If you work in a tech company — as a PM, analyst, marketer, or business leader — you interact with the modern data stack every time you look at a dashboard, ask for a report, or wonder why a number does not match what you expected. Here is what each layer does, in plain language.
Why every tech professional needs to understand this
Whether you are a PM deciding what events to track, a data analyst trying to understand why a number is wrong, or a business analyst requesting a new report, you will interact with the data stack constantly. Understanding the layers helps you ask better questions, scope data requests more precisely, and debug problems without escalating everything to engineering.
The five layers
Sources are where the raw data lives — production databases, third-party SaaS tools like Stripe and Salesforce, marketing platforms, event tracking systems. This is the origin of every number you will ever see in a dashboard.
Ingestion tools like Fivetran and Airbyte automatically copy data from those sources into the warehouse on a schedule. They handle the plumbing so engineers do not have to write custom connectors for every data source.
The warehouse — Snowflake, BigQuery, Redshift — is a database optimized for analytical queries rather than transactional operations. You write SQL against it. It is designed to scan billions of rows fast, not to process millions of small writes per second like a production database.
Transformation is where raw ingested data becomes clean, joined, analyst-ready tables. dbt (data build tool) is the dominant tool here. It runs SQL transformations that apply business logic — joining the orders table to the customers table, calculating monthly recurring revenue, defining what counts as an active user. This is where your company's definitions of its most important metrics live.
Visualization tools — Looker, Tableau, Mode, Metabase — sit on top of the warehouse and let non-engineers query and visualize data without writing SQL directly. They are the layer most business users actually touch.
The most common breakage points
Source schema changes break ingestion when a SaaS tool renames a column or removes a field. Pipeline delays make real-time data unreliable — most warehouses are not truly real-time. Transformation logic bugs produce wrong numbers in dashboards that look correct. And warehouse cost optimization — throttling query concurrency to save money — can make reports slow or time out during peak usage.
What you should know as a non-engineer
Learn how to read a table schema so you understand what columns exist and what they mean. Learn enough SQL to run a basic count or sum against a table to verify a number you see in a dashboard. And learn how to describe a data request precisely enough that engineering can scope it — including the exact definition of the metric, the grain of the data you need, the time range, and any filters that should apply. Vague requests are the biggest source of wasted time between data consumers and data producers.