Natural language analytics needs a semantic layer

Natural language analytics is fragile when it talks directly to raw tables.

A useful AI/BI system needs a semantic layer: business definitions, joins, metrics, filters, ownership, and acceptable query patterns. Without that layer, the model guesses intent from schema names and returns plausible but brittle answers.

The hard part is not asking in English. The hard part is defining what the data means.

Related:

• [AI/BI] Plug and Play Data Analytics