The Composable CDP Architecture (Definitive Stack)
The composable CDP stack (ELT → Warehouse → dbt → Reverse ETL → Destinations) is now the standard for data warehouse-native activation, but its warehouse round-trip makes it unsuitable for in-session personalization.
Scale
Industry standard for data warehouse-native customer data activation
Before
Purpose-built CDPs with proprietary data storage — vendor lock-in, data replication costs, limited flexibility
After
ELT (Fivetran/Airbyte) → Cloud Warehouse (Snowflake/BigQuery/Databricks) → dbt (profile modeling) → Reverse ETL (Hightouch/Census) → Destinations (ESPs, paid media, CRM)
Key Insight
Composable CDP ≠ real-time personalization. The warehouse round-trip introduces hours of latency — purpose-built CDPs avoid this for in-session use cases. Know the constraint before recommending the architecture.
In a Snowflake Conversation
The composable CDP is excellent for segmentation, audience building, and next-day activation. It's not the right architecture for in-session personalization or millisecond decisioning.
My Read
Practitioner commentary coming soon.
Relevant Conversations