AI glossary · Basic · Data engineering
ETL·ELT
ETL and ELT are ways of pulling data scattered across many places (Extract), reshaping it into a usable form (Transform) and moving it into one place (Load). If the transformation happens before loading, it is called ETL; if it happens after loading, it is called ELT.
About this term
- Level
- Basic · Technical terms you often meet in AI news
- Category
- Data, ops & platforms
- Fields
- Data engineering
- Added
- 2026-10-09
How it works
First, the needed data is pulled from where it lives, such as a member DB, order system, web logs or external services (extract). Next, it is shaped for analysis, for example by unifying date formats, cleaning up missing values and duplicates, and joining multiple tables (transform). Finally, it is loaded into a store such as a data warehouse or data lake (load). ETL finishes the transformation on a separate processing server and loads only the clean results, while ELT loads the raw data into the store first and transforms it using the store's own computing power.
The difference between ETL and ELT
ETLExtract → Transform → Load
- Cleaned in advance on a separate server
- Only clean results are stored
- Better when storage is expensive
ELTExtract → Load → Transform
- Raw data is loaded into the store first
- Cleaned using the store's computing power
- Easy to transform again later
공통 Both gather scattered data in one place in a form ready for analysis
Why it matters
Whether it is an AI model or an internal dashboard, it ultimately needs "organized data" to work properly. Raw data differs in format and naming from system to system, so combining it as-is leads to wrong numbers or makes analysis impossible. ETL and ELT let this cleanup run automatically and repeatedly in a set order instead of being done by hand each time. That is why they are considered the most basic backbone of a data pipeline.
Things to know
In the past, storage and compute were expensive, so ETL, which cleans data in advance and loads only what is needed, was the norm. As stores like cloud data warehouses made large-scale computation inside the store cheap and fast, ELT, which loads raw data first, became widely used. Because ELT keeps the raw data, it is easy to transform it differently later, but unorganized data can pile up and become hard to manage. Data that must not go into the store as-is, such as personal information, still needs to be filtered out before loading, even with ELT.
Example
A typical ETL/ELT job is an online shopping site pulling data from its order system, member DB and ad click logs every night, gathering it in a data warehouse, and shaping it into a "per-customer purchase summary table" used the next morning for the sales dashboard and for training the recommendation model. Building a search index for RAG from internal documents follows the same thinking: documents are pulled, split into chunks and cleaned, and loaded into a vector DB.