Anthropic reports that Claude now handles approximately 95% of its internal analytics queries with ~95% accuracy, up from just 21% without structured skills. The key insight is that success came not from model improvements but from strong data governance: governed canonical datasets, semantic metric definitions, centralized metadata, and encoded analytical workflows (skills). The setup uses four layers — data foundations, a knowledge layer, skills, and validation systems — to reduce ambiguity and prevent metric drift. Community reaction is mixed, with some praising the semantic layer approach and others questioning whether AI-driven analytics can meet the deterministic standards expected of BI systems.
