Keynote Interview
summary
Describes how Shopify moves from one model to hundreds by embedding data scientists in product teams, building an in‑house Python and Spark platform since 2014, and enforcing unit tests, model versioning, shadowing and human‑in‑the‑loop monitoring. Gives concrete products: fraud detection in checkout (first production model), a forecasting service used across teams, a fulfillment network with forecasting and allocation models, and Shopify Capital that funds merchants (nearly $250M Shopify Capital in Q3). Emphasizes shipping simple models first and encoding best practices into the platform.
AI-written summary · about
chapters (28)
- 0:00Intro
- 0:25How we met
- 2:20Impact of COVID on Shopify
- 4:44Solmazs role at Shopify
- 5:33Use cases for machine learning
- 7:35Role of platforms at Shopify
- 11:16How Shopify supports the business
- 13:19Live QA
- 13:42How has Shopify evolved
- 15:36What happens when a team isnt focused on impact
- 17:55Successes at Shopify
- 19:33Building a team
- 21:15Platform team
- 22:54Managing conflicts
- 24:45Shipping early and often
- 26:40Lowhanging fruit
- 27:38Model evolution
- 29:16Early experimentation
- 30:55Rituals
- 32:29Tech difficulties
- 33:16Data science culture
- 35:59Model testing
- 38:25Covid impact
- 39:42Automated Retraining
- 41:13Privacy Fairness Bias
- 42:48Culture and Process
- 44:30Embedded vs Centralized
- 47:57Data and Machine Learning
