offering strategic insights for organizations seeking to navigate and automate within complex SaaS environments.