Scaling Data Stacks: How To Build a Great Data Model
summary
Claims that starting from business value and building a Domain Modelling layer produces a stable static target that survives changing sources. Uses dbt plus ingestion tools (Fivetran/Stitch/Airbyte/Snowplow) and warehouses like Snowflake, with Looker/Lightdash/Metabase for presentation. Cites 85% failure rate for data projects and six years with ~60 clients as motivation. Shows a thin, well‑tested central domain layer and skinny downstream marts that prevent 85,000‑model sprawl and let downstream reports keep working when upstream systems change.
AI-written summary · about
chapters (20)
- 0:00Introduction
- 1:03Why data projects fail
- 3:10Data team roles
- 5:58Data modeling philosophy
- 9:48Modern stack limits
- 11:56Static targets and models
- 13:39Separation of concerns
- 16:30History of modeling
- 18:14Entity vs dimensional
- 21:45Startup modeling mistakes
- 22:06Domain-aligned design
- 24:36Conceptual vs logical models
- 28:06Models to physical structures
- 30:55Insulate models from sources
- 32:23Thin-waist architecture
- 33:28Domain modeling challenges
- 36:59DBT best practices
- 39:52Summary and best practices
- 40:14Resources
- 40:56Q&A
related talks
41mModern Data Stack: Optimising and Scaling Data in a Tech Company
41mModern Data Architectures – What They Are and How to Implement?
42mData Modelling: Navigating the Data Seas with a Data Vault Compass
34mDimensional Data Modeling in the Modern Era
31mData Unleashed: A Developer's Perspective on Navigating the Architecture Maze
42mBuilding Tomorrow's Warehouse Today: Modern Data Warehouse Modelling Insights
56mData Unleashed: A Developer's Perspective on Navigating the Architecture Maze
18mDemystifying Data Modeling