What it is
Conversational analytics / text-to-SQL: you ask a question about your data in natural language, the system parses it, understands the database schema, generates SQL and returns an answer or a chart. BI without SQL and without queueing for an analyst.
Where it came from
The "search your data" idea lived inside ThoughtSpot for years, but the LLMs of 2024-25 made NL queries genuinely accurate. Now everyone has it: ThoughtSpot Spotter, Databricks Genie, Snowflake Cortex, Power BI Copilot, Looker + Gemini, Google Conversational Analytics API.
Why it took off
The new generation is not a question-answer box but an agentic analyst: ThoughtSpot Spotter breaks a complex question into steps itself, runs multi-step analysis and synthesizes the result. That removes the bottleneck — every employee can "ask the data" without pulling on the data team.
How to use it right now
- Turn on the conversational layer in your BI stack (Power BI Copilot / Looker+Gemini / ThoughtSpot).
- The key condition for accuracy is a governed semantic layer: single definitions for metrics (what "ACV" or "active user" actually means).
- Give non-technical teams direct access to asking questions, and free your analysts up for deep work.
- Review the generated SQL on business-critical reports.
What to watch out for
Without governance the model confuses metric synonyms ("ACV" vs "annual contract value") and returns different numbers — trust collapses. AI can confidently generate wrong SQL. For decisions where mistakes are expensive, keep an analyst in the verification loop.