Noor sat alone at 11:47 p.m. Tomorrow's QBR: the CEO had texted “One slide. An AI initiative that moved a needle — messy details. No hype.” If she glossed over failures, she'd lose the room.
Netflix — $1 Billion Retention Machine
Lila Patel, former Personalization PM. Mandate: every user finds something in under 60 seconds. 80% of viewing already from recs.
Trinity Breakdown
▸ Data: Billions of implicit signals — pause duration, rewind, device, time of day.
▸ Models: Matrix factorization → deep learning ensembles. Non-deterministic by design.
▸ UX: “Continue Watching” rows felt psychic. Low-confidence recs buried.
One A/B swapped “Top Picks” with “Trending Now.” Retention tanked. Novelty kills habit.
Tesla — Shadow Mode
Jason Chen, ex-Autopilot Product. 300M+ miles of fleet data. Every car became a sensor. Shadow mode: neural net predicted, human acted, every mismatch = labeled training data.
▸ Data: Real-world, always-on, edge-case rich.
▸ Models: End-to-end neural nets improving weekly. Drift = feature.
▸ UX: “Disengage” button sacred. Every intervention instrumented.
“We built the world's largest real-time driving university.”— Jason Chen