ETL / ELT

ELT has become the default because warehouses are now cheap and fast. Keeping the raw copy means a business can change its definitions later, such as what counts as a closed deal, and rebuild every report from the original data without extracting it again. ETL made sense when storage was expensive, and it still fits when data must be cleaned or anonymised before it may be stored at all.

Without a routine like this, reporting means exporting files by hand and reconciling them in a spreadsheet. The numbers drift apart, and nobody trusts any of them.

• Start with ELT, and keep the raw layer untouched so any transform can be rebuilt.
• Use a ready-made connector for extract and load. Tools such as Fivetran or Airbyte exist for this, so the team only writes the transforms.
• Land the sources that matter on a schedule: the product database, payments, ad platforms and the CRM.
• Write each business definition once, as a transform that anyone can read, and build reports on top of it.

Both names are three steps in a different order. Extract means pulling data out of a source such as a payment tool, a CRM or an ad platform. Load means putting it into a data warehouse. Transform means cleaning and reshaping it, for example turning payments in three currencies into one revenue figure.

In ETL the transform happens before the load. In ELT the raw data lands first and is transformed inside the warehouse, usually with SQL.

• Transforming in the spreadsheet instead of the warehouse, which brings back the drift the pipeline was meant to remove.
• Copying personal data into the raw layer without a plan for deletion requests.
• Building a pipeline for every tool before knowing which questions need answering.