Stop Arguing About Numbers: How Zalando Scales Governance With Unity Catalog and Metric Views

Fabian Halkivaha, Mukrram Ur Rahman17 Jun 2026

Problem

Zalando serves more than 50 million customers in Europe through thousands of microservices. For years domain teams owned their data products, which sped up delivery but left hundreds of teams sharing data in unclear ways and defining metrics in separate BI tools. Metrics were hard to find and hard to trust, and teams quoted different numbers for the same KPI. Centralizing everything would have created a bottleneck.

Example from the article: the marketing dashboard showed a different net revenue than the finance report.

Solution in brief

Zalando splits data into private team catalogs and one shared catalog with governed dynamic views, then defines every business metric once as a Databricks metric view in YAML. The metric views live in a central GitHub repository, pass automated and LLM review, and are deployed to the shared catalog only. A two-level hierarchy (broad foundation views, lighter business-local views) lets teams localize metrics without redefining them.

Questions this case answers
  • How to give autonomous teams one definition of each business metric while sharing data with governed access control across the organization
  • How to keep metric definitions in version control with review and automated checks instead of inside presentation tools
  • How to make every team and dashboard use one definition of each business metric
  • How to keep metric definitions in Git with review and CI instead of locking them in BI tools

Tools and methods

  • Unity Catalog — identity-based access control, lineage and auditing
  • Dynamic views — the only way to read across teams; GDPR rules sit inside the view
  • Metric views (YAML) — metrics and dimensions defined once, queried from SQL, dashboards and Genie
  • Kimball dimensional model (Zalando Commons) — facts, conformed dimensions, surrogate keys
  • GitHub CI/CD with an LLM review bot — validation, review and per-PR test environments
Continue reading — it's free

Google or a one-time email code. No password.