Lineage first
Traced upstream sources through Airflow-orchestrated loads to isolate where duplicates compounded.
Rebuilt trust in marketing tag tables — traced lineage upstream, then fixed the load strategy that compounded duplicates downstream.
Problem
Tag-based tables were heavily inflated with duplicates from historical data modeling. Incremental loads kept compounding bad rows until marketing analytics on campaign and canvas tags was untrustworthy.
Approach
Trace lineage from upstream tables to find where duplicates entered, then change the Airflow-orchestrated downstream loads from incremental merge to full overwrite so row counts stay reliable for every consuming team.
Architecture
Marketing data pipeline
Full-overwrite upstream; Airflow downstream with full overwrite on tag tables.
Drag nodes to explore the flow
Why incremental downstream kept inflating duplicates.
Before
Incremental merges on tag tables compounded duplicate rows even when upstream tables only had a handful of bad records.
After
Full overwrite downstream aligned with upstream full-refresh semantics — reliable row counts for marketing and CS analytics.
Solution highlights
Traced upstream sources through Airflow-orchestrated loads to isolate where duplicates compounded.
Tag tables switched from incremental merge to full overwrite, matching upstream refresh semantics.
Upstream vs downstream semantics written down so the fix held for every consuming team.
Reliable row counts for marketing and customer-success analytics.