Netflix · AI explanations · 2021 to present
Personalizing Netflix
Recommendations
Adapting how a title is explained to who you are. I led the design of the first AI-written words Netflix put in front of members, and the grading system that made them safe to ship.
157,000 explanations a day · 96 to 100% agreement with human reviewers · 0 rollbacks
At a glance
- My role
- Product Design Lead for personalization and explanations, Netflix member experience.
- Scope
- The first generative AI words Netflix put in front of members, the grading system that decides which ones ship, and a three year north star for one recommendations system.
- Team
- I lead design with a content designer, a PM, engineering and data science. For the north star I pulled together a tiger team: a lead engineer, a data scientist, a content designer and a PM.
- How we split it
- I designed the content principles, the three checks and the rule for what happens on a fail. Data science built the models, and we tuned the rules together.
- When and where
- 2021 to present, on TV, mobile and web: detail pages, rows, billboards and post play. AI explanations launched in May 2025.
- The one thing
- I replaced a match score nobody believed with AI-written reasons a title is for you, and designed the grading system that lets them run at about 157,000 a day with zero rollbacks.
Context
The Netflix evidence system in thirty seconds
If you have never worked on Netflix, here is the screen this whole story happens on. Point at each dot.
A few hundred million
members, each with a different home screen
About 40
rows on one home screen, each one ranked for you
About 2 minutes
before a member gives up and watches something familiar
- 1 · The titleThe art and the name.
- 2 · Metadata barNew, 2021, rated R, one season, Dolby Vision, 5.1. Facts about the title, not reasons to watch it.
- 3 · SynopsisWhat the story is about, the same for every member. Why this title is for you never made it to the screen. That gap is this case study.
- 4 · TriviaAn evidence card: a true line that helps you decide, like who stars in it. Mine to design across the product.
- 5 · A rowOne of about forty, each picked and ordered for you by the recommendation system.
- 6 · A badgeTop 10, New Episodes. Small signals that compete for the same few pixels.
Enter the password to keep reading
Netflix work under NDA. The password is in my application.
