Beyond layers and medallions: an explicit data modeling framework
Bolt's head of analytics engineering explains why medallion layers alone weren't enough to stop a growing dbt project from drifting, and the 14 explicit rules that replaced them.
Overview
Using dbt alone doesn't guarantee effective data modeling. Shares how standard layering failed a growing analytics team for lack of clear guidance, and the move to explicit patterns for intermediate models, naming conventions, mart-layer design, and maintenance at scale.
What you'll take away
- 01
Having dbt doesn't guarantee a good data model: Bolt wrote down an explicit 14-rule set once its analytics engineering team grew past 20 people and answers started to diverge.
- 02
Every table exposed to consumption tools must be one of exactly three shapes, and intermediate models are limited to spine, feature, or metrics, with metrics auto-generated from one centralized definition rather than hand-built.
- 03
Bolt's core advice wasn't about the specific rules: "you don't have to follow our rules, but I think you should have a written set of your own."
Session speaker
Silja Märdla
Bolt