Document Type
Article
Publication Date
5-9-2026
Abstract
Service organizations increasingly seek to automate customer service workflows, yet most customer communications arrive as unstructured data, such as email and chat. Although large language models (LLMs) offer new possibilities for interpreting such data, their non-deterministic behavior creates reliability risks that can undermine automation performance, as error rates as low as 10% can cause automated flows to perform worse than manual ones. This research discusses how firms can build IT capabilities that convert LLM-based interpretation of unstructured data into reliable automated customer service workflows. Using the Resource-Based View (RBV) as the theoretical lens, the study examines how knowledge sharing and governance contribute to an IT capability for large-scale workflow automation. The study also develops and evaluates an N8N-based prototype that combines LLM agents with state model definitions to preserve deterministic workflow execution as a practical example of best design practices.