What Worked, What Didn’t: Virgin Atlantic’s Flight to Trusted Generative BI

Tom Barber17 Jun 2026

Problem

Virgin Atlantic leaders wanted answers in hours, and their questions cut across commercial, operational and customer data. The BI team was a "report factory" with a Friday-to-Monday deck cycle. After the team connected an AI assistant, answers arrived in seconds. But business meaning lived in analysts' heads as tribal knowledge, so users stopped trusting the numbers. As speed grew, confidence fell, which the speaker calls the "trust gap". Strong governance (Unity Catalog, data quality frameworks) did not close the gap.

Example from the article: after getting an AI answer that "looked and sounded right", users opened one dashboard, then another, then asked an analyst whether the answer was correct.

Solution in brief

Make trust consumable by AI: every metric gets one signed-off definition and one authoritative Unity Catalog metric view, and the AI answers from that layer. Metrics are managed as products. Analysts and leaders get different entry points on the same governed foundation. Explainability is part of the user experience.

Questions this case answers
  • How to make answers from a generative BI assistant trusted by business users by grounding every metric in one signed-off definition
  • How to move a reporting team from one-off report production to governed data products that serve both analysts and executives
  • How to make answers from a generative BI assistant trusted by business users
  • How to govern metric definitions so that AI and analysts use the same numbers

Tools and methods

  • Unity Catalog metric views — the single authoritative asset per metric definition.
  • Metric sign-off — three parties approve each definition.
  • Markdown plus Git — storage and version control of definitions.
  • Databricks Genie — conversational access and explanation of answers; certified Genie spaces for leaders.
  • "Start here" notebook — per-domain guide to certified metrics.
  • Genie Code, OpenAI Codex — agents that draft assets for new analytics products.
Continue reading — it's free

Google or a one-time email code. No password.