
A long-running language learning product built around immersive lessons.
Rosetta Stone is cited mainly as a contrast. It offered little speaking early because speech models erred on about 80% of accented speech, and it locked content behind a paywall, which let free Duolingo grow. The lesson: pick a narrow technical wedge and let distribution fund the bigger vision, and keep core features free so users fuel growth.
From Speak's strategy, from this page's notes
Rosetta Stone offered little speaking because speech models failed on accented speech. The speech models available at the time had an error rate of about 80% on accented speech, so Rosetta Stone and Duolingo both offered little speaking practice early on.
Speak treated that gap as its wedge: it started by understanding strongly accented learner speech, which brought distribution and data. That funded compute and better models, with the aim of a full end-to-end tutor over about 10 years. Quote: "we saw that that could be the initial wedge to building a product with more distribution, more data"
Paid, expensive Rosetta Stone left room for free, fun Duolingo to grow. Rosetta Stone locked content behind a paywall. Duolingo grew early by being free and fun against that paid, expensive product, and monetized only the bells and whistles.
Cem Kansu, Duolingo's CPO, warns that paywalling previously free features turns metrics green for 6 months to a year, then stops working, and lets a competitor offer the same thing for free. Many subscription businesses nickel and dime users and burn their growth engine even as revenue keeps rising. Quote: "it does kind of work for 6 months or maybe a year but eventually it stops working"
Like Duolingo, it offered little speaking early on because the speech models available had an error rate of about 80% on accented speech.
Rosetta Stone offered little speaking because speech models failed on accented speech. The speech models available at the time had an error rate of about 80% on accented speech, so Rosetta Stone and Duolingo both offered little speaking practice early on.
Speak treated that gap as its wedge: it started by understanding strongly accented learner speech, which brought distribution and data. That funded compute and better models, with the aim of a full end-to-end tutor over about 10 years. Quote: "we saw that that could be the initial wedge to building a product with more distribution, more data"
It locked content behind a paywall, which let free Duolingo grow early and led Duolingo to monetize only the bells and whistles.
Paid, expensive Rosetta Stone left room for free, fun Duolingo to grow. Rosetta Stone locked content behind a paywall. Duolingo grew early by being free and fun against that paid, expensive product, and monetized only the bells and whistles.
Cem Kansu, Duolingo's CPO, warns that paywalling previously free features turns metrics green for 6 months to a year, then stops working, and lets a competitor offer the same thing for free. Many subscription businesses nickel and dime users and burn their growth engine even as revenue keeps rising. Quote: "it does kind of work for 6 months or maybe a year but eventually it stops working"
EducationA language learning app built around daily streaks and short lessons.
EducationA book summary and self-growth app that sells subscriptions through a long web quiz funnel.
EducationA nonfiction book summary app that turns books into short audio and text reads.
EducationAn AI study platform that turns course materials into practice tests, flashcards and tutoring sessions.
EducationAn AI study app that records and transcribes lectures and creates notes, summaries and flashcards.
EducationA company offering courses on product and growth, now also building AI products like Reforge Insights.
EducationAn AI learning app that teaches beginners practical AI tools through short lessons and challenges.
EducationAn Indian online learning platform offering courses for exam preparation and other subjects.