Data Engineering — Fri Feb 13

← Home | ← data-engineering

ETL vs ELT: Same Letters. Different Eras.

Fri Feb 13
#Data-Engineering #Pipeline

Why I’m writing this

When I started in data, nothing was elastic.

Storage had limits. Compute had limits. And if you broke the warehouse, you were going to hear about it.

So we were careful.

We shaped data before it earned the right to live in production.

That discipline created ETL.

Then the cloud happened.

And the order flipped.


ETL – Extract → Transform → Load

ETL means:

  1. Extract from source systems
  2. Transform in staging
  3. Load only clean, structured data

Think traditional enterprise stacks:

  • SQL Server
  • Oracle Database
  • SQL Server Integration Services
  • Informatica PowerCenter

The warehouse was protected.

You didn’t dump raw logs in there. You didn’t “figure it out later.” You shaped it first.

Very DBA energy.

It was built for a world where:

  • Storage was expensive
  • Compute was limited
  • Governance meant strict control

The mindset:

Clean it before it enters the building.


ELT – Extract → Load → Transform

ELT flips the order:

  1. Extract
  2. Load raw
  3. Transform inside the platform

Think modern platforms:

  • Snowflake
  • Databricks
  • Amazon Redshift
  • BigQuery

Now the warehouse (or lakehouse) is the transformation engine.

You land raw data.

Bronze → Silver → Gold.

You keep the messy stuff — because you might need it later.

This only works because:

  • Storage became cheap
  • Compute became elastic
  • Reprocessing became viable

The mindset:

Store first. Refine later.

Very cloud-native energy.


What Actually Changed?

Not the letters.

The economics.

ETL WorldELT World
Hardware constrainedElastic compute
Schema-firstData-first
Protect the warehouseUse the warehouse
Transform outsideTransform inside
Risk-averseIteration-friendly

ETL was discipline born from constraint.

ELT is flexibility born from scale.


Which Should You Use?

It depends on:

  • Regulatory pressure
  • Data volume
  • Cost model
  • Team maturity
  • Failure tolerance

If you’re running tightly governed legacy workloads, ETL still makes sense.

If you’re building modern platforms in AWS or Databricks, ELT is often the natural pattern.

This isn’t religion.

It’s architecture.


The Data & Grit Take

Good engineers don’t argue acronyms.

They understand trade-offs.

ETL wasn’t primitive. ELT isn’t revolutionary.

They are reflections of the infrastructure available at the time.

The job isn’t to follow fashion.

It’s to design systems that survive reality.


Gareth Winterman