Making it usable.
WHO THE CONSUMERS ARE
Analysts, querying directly Reporting tools Applications Machine learning systems External partners
WHAT EACH NEEDS
- Analysts: understandable models and documentation
- Tools: stable structures and acceptable performance
- Applications: low latency and reliability
- Models: consistent features and historical accuracy
- Partners: defined interfaces and agreed formats
WHAT TO PROVIDE ANALYSTS
Well-named modelled tables, not raw data.
WHY
Raw data requires knowledge they should not need.
WHAT TO AVOID EXPOSING
The raw layer, generally Intermediate models Anything whose grain is unclear
WHAT REVERSE PIPELINES DO
Send warehouse data back into operational systems.
WHAT THAT SUITS
Enriching customer records, driving campaigns, populating applications.
WHAT IT REQUIRES
Care about volume and rate limits in the destination.
WHAT TO PROVIDE FOR APPLICATIONS
A serving store suited to their access pattern, rather than querying the warehouse directly.
WHY
Warehouses are not built for many small low-latency queries.
WHAT TO AGREE WITH EVERY CONSUMER
Freshness expectations What changes require notice
WHAT TO PUBLISH
Data freshness, visibly.
