Encoding Your Domain Expert: The Context Layer Behind Spotify's Data Assistant

Spotify's data assistant depends on experts teaching it what company data means. Each subject area has curated table information, approved question-and-query examples, and business definitions. The assistant selects that context, writes and executes a database query, and returns the answer with its query and sources. Spotify's engineering account explains how this makes expert knowledge reusable.

Curation proved selective: experts accepted only 12.5% of examples proposed from historical queries. Rejected examples included debugging and one-off exploration, as well as misleading patterns. That percentage measures suitability as teaching material, not the accuracy of employees' queries. A query can answer one person's immediate question while being a poor template for future answers.