ETL

ETL stands for Extract, Transform, Load. It describes the three-step process of moving data from source systems (your app database, SaaS tools, APIs) into a destination system (usually a data warehouse) for analysis. Extract pulls raw data from the source. Transform cleans, restructures, and enriches it. Load writes it to the destination.

Why it matters: modern marketing and growth teams rely on data from dozens of sources. Your email engagement data lives in Mailchimp or HubSpot. Product analytics data is in Mixpanel or Amplitude. Revenue data sits in Stripe. Ad performance is spread across Google Ads, Meta Ads, and LinkedIn Ads. Without ETL, this data remains siloed, and your team spends hours manually exporting CSVs and building spreadsheets instead of analyzing.

ETL vs. ELT: the traditional ETL approach transforms data before loading it into the warehouse. The modern ELT approach loads raw data first and transforms it inside the warehouse. ELT has become dominant because cloud warehouses (BigQuery, Snowflake) have massive compute power, making in-warehouse transformation faster and more flexible. dbt (data build tool) is the standard for the Transform step in ELT workflows, letting analysts define transformations as version-controlled SQL models.

Tools in the ecosystem: Fivetran and Airbyte are the leading extraction and loading tools, offering pre-built connectors for hundreds of data sources. Stitch (by Talend) is another option. For more custom or complex pipelines, teams use Apache Airflow, Dagster, or Prefect as orchestration tools. On the transformation side, dbt dominates, with dbt Cloud offering a managed environment and dbt Core being open-source.

Common mistakes: building custom ETL scripts when off-the-shelf connectors exist. Maintaining a custom Python script that pulls data from the HubSpot API is a maintenance burden that Fivetran or Airbyte handles out of the box. Another mistake is not monitoring pipeline health. When a source API changes or credentials expire, your data silently goes stale, and downstream dashboards show outdated numbers without anyone noticing.

Practical example: a growth team uses Fivetran to sync Google Analytics, Stripe, and HubSpot data into Snowflake every hour. dbt models join this data into a unified "customer 360" table. A Looker dashboard built on top shows real-time CAC, LTV by channel, and churn risk scores, updating automatically without any manual data wrangling.