Building a Data Experimentation Platform Without a Platform Team
summary
Builds a full experimentation stack built twice in five years without a platform team, combining client-side feature toggles, caching and event tracking with a semantic layer (DBT/metricflow/cube) plus SQL/YAML/git driven launches to make experiments reproducible results. Pushes an MVP slice approach: launch, measure impact on business metrics, iterate. Notes buy-in, sponsor and daily reliable pipelines as prerequisites and flags cost tradeoffs — commercial options can run 30–40K/year estimate.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 2:20Why build data platform
- 4:08Key questions before building
- 7:43Data infrastructure needs
- 9:11Building MVP
- 11:16Defining MVP
- 13:41Functional vs reliable layers
- 15:48Engineering platform as product
- 16:10Invent and simplify
- 18:41Reusing tools for efficiency
- 20:09Building components
- 21:55Three pillars for success
- 23:19Final thoughts
- 23:41Q&A: Measuring platform impact
- 27:18Q&A: Build vs buy
related talks
21mWhen a platform isn’t enough: Building an experimentation culture
24mEffective Product Experimentation: Using Data-Driven Techniques for Iteration
43mEngineering Platforms Evolved
43mEverything You Need, Nothing You Don't: How Do You Actually Build a Lean Platform
35mBuilding Data-Intensive Teams
54mDeveloper experience platforms
47m10K Metrics, 500 A/B Experiments and 500M p‑values Every Day — How Is That Possible?
35mBuilding a data analytics platform on strong opinions held loosely