Looker

Introduction to Looker

In the mid-2010s, self-service BI tools like Tableau and Power BI empowered business analysts to build dashboards rapidly. However, this empowerment led to a massive enterprise crisis known as Metrics Chaos.

Because every analyst could download their own data and write their own math inside their own dashboard, the company lost its single source of truth. The Marketing dashboard said revenue was $10M, while the Finance dashboard said revenue was $8M.

Looker (acquired by Google Cloud in 2019) was designed specifically to solve Metrics Chaos. Looker is a cloud-based enterprise BI platform that operates on a completely different architectural philosophy than its competitors: it fundamentally separates the definition of the metrics from the visualization of the metrics.

The Architecture: The Semantic Layer (LookML)

Looker’s defining feature is its proprietary modeling language: LookML.

Instead of analysts writing SQL directly inside a dashboard, Looker forces the organization to define all of its business logic centrally using code. A data engineer uses LookML (which looks like a simplified version of YAML/SQL) to define the relationships between tables and the exact mathematical formula for every metric.

• LookML Definition: measure: total_revenue { type: sum, sql: ${TABLE}.price * ${TABLE}.quantity ;; }

Once this metric is defined in LookML, it is locked. The code is committed to Git, heavily version-controlled, and reviewed by senior engineers.

When a non-technical business user opens the Looker UI to build a dashboard, they simply drag the total_revenue field onto the canvas. They cannot change the math. Looker guarantees that every single dashboard in the company using the total_revenue field is executing the exact same mathematical formula.

The 100% In-Database Philosophy

Looker was the first major BI tool to completely abandon the concept of an “Extract” or a proprietary in-memory engine. Looker does not store your data.

Looker operates strictly as a SQL generator. When a user clicks a filter on a Looker dashboard, Looker reads the LookML definitions, dynamically generates an incredibly complex, highly optimized SQL query, and pushes it down to the underlying Data Warehouse (like Google BigQuery, Snowflake, or an Iceberg Lakehouse powered by Dremio).

This architecture has two massive advantages:

1. No Data Silos: Because Looker never extracts data, the organization’s data remains perfectly centralized and governed in the main Data Warehouse.
2. Infinite Scalability: Looker’s speed is entirely dependent on the underlying database. If you connect Looker to a 100-node BigQuery cluster, Looker can instantaneously query petabytes of data, a feat impossible for tools that try to load data into local RAM.

The Headless BI Movement

Looker’s architectural philosophy pioneered what is now known as Headless BI or the Universal Semantic Layer.

Because the metrics are defined in code (LookML), they don’t just have to be viewed in Looker dashboards. An engineer can use Looker’s API to extract the mathematically perfect total_revenue number and embed it directly into a custom React web application, or feed it into a Python machine learning script. The “Head” (the visualization) is decoupled from the “Body” (the math).

Conclusion

Looker is not a visualization tool; it is a data governance platform disguised as a visualization tool. By forcing organizations to define their business logic in version-controlled code (LookML) and executing 100% of its queries directly against the cloud data warehouse, Looker eradicated Metrics Chaos. It remains the gold standard for organizations that prioritize strict data governance and a single, unassailable source of truth across their entire enterprise.