Metrics-driven Data Architectures
summary
Argues that metric sprawl causes rework and bad decisions and that treating metrics as first-class, versioned artifacts fixes it. Surveys concrete tooling: dbt for transformations, LookML for metadata, Cube.js for embedded APIs, Airbnb’s Minerva and in-house stacks at Uber/LinkedIn as large-scale precedents, and emergent vendors like Transform (GraphQL metrics API), MetricQL (strong dbt integration), GoodData and Supergrain. Recommends metrics as code, a centralized metrics layer with pre-aggregation and discoverable APIs to ensure consistent, low-latency answers.
AI-written summary · about
chapters (11)
- 0:00Introduction to metrics
- 2:50Why metrics matter
- 4:15Brief history of data platforms
- 7:26Problems with metric sprawl
- 10:33Unified semantic layers
- 12:41Rise of analytics engineering
- 15:45Metrics layer explained
- 18:50Building your own metrics layer
- 20:58Metric stores overview
- 23:46Design choices for metrics
- 25:09Summary and key takeaways
related talks
40mMetrics and Semantics: An evolution in data consumption
27mMetric Store
29mModern Data Stack: New Terminology Made Clear
39mBREAKING YOUR DATA TEAM OUT OF THE SERVICE TRAP
28mStop querying tables: give AI governed metrics instead
54mBehind the Hype - Architecture Trends in Data
50mMasterclass: Building a data stack in 2021
41mModern Data Architectures – What They Are and How to Implement?