Extract
Data is read from source systems such as SAP S/4HANA, Dynamics 365, Salesforce, NetSuite, databases and files, with as little load on those systems as possible.
Load
Raw data lands in a cloud warehouse or lake such as Snowflake, Databricks, Microsoft Fabric, Amazon S3 or ADLS. Loading first keeps a full-fidelity copy and decouples ingestion from modeling.
Transform
Transformation then runs where compute is elastic: in the warehouse, or with a Python step during the pipeline for cleansing and enrichment that shouldn't wait.
Why ELT over ETL
Cloud platforms make storage cheap and compute elastic. ELT lets teams re-model historical data without re-extracting it, and keeps pipelines simpler to operate.