Your data platform mirrors your org
Conway's Law, applied to data
1) Centralised data team
→ One warehouse. One semantic layer.
→ Long backlog. High coupling.
→ Business builds spreadsheets “temporarily”.
→ Data people context-switch across every team.
Bottleneck: competing priorities for analytics.
2) Fully embedded, centrally managed
→ Central warehouse. Distributed analytics.
→ Bloated data models to satisfy everyone.
→ Fields added “just in case”. Nothing ever removed.
Bottleneck: changes break things for everyone else.
3) Fully siloed teams
→ Separate warehouses. Separate definitions.
→ Teams answer the same question differently.
→ Nobody reconciles. They just stop comparing.
Bottleneck: data is not trusted at the top.
4) Platform + embedded (data as product)
→ Shared foundations. Analytics treated as a first-class need.
→ Clear ownership of datasets. Explicit consumers.
→ Works only when analytics isn't an afterthought.
Works very well...
Bottleneck: ...but takes a long time to build.
5) Tool-driven architecture
→ “This tool will fix everything.”
→ Everything is half-migrated.
→ Old systems linger “until confidence is higher”.
Bottleneck: turnover. Nobody dares to actually fix things.
Data architecture usually just happens.