This paper proposes and evaluates a three-tier decision architecture for automated fault triage in SAP ERP environments. Tier one uses deterministic rule matching, tier two applies gradient boosting classification, and tier three uses Large Language Model reasoning with retrieval augmentation for complex cases. Grounded in the diagnostic framework of Veershetty (2026) and deployed following the clean-core extension principles of Veershetty (2025), the architecture is evaluated across five SAP ERP installations over 15 months, achieving 84% diagnostic accuracy and resolving 91% of incidents without human escalation.
Subramaniam et al. (Mon,) studied this question.