Looker as an Enterprise Governance Surface

LookML in git, content validator, code review — why Looker became the ad-hoc semantic layer of the 2010s.

0/1 done

Overview

The cultural reason Looker won

Tools like Tableau and Power BI treat the data model as an artefact created inside the dashboard. Looker put the model in git: every LookML file is version-controlled, reviewed via PR, and validated automatically.

The platform features that turn this into governance:

  • Content Validator — every dashboard, look and scheduled report references LookML fields by name; if a PR renames a field, Content Validator flags every consumer that would break. This is the pre-flight check the dbt SL and Cube are still catching up on.
  • group_label / view_label / description — labels and docs travel with the field, not with the dashboard, so the catalog is an emergent property of the model.
  • Permission sets + access grants + access filters — RBAC is declared in LookML, not bolted onto each dashboard.
  • CI/CDlookml-validator + Spectacles run as PR checks; merging to main deploys the model.

The downside

Looker is a closed runtime. The model lives in git, but the engine is Google's. Cube and the dbt SL are bets on an open runtime — same modelling discipline, but the execution layer is also yours.

Analogy — building codes for metrics

LookML-in-git is building codes plus a city inspector for analytics. Anyone can nail boards together in an afternoon (a Tableau workbook), but a permitted building has reviewed plans in the public record (LookML in git, merged via PR) and an inspector who — before you knock down a wall — tells you exactly which rooms upstairs are resting on it (Content Validator tracing every dashboard, look and schedule back to the field it depends on). Analytics tools without those workflows are unpermitted construction: perfectly fine to live in, right up until the renovation that quietly collapses the floor above.

Reflect

Whether or not you adopt Looker itself, the practices Looker normalised — model-in-git, PR-reviewed metric definitions, automated validation of consumers against model changes — are the table stakes of any modern semantic-layer rollout. Tools without those workflows regress to the five-revenues problem within a year.

  • Is your semantic layer (whichever tool) actually under code review today?
  • Do you have a Content Validator equivalent — something that fails CI when a renamed metric or dimension would break a downstream dashboard?

Reading in progress · 0 of 1 activity done