Business Problem & Solution

"What was revenue by region last quarter?" is a thirty-second question that, in most companies, takes a day: someone files a ticket, an analyst writes SQL, and the answer comes back stale. Self-serve BI promises to close that gap — but the moment you point natural language at a warehouse, you have handed every executive a way to accidentally query row-level guest data.

Executive BI Chat solves the governance problem first and the chat second. It answers in plain English, but it can only ever read four governed, PII-free metric views. The natural-language layer chooses a view, a measure, and a dimension; it never sees a guest table.

It proves that a safe, self-serve analytics surface is a governance design, not a model trick: the question is routed to a governed view by deterministic logic, the generated query is shown in full for transparency, and the narrative explicitly states that no row-level guest data was accessed.

It simplifies the model layer. The query plan and the aggregation are deterministic over a synthetic dataset; the metered AI step only phrases the narrative. That keeps the demo honest about where the intelligence is and where the guardrail is.

Synthetic aggregates only — invented bookings, reviews, segments and partner scorecards across a handful of regions and months. No real customers, no PII, no production prompts. Every figure shown is representative and labelled; the four governed views are the entire surface area the assistant can touch.