This appendix summarizes the platform patterns discussed in Chapters 6, 7, 9, and 22, with the conditions under which each is most appropriate and the principal risks to manage.
| Pattern | Description | Use when | Watch for |
|---|---|---|---|
| Centralized enterprise platform | One platform and team provide common data capabilities for all domains | Early maturity; strong consistency needs; limited domain capacity | Bottlenecks; distance from domain knowledge |
| Federated platform | Domains own data and products on shared services and standards | Many domains with distinct needs and capable teams | Incompatible definitions without strong standards |
| Hub-and-spoke | Central hub for shared infrastructure and governance; domain spokes for specialized needs | Need for autonomy without uncontrolled sprawl | Hub becoming a hidden bottleneck |
| Lakehouse-centric | Open object storage with managed tables serves BI, data science, and AI | Diverse workloads on shared data; desire to reduce copies | Treating the lakehouse as a substitute for information architecture |
| Warehouse plus lake | Governed warehouse for reporting; lake for raw and exploratory data | Established warehouse with strong reporting value | Duplicated pipelines and inconsistent definitions |
| Event-driven data architecture | Business events distribute information to many consumers | Timely operational intelligence; many independent consumers | Event semantics, ordering, replay, and schema evolution |
| Separation of storage and compute | Durable storage with workload-specific compute | Variable or mixed workloads; cost optimization | Data movement, caching, and concurrency effects |
| Hybrid cloud | On-premises and cloud environments operate together | Legacy, regulatory, or latency constraints | Excessive data movement and inconsistent controls |
| Intentional multi-cloud | Workloads distributed across providers for clear reasons | Regulatory, geographic, resilience, or capability needs | Multiplied skills, security models, and operating costs |
| Medallion-style refinement | Raw, standardized, curated, and published stages | Clear lifecycle and responsibility boundaries | Treating stages as mandatory physical zones for every case |
Selection questions. What workloads and latency are required? What data types are involved? What governance applies? Where must data reside? What does the operating model support? What are the cost and portability implications? Which decisions are hard to reverse?