Kimball vs Inmon: Choosing a Data Warehouse Architecture
A practical comparison of Bill Inmon's top-down enterprise approach versus Ralph Kimball's bottom-up dimensional modeling for building data warehouses.
When building a data warehouse, two dominant philosophies emerge: Bill Inmon’s top-down enterprise approach and Ralph Kimball’s bottom-up dimensional modeling. Picking the wrong one costs months of rework.
The Two Philosophies
Bill Inmon: Top-Down Enterprise Approach
Inmon recommends starting with a big centralized enterprise data warehouse where all available data from transaction systems are consolidated into:
- Subject-oriented
- Integrated
- Time-variant
- Non-volatile
Data marts are then built for the analytic needs of individual departments.
Ralph Kimball: Bottom-Up Dimensional Approach
Kimball recommends starting with mission-critical data marts that serve immediate analytic needs of departments. These data marts are then integrated for data consistency through a “bus architecture.” The key tool here is the dimensional model.
Decision Framework: When to Choose Each
| Characteristic | Favours Kimball | Favours Inmon |
|---|---|---|
| Business decision support | Tactical | Strategic |
| Data integration requirements | Individual business requirements | Enterprise-wide integration |
| Structure of data | KPIs, business performance measures, scorecards | Data meeting multiple/varied needs |
| Source system stability | Source systems are quite stable | High rate of change |
| Team composition | Small team of generalists | Bigger team of specialists |
| Time constraints | Urgent need for first data warehouse | Longer time allowed |
| Cost to build | Low start-up cost | High start-up costs |
Key Takeaways
- Start with business needs, not technical preferences - If your org needs quick tactical insights, Kimball’s approach delivers faster
- Consider your team - Kimball works better with smaller generalist teams
- Plan for change - If source systems change frequently, Inmon’s normalized approach handles this better
- Cost vs. time tradeoff - Kimball is faster to first value; Inmon is more comprehensive long-term
My Experience
Working at Disney Streaming, we saw this tension play out. Marketing, DPA, and EPA were semi-decentralized data marts (Kimball-style), but this caused problems when we needed enterprise-wide consistency for subscription data across Disney+, ESPN+, and Hulu.
The lesson: hybrid approaches often work best. Start with Kimball for speed, but plan for eventual integration.
Sources: Kimball Group, zentut.com