The Modern Data Stack — a Map, not a Religion

Five boxes you will always recognise

Every data platform — whether you call it ETL, ELT, lakehouse, or warehouse — reduces to five boxes:

1. Ingest — Moving raw data from sources. Tools: Fivetran, Airbyte, Debezium, Kafka.
2. Store — Landing the data. Object storage (S3/GCS) for raw files, or warehouses (Snowflake/BigQuery) for structured analytics.
3. Transform — Cleaning and joining. Converting raw forms into clean SQL tables. Tools: dbt, Spark, SQL pipelines.
4. Serve — Delivering the insights out of the warehouse. Tools: BI dashboards (Looker/Preset), reverse-ETL, or API endpoints.
5. Observe — Monitoring the pipes. Tools: Monte Carlo, Soda, OpenLineage for schema drift and freshness alerts.

Use Case Example (E-commerce): You use Airbyte (Ingest) to pull daily Shopify sales strings into Snowflake (Store). You run a dbt (Transform) job to clean the data and calculate 'Customer Lifetime Value'. Finally, the Marketing team logs into Looker (Serve) to see if their recent ad campaign brought in high-value users. If Airbyte fails overnight, your Soda (Observe) tests alert you via Slack before the Marketing team wakes up!

The modern data stack is a restaurant kitchen. Ingest is the delivery dock where raw produce arrives from suppliers. Store is the walk-in fridge and pantry. Transform is the prep line where raw ingredients become clean, chopped, ready-to-cook components. Serve is the pass where finished dishes go out to diners. Observe is the health inspector and the order tickets — checking that nothing is spoiled and every plate left on time. Swap a supplier (Fivetran→Airbyte) or a stove (Spark→dbt) and the stations stay exactly the same. That's why the map outlives every tool that fills it.

The five boxes, visualised

Every modern platform fits this skeleton. The arrows are data flow; the dashed arrow is metadata flow (lineage, quality, freshness) — a property of mature platforms only.