Extract, transform, load: an ETL is a common operation in data processing systems that reads data from sources, massages it, and stores it in another system. For example, an ETL job might read web server logs and a customer database to fill up a table of user sessions.
Almost every analytics setup runs on one. Your data starts scattered across a production database, a payments processor, a CRM, and a support tool; a pipeline is what brings it together somewhere you can query it.
The three steps
- Extract. Pull raw data from each source: a database read, a nightly CSV drop, or — most often now — a paginated call to a vendor’s API returning JSON.
- Transform. Clean and reshape it. Cast data types, standardize currencies and timestamps, deduplicate, rename columns to something consistent, and join related records into the shape your analysts expect.
- Load. Write the result into the destination, usually a data warehouse or data lake.
Pipelines are typically modeled as a DAG, where each step declares what it depends on, so the orchestrator knows what can run in parallel and what has to wait.
ETL vs. ELT
The classic order is extract, transform, load — transform the data on the way in, so only clean, modeled tables land in the warehouse. That made sense when warehouse storage and compute were expensive: you didn’t want to pay to store raw data you’d never query.
Cloud warehouses inverted the economics, and ELT became the default. You extract and load raw data first, then transform it in the warehouse itself with SQL:
CREATE TABLE analytics.user_sessions AS
SELECT
user_id,
min(event_time) AS session_start,
count(*) AS event_count
FROM raw.web_events
GROUP BY user_id, session_id
The trade-offs:
- ETL keeps the warehouse tidy and can strip sensitive fields before they’re ever stored — useful when compliance rules limit what you may retain.
- ELT keeps the raw data around, so when a transform turns out to be wrong you can rebuild from source instead of re-extracting everything. It also lets analysts, not just data engineers, own the transform layer.
Most teams end up doing both: light cleanup in flight, heavy modeling in the warehouse.
Reverse ETL
A reverse ETL runs the pipeline backwards, pushing modeled data from the warehouse back into operational tools — a health score into your CRM, a churn risk flag into support software — so the people doing the work see it where they already are.
Getting started
If you’re standing up your first pipeline, our guide to building a data pipeline walks through the moving parts, and the individual integration guides cover how to pull data out of specific tools.
Key article
Related terms
Further reading
Put it to work
- Build a data pipeline — Integration
- Stripe — Integration
- Revenue analytics — Overview
