Data Flywheels — How Spotify, Uber, and Duolingo Compound AI Value
Episode 3 of the Crafting Product Excellence podcast, hosted by Torsten Feld, titled "Data Flywheels — How Spotify, Uber, and Duolingo Compound AI Value" was published on April 6, 2026 and runs 22 minutes.
April 6, 2026 ·22m · Crafting Product Excellence
Summary
The most powerful AI products don't just use data — they create self-reinforcing loops where more users generate more data, which trains better models, which attract more users. This is the data flywheel, and it's the closest thing to a sustainable competitive moat in AI.In this episode, we go deep on three companies that have built some of the most effective data flywheels in tech — and what AI product managers can learn from each.Companies analyzed:• Spotify — How 600M+ users and billions of daily listening events power a recommendation engine where 75-80% of all listening is algorithmically driven. Why personalization isn't a feature at Spotify — it IS the product. How Discover Weekly, Daily Mix, and AI DJ create different flywheel speeds across discovery and lean-back listening.• Uber — How millions of daily trips across 70+ countries feed real-time pricing, ETA prediction, and route optimization. The evolution from Michelangelo (ML platform) to GenAI 3.0, serving 15M+ predictions per second. Why Uber's shift to an asset-light autonomous vehicle aggregator model (18 AV partners, 28 cities by 2028) is itself a flywheel play.• Duolingo — How 100M+ monthly active users generate the learning data that makes AI tutoring better for everyone. The radical AI-first transformation: replacing contractor content creation with AI (4-5x output, same headcount), rebuilding product tiers around AI capabilities, and navigating the public backlash of workforce changes.For each company, we examine the flywheel mechanics, the cold-start problem, the competitive moat, and the product decisions that accelerate or stall the loop.Part of the Crafting Product Excellence series on AI product management.
Episode Description
The most powerful AI products don't just use data — they create self-reinforcing loops where more users generate more data, which trains better models, which attract more users. This is the data flywheel, and it's the closest thing to a sustainable competitive moat in AI.In this episode, we go deep on three companies that have built some of the most effective data flywheels in tech — and what AI product managers can learn from each.Companies analyzed:• Spotify — How 600M+ users and billions of daily listening events power a recommendation engine where 75-80% of all listening is algorithmically driven. Why personalization isn't a feature at Spotify — it IS the product. How Discover Weekly, Daily Mix, and AI DJ create different flywheel speeds across discovery and lean-back listening.• Uber — How millions of daily trips across 70+ countries feed real-time pricing, ETA prediction, and route optimization. The evolution from Michelangelo (ML platform) to GenAI 3.0, serving 15M+ predictions per second. Why Uber's shift to an asset-light autonomous vehicle aggregator model (18 AV partners, 28 cities by 2028) is itself a flywheel play.• Duolingo — How 100M+ monthly active users generate the learning data that makes AI tutoring better for everyone. The radical AI-first transformation: replacing contractor content creation with AI (4-5x output, same headcount), rebuilding product tiers around AI capabilities, and navigating the public backlash of workforce changes.For each company, we examine the flywheel mechanics, the cold-start problem, the competitive moat, and the product decisions that accelerate or stall the loop.Part of the Crafting Product Excellence series on AI product management.
Similar Episodes
Feb 2, 2026 ·126m
Dec 23, 2025 ·82m