Reverse ETL

Reverse ETL sends modeled analytical data to operational applications. For example, an owned customer segment is computed in a warehouse and synchronized to an email service. Reverse describes the direction relative to conventional analytical ingestion; it does not undo earlier transformations.

Match source entities to destination records using stable keys, and agree which system owns each field. Define insertion, update, removal, and conflict behavior. A customer disappearing from a query result could mean lost eligibility or a source failure; it is not enough evidence by itself to delete the destination record.

A failed request may have been partly applied. Track per-record outcomes and retry only under the destination’s documented semantics. Reapplying the same segment value can be idempotent, while sending the same campaign again can create a second message. State which side effects a retry can trigger.

Measure freshness through destination visibility, not just query completion. Rate limits, mapping errors, and partial failures can delay individual records after the job starts. Send only authorized fields and check that the intended destination state was reached.

Reference: Hightouch: data activation concepts. See Designing End-to-End Data Architecture for the worked example.