Lean Six Sigma (LSS) improvement work increasingly depends on information-intensive activities such as document handling, data interpretation, reporting, and communication, yet current discussions of Artificial Intelligence in LSS remain largely technology-centric. This paper proposes a task-first, process-centric framework to support the governed application of Large Language Model (LLM)-enabled tools in such environments. The study makes three contributions: (i) a set of cross-functional organizational process types relevant to LSS practice, (ii) a functional classification of recurring tasks and LLM-enabled tool categories, and (iii) a dual-encoded task–tool matching matrix that separates alignment strength from interaction mode, distinguishing capability fit from governance logic. The framework is empirically anchored through two real-world industrial applications: customs document processing and shop-floor data digitalization and reporting. The results show that (i) stronger outcomes emerge when LLM-enabled support is matched to bounded, repetitive, and structured work, or when analytical support is built on stable and traceable data layers; (ii) operational value depends not only on technical capability, but on workflow embeddedness, data readiness, and human validation checkpoints. The framework also clarifies where support, augmentation, and partial automation are appropriate for different task classes and under explicit accountability constraints in information-intensive administrative work connected to improvement practice and governance.
Carvalho et al. (Mon,) studied this question.