Looker is the BI tool that decided dashboards should be code. Where Tableau let analysts drag fields onto a canvas and Power BI let them write DAX, Looker said: no, the right move is to define your entire data model — every metric, every join, every dimension — in a typed configuration language called LookML, version-control it in Git, and make every chart in the company derive from that single source of truth.

This was a radical idea in 2012. By 2018, it was the founding principle of an entire category called the semantic layer, and every modern data team that takes governance seriously is doing some version of what Looker invented. Google bought Looker in 2020 for $2.6 billion. Whether the product still leads the category it created is debatable — but the idea won completely.

Origin Story

Looker was founded in 2012 by Lloyd Tabb and Ben Porterfield in Santa Cruz, California. Tabb is a deep database veteran — he was the second engineer at Netscape, an early engineer at Borland, and the CTO of LiveOps. He came to Looker with a specific frustration: every BI tool he had ever seen produced inconsistent answers for the same business question because every analyst recreated the metric calculation from scratch. Two analysts would both query "monthly active users," get different numbers, and waste hours arguing about which one was right.

Tabb's insight: the problem isn't the dashboards. The problem is that the metric is defined in the dashboard. Pull the metric definition out of the visualization layer, put it in code, version-control it, code-review it, test it, and now you have one definition of MAU that every dashboard inherits from.

To make this work, Tabb invented LookML — a YAML-like declarative language for defining models, views, dimensions, measures, joins, and explores. The Looker server compiles LookML into SQL and runs it against your warehouse on demand. There's no extract, no in-memory cube, no separate semantic store. The warehouse is the source of truth; LookML is the lens.

Looker raised about $280M in venture funding from Redpoint, Kleiner Perkins, Goldman Sachs, and others, reaching a peak private valuation of around $1.6B before Google acquired it in February 2020 for $2.6B in cash. The acquisition closed despite a long antitrust review.

What LookML Actually Is

The best way to understand Looker is to understand LookML. Here is the rough mental model:

• A view is a definition of a database table, with the columns you care about and how to interpret them.
• A dimension is a column you can group by ("user country," "order date").
• A measure is an aggregation ("count of orders," "sum of revenue," "distinct count of users").
• An explore is a starting point for querying — a fact view plus the dimensions and other views it can join to.
• A model is a collection of explores, scoped to a database connection.

All of this lives in .lkml files inside a Git repository. Changes go through pull requests. Bad LookML throws compile errors. Good LookML produces a single, governed definition of every metric in the company that every dashboard, every alert, and every API call inherits from.

The aha moment: when a finance analyst and a marketing analyst both ask "what was MRR last month," they both get the same number, because both queries are generated from the same measure: mrr in LookML. This sounds obvious. It is not what happens in Tableau or Power BI without serious effort.

Strengths

• The semantic layer is real and works. Looker's signature feature is also its best feature. For organizations that care about consistent metrics, nothing else comes close — at least not without bolting on an external semantic layer like dbt Semantic Layer or Cube.
• Git-based workflow. LookML lives in Git, with branches, PRs, and code review. This is the only major BI tool with a real software engineering workflow.
• Warehouse-native. Looker doesn't extract data. Every query hits your warehouse live, which means your dashboards always reflect fresh data and you only pay for one copy of the truth.
• API-first. The Looker API and embedded Looker have made it the BI tool of choice for SaaS companies that need to ship dashboards inside their own products.
• Composable. LookML supports inheritance, refinements, and reusable derived tables, which lets large teams build maintainable models.

Weaknesses

• The visualization layer is mid. Looker's charts are functional but visually unimpressive next to Tableau or Sigma. If you're optimizing for executive presentation polish, Looker is not your tool.
• LookML is a real language. Onboarding new analysts takes weeks. Most business users can't and won't learn it, which means analysts become a bottleneck for new metric requests.
• Slow on bad warehouses. Because every query is live SQL, Looker is only as fast as your warehouse. Cheap warehouses + complex LookML = slow dashboards and big bills.
• Pricing is opaque and high. Looker doesn't publish prices. In practice, deployments routinely cost $60K–$500K+ per year depending on user count and platform tier.
• Post-Google product velocity has been slow. This is the big one. See below.

The Google Era: A Tighter, Slower Looker

When Google acquired Looker in February 2020 for $2.6B, the strategic logic was clear: Google Cloud needed an analytics layer that played well with BigQuery. Looker fit perfectly. The deal closed, Looker became part of Google Cloud, and a slow but unmistakable transformation began.

Since the acquisition:

• Looker has been more deeply integrated with Google Cloud. Single sign-on with GCP IAM, billing through GCP, deeper BigQuery optimization, integration with Vertex AI.
• A new product called "Looker Studio" (formerly Google Data Studio) confused the market. Looker Studio is the free, lightweight, Google-Sheets-style dashboarding tool — completely unrelated architecturally to "Looker (original flavor)" with LookML. Many users still don't know which is which.
• Looker Modeler was announced as a way to expose the semantic layer to other BI tools (Tableau, Power BI), positioning Looker as a headless semantic layer rather than a full BI tool. It has been slow to deliver and unclear in positioning.
• Product velocity slowed. Several long-time Looker leaders, including founder Lloyd Tabb (who departed Google in 2021), have left. The community has grown quieter.
• Competitors moved. dbt's Semantic Layer (acquired from Transform in 2023), Cube, and MetricFlow have all positioned themselves as the open, BI-tool-agnostic alternative to LookML. They are eating Looker's intellectual lunch.

The honest read: Looker became a feature of Google Cloud rather than the standalone category leader it once was. The semantic layer concept exploded; Looker the product captured less and less of that growth.

Where Looker Sits in the Data Stack

Looker sits directly on top of the warehouse. It connects to BigQuery (the most common pairing today, especially after the Google acquisition), Snowflake, Redshift, Databricks, Postgres, and others. Unlike Tableau or Power BI, it does not extract data — every query is generated live from LookML and pushed to the warehouse. Increasingly, Looker is paired with dbt for warehouse-side transformations, with dbt models exposed as Looker views.