ELT
Aussi : Extract, Load, Transform
ELT (Extract, Load, Transform) is a data integration pattern where raw data is loaded into a target system first, then transformed inside it using the platform's compute power.
What it is
ELT stands for Extract, Load, Transform. It is a data integration approach where data is pulled from source systems, loaded in its raw form into a target platform (typically a cloud data warehouse or lakehouse), and only then transformed into clean, modeled datasets. This reverses the order of the older ETL pattern, where transformation happens before loading.
The shift to ELT was driven by the rise of scalable, low cost cloud storage and powerful in warehouse compute engines (such as columnar query engines). When transformation can run cheaply and quickly inside the destination, there is little reason to do it beforehand in a separate processing layer.
Why it matters
- Speed of ingestion: Raw data lands fast because loading is decoupled from transformation logic.
- Flexibility: Keeping raw data means you can re-model it later without re-extracting from sources.
- Scalability: Transformation leverages the elastic compute of modern warehouses instead of a fixed ETL server.
- Auditability: The original raw data is preserved, which helps with debugging, reprocessing, and compliance.
- Separation of roles: Engineers manage ingestion; analysts and analytics engineers own transformations using SQL based tools.
For a Chief Data Officer, ELT supports a governed, single source of truth while letting teams iterate on business logic without bottlenecking on a central pipeline team.
How it is used in practice
A typical ELT stack combines:
1. Extract and Load tools that connect to APIs, databases, and files, replicating data into the warehouse.
2. A cloud warehouse or lakehouse as the central store.
3. A transformation framework (commonly SQL based) that builds layered, tested, documented models.
Transformations are usually organized in layers: raw, staging (cleaned), and marts (business ready tables for reporting and AI features).
Concrete example
A marketing team wants a unified view of campaign performance:
- Extract: Pull raw rows from an ad platform API, a CRM, and web analytics.
- Load: Drop all three raw datasets, unmodified, into the warehouse.
- Transform: Run SQL models that standardize date formats, deduplicate leads, join spend to revenue, and produce a `campaign_roi` table.
If a new metric is needed next quarter, analysts simply add a model on top of the raw data already loaded, with no new extraction required.