Architecture slides show ingest, lake, warehouse, BI, reverse ETL. Team of three spends weekends upgrading components. SME data stack principles: few components (business DB plus controlled warehouse—no extra layer if not needed), strong definitions (metrics have owners), exportable (not tool-locked), auditable (who viewed and changed what). Fashionable stacks solve scale; SMEs first have definition and trust problems.
More components mean definitions fork in the gaps. After forking, prettier dashboards are just more truths.
Four Acceptance Tests
Few: default no more than three consumption layers. Strong definitions: core metrics in a dictionary, calculations reproducible. Export: standard tables to CSV or DB—migration without begging vendors. Audit: authorization and query logs retained. If any one fails, do not add components.
- Reuse business system fact tables—avoid parallel shadow databases.
- Personal spreadsheets can analyze—they cannot be official definitions.
- Register models and retirements—avoid "who ran this number on a laptop."
Intelligence Builds on These Four
Assistants and forecasts eat the same definitions. The XYN digital intelligence system keeps business facts, metrics, and audit in one manageable layer—a stack SMEs can afford. The stack can grow later; definitions cannot wait. Fixing definitions usually costs more than the components you skipped.
List running components and official metrics. Misaligned? Subtract first. Add when scale truly arrives.
