Building a Data Culture at EF Tours, and Learning What Our Data Products Are Worth...
Abstract
At EF Tours, we operate in a deliberately decentralized world; domains own their data, their tools, and their decisions. That autonomy drives innovation, but it also created silos, inconsistent definitions, and a lot of reinvented wheels. This talk is about how we turned that decentralized chaos into a real data culture, and what we learned once that culture matured enough to ask a harder question: what is our data actually worth?
We’ll walk through how Fivetran & dbt became the technical and cultural foundation for that shift; first dbt Core as a shared control plane for testing, documentation, and transformation, then dbt Platform as a catalyzer for unified metadata, observability, and cataloging across a decentralized mesh of teams. We’ll cover the medallion architecture that let domains keep ownership while still building on a common foundation, and the “you own your domain, we enable your success” model that made adoption stick.
Once that foundation was in place, a new question emerged from the business: not just “is the data trustworthy?” but “what is this data product costing us, and who’s actually using it?” We’ll show how we answered that with mart_metadata, a dbt-native observability layer that tracks build consumption, Snowflake credit costs, Power BI usage, and downstream query patterns; stitched into a single model of cost and usage per data product. We’ll cover the scoring methodology behind it, the cost attribution layer in dbt, the semantic layer in Snowflake, and how a Cortex Agent turned that intelligence into something conversational, so anyone can get a grounded answer without writing a query.
You’ll leave with a practical model for building data culture in a decentralized organization, a concrete scoring formula for dbt model performance, a pattern for attributing Snowflake credits to individual data products, and a working example of a Cortex Agent as the interface layer.