Problem
Ticketmaster's operations teams watch high-demand on-sales in real time, and many sell out in 15 minutes or less. Questions come up mid-sale, but the data sits with engineers, PMs and data scientists, so getting an answer means finding the right person. The delay was human, not query latency. Engineering teams reported about 2 weeks per quarter spent on ad hoc data pulls, and a dashboard covers only the questions you can predict.
Example from the article: during an on-sale, someone asks "how many fans showed up to buy tickets last week?" and an engineer drops their work to write a throwaway SQL query. The answer arrives hours or days later.
Solution in brief
The product manager taught a Databricks Genie space the business: metric definitions, terminology and validated queries. Engineering then connected it to the operations tool through an API, and later a Slack bot. Suggested questions, live status messages and a weekly log-driven improvement loop built trust.
Questions this case answers
- How to let non-SQL business users get answers to ad hoc questions during live operations without waiting for an engineer
- How to keep a text-to-SQL assistant accurate after launch using conversation logs and validated question-to-query pairs
- How to let non-SQL business users get answers from data themselves with a text-to-SQL assistant
- How to cut ad hoc data requests to engineers and data scientists
Tools and methods
- Databricks Genie — text-to-SQL engine; a Genie space holds instructions, labels and validated queries
- Unity Catalog — governed data and access behind Genie
- AWS Lambda, GraphQL, React — backend, API and chat side panel
- Slack bot on Databricks Apps — second surface on the same Genie space
- Claude Code — runs the weekly improvement loop
Google or a one-time email code. No password.