GLOSSARY

ETL / ELT

ETL / ELT — definition, when each pattern fits, and how NUUN Digital builds pipelines.

Last updated:

Quick answer
ETL stands for extract, transform, load — the process of pulling data from source systems, transforming it to a target schema, and loading it into a destination, typically a data warehouse. ELT is the modern variant: load first, transform in-warehouse using SQL and tools like dbt. ELT is the dominant pattern in cloud-native stacks because cloud warehouses make in-warehouse transformation cheap and fast.

WHAT IT IS

ELT has become the dominant pattern in cloud warehouses because compute is cheap, storage is cheap, and modern engines can transform at scale inside the warehouse (dbt on Snowflake, BigQuery, Databricks). ETL still applies where source schemas must be masked before landing, where network egress is expensive, or where strict compliance forbids raw landing.

HOW IT WORKS

A production pipeline — ETL or ELT — needs observability (lineage, run status, data volume), idempotency (safe to rerun), schema evolution handling, and alerting. Tools include Fivetran, Airbyte, Stitch, and native CDC; transformation is typically owned by dbt, dataform, or SQLMesh.

WHEN TO USE

Choose ELT when the target is a modern cloud warehouse. Choose ETL when governance, latency, or compliance requires transformation before load.

RELATED

SOURCES

Related questions.

What is ETL?
ETL stands for extract, transform, load. It is the process of pulling data from source systems, transforming it into a target schema (cleaning, conforming, deduplicating, aggregating), and loading it into a destination — typically a data warehouse. ELT is the modern variant: load first, transform in-warehouse.
What is the difference between ETL and ELT?
In ETL, transformation happens before loading, typically in a dedicated processing layer. In ELT, raw data lands in the warehouse first and transformation happens there, using SQL and tools like dbt. ELT is the dominant pattern in modern cloud stacks because cloud warehouses make in-warehouse transformation cheap and fast.
What tools are used for ETL/ELT today?
Fivetran, Airbyte, Stitch, and Hevo dominate managed ingestion. dbt is the de facto tool for in-warehouse transformation. Orchestration sits in Airflow, Dagster, or Prefect. Reverse ETL (sending warehouse data back to operational systems) is handled by Hightouch or Census.
How often should ETL run?
Depends on the freshness requirement. Most BI workloads run hourly or daily. Operational use cases (marketing activation, fraud detection) often need minute-level or streaming pipelines. Pay for the freshness the business actually uses, not the freshness it aspires to.
How does NUUN Digital build pipelines?
We default to managed ELT (Fivetran or Airbyte) into a cloud warehouse, transform with dbt, orchestrate with Airflow or Dagster, and instrument data quality checks in the pipeline itself. We avoid custom pipelines unless a specific workload demands it — custom code is a liability masquerading as flexibility.

Need this term in action?

Book a call