ELT (Extract, Load, Transform)

ELT (extract, load, transform) is a data integration approach that reverses the order of the traditional process. Instead of cleaning and shaping the information before storing it, it first extracts the data from the sources, deposits it raw in the destination and only then runs the transformations inside the warehouse itself.

An everyday analogy helps: instead of cooking each ingredient before putting it in the fridge, you store everything raw and cook on demand, following each day's recipe. That change in sequence became viable with cloud data warehouses, which separate storage and compute and scale elastically, something older architectures did not allow.

It is worth drawing a line from the start: ELT is the process, not the place. It should not be confused with the data warehouse (the structured destination) or the data lake (the repository of raw data). ELT describes how data moves and is organized on its way to those destinations, not the destination itself.

ELT emerged as a response to a fundamental shift in data infrastructure. For decades, transforming before loading was a technical necessity: the destination warehouse was expensive and limited, so the data had to arrive already clean and modeled. When cloud warehouses appeared with elastic compute and parallel processing, it no longer made sense to set up an intermediate transformation server. Loading first and transforming inside became faster, more flexible and, often, cheaper.

How it works, step by step

1. Extract: data is collected from the available sources (relational databases, files, SaaS applications or event streams) without processing it yet.
2. Load: the raw data is deposited directly in the cloud data warehouse, without going through a separate transformation server.
3. Transform: the transformations happen inside the destination, with SQL or tools that run on the warehouse, taking advantage of its parallel compute. They can be repeated as many times as needed.

The key difference from the classic approach lies in that third step: since the raw data stays stored, redoing a transformation does not force you to extract again from the source.

Why it matters to a company in Latin America

Many midsize companies in the region have their data scattered across several systems (the point of sale, the e-commerce store, a wholesale spreadsheet, an old ERP) and they need fast answers without setting up heavy infrastructure or a large engineering team. ELT shortens the time until data is available: you load first and transform later, according to the business question of the moment. By keeping the raw data, it lets you redo transformations when needs change, something the ETL approach makes harder because it discards whatever it did not model up front. Organizing that scattered data and making it queryable in natural language is part of what an analytics layer like Metrix solves.

Differences from ETL and related terms

ETL transforms before loading, on a separate server, and is usually preferred when you need to mask or filter sensitive data before it reaches the destination, for example because of compliance requirements. ELT loads first and transforms in the warehouse. They are not rivals: they coexist depending on the case.

Nor should you confuse the process with the destination: the data warehouse is the structured store where the data lands, and the data lake is the repository of raw data. ELT is the road that brings the data there and organizes it, not the building.

Common mistakes

• Believing ELT removes the need to govern and clean data: it only changes when that happens, not the obligation to do it.
• Loading raw sensitive data without access controls in the destination, a risk that ETL mitigated by filtering before loading.
• Assuming ELT performs the same in an on-premise environment without elastic compute: its advantage depends precisely on the cloud's parallel processing.
• Confusing the process (ELT) with the destination (the data warehouse or the data lake) and expecting one to do the other's job.

Applied well, ELT is not a technical fad: it is the natural way to work when the destination warehouse has muscle to spare and the business needs to change the question without rebuilding the pipeline every time.