Podcast
Designing High-Impact Data Science Teams: Centralized vs Embedded Models, Experimentation & Staffing
Open original DataTalks.Club episode
Designing High-Impact Data Science Teams: Centralized vs Embedded Models, Experimentation & Staffing
Original Episode
Use these links for the canonical episode and media sources.
- Open the original DataTalks.Club podcast page
- Watch on YouTube
- Listen on Spotify
- Listen on Apple Podcasts
Episode Overview
How should you structure a data science organization to maximize product impact: centralized, embedded, or a hybrid of both? In this episode, Lisa Cohen, Director of Data Science at Twitter who leads 70 data scientists and previously led Azure Customer Growth Analytics at Microsoft, walks through practical tradeoffs and implementation patterns for designing high-impact data science orgs.
People
Use these links to connect the episode to guest notes.
Chapter Summary
Use these checkpoints to decide whether to open the source transcript.
- 1:17 - Guest Introduction: Lisa Cohen, Director of Data Science at Twitter
- 1:42 - Career Background: Applied Math, Microsoft telemetry, Azure to Twitter
- 6:27 - Org Models Overview: Centralized vs decentralized data science organization
- 8:34 - Embedding Explained: Reporting lines vs day-to-day integration with feature
- 10:41 - Hybrid Structure: Centralization per division and multiple DS orgs
- 15:26 - Reporting Structure: Embedded teams vs centralized data science reporting
- 18:43 - Team Rhythms & Planning: Cross-functional ceremonies and dependency management
- 21:58 - Cross-Functional Alignment: OKRs and aligning goals across levels
- 24:53 - Twitter’s Approach: Hybrid per-division model for product and ads
- 25:48 - Decentralized Model: Immersive domain context, faster decisions, career tradeoffs
- 29:25 - Centralized Model: Knowledge sharing, consistency, and context-building challenges
- 30:52 - Communicating Insights: Translating metrics for product, engineering, and
- 33:08 - Starting Data Science: Foundations—data pipelines, data quality, and analytics
- 36:49 - Staffing Guidance: Engineers-to-data-scientist ratios and ML partnerships
- 42:19 - Knowledge Sharing & Publication: Research archives, Slack channels, and push
- 46:09 - Product Partnership: Co-ownership with product, engineering, design, and
- 47:20 - Metrics & Experimentation: Defining success metrics, ship criteria, and experiment
- 50:44 - Analytics vs Data Science: Analysts driving dashboards vs ML-heavy DS work
- 52:30 - OKRs & Exploration Time: Using objectives to prioritize and allocate research
- 54:16 - Resolving Conflicts: Data-driven opportunity sizing for prioritization decisions
- 55:48 - Data-Driven Product Innovation: Guiding roadmap decisions with trusted data
- 57:31 - Qualitative Research Collaboration: Bridging user studies with quantitative
- 59:38 - Contact & Resources: Lisa on Twitter, LinkedIn, and Medium