Semantic layers are in
Main idea: define your dataset and metrics, map out their relationships, translate semantic query into SQL
Example metrics: revenue per country, churn rate …
One example
in dbt:
Precision & consistency
Many people, many teams but only one way to define revenue
Avoid repeating work or copy-paste, and inconsistency can arise
Bridging the gap
Standardisation: feature store for ML training and serving
Semantic layers: feature store in the BI world, output for analytics
Doing it once and correctly
get reuse and consistency