Data Integration
Data integration is the work of getting data out of the systems where it is created (application databases, SaaS tools, files, APIs) and into the systems where it is used (a warehouse, a lake, a search index, a CRM), in a shape those systems can use. Almost every data platform is built from four patterns:
| Pattern | Direction | Where transformation happens | Freshness | Typical tools |
|---|---|---|---|---|
| ETL | Sources → warehouse | In a separate engine, before loading | Hours (batch) | Spark, Informatica, SSIS, custom Python |
| ELT | Sources → warehouse | Inside the warehouse, after loading raw data | Minutes to hours | Fivetran / Airbyte + dbt |
| CDC | Database → anywhere | Usually downstream | Seconds | Debezium, Kafka Connect, cloud DMS |
| Reverse ETL | Warehouse → business tools | In the warehouse, before syncing | Minutes to hours | Hightouch, Census, RudderStack |
They are not rivals. A typical 2026 stack uses CDC or an ELT connector to land raw data, dbt (ELT) to model it, and reverse ETL to push the results into the tools sales and marketing use. Classic ETL survives where data must be cleaned or masked before it is allowed into the warehouse.
What’s in this section
- ETL: extract, transform in a staging engine, then load clean tables.
- ELT: load raw data first, transform it with SQL inside the warehouse.
- CDC: stream every insert, update and delete from a database’s log.
- Reverse ETL: sync modeled warehouse data back into operational tools.