Latest / Startup Stories with Fexingo: Conversations About Founders, Funding, and Building Companies from Zero / How Duolingo Gamified Language Learning to 500 Million Users
Transcript
- Lucas: So there's this moment in Duolingo's history that I think sums up their entire approach. In 2015, they noticed that a huge chunk of users were dropping off after just a few days. The typical response would be to improve the content or add more languages. Instead, the product team asked a different question: what if we just made people feel guilty for quitting? Luna: Guilty? Like, shame them back into learning? Lucas: Exactly. They introduced the streak feature, but more importantly, they started sending push notifications that said things like 'Your owl is sad you missed a day.' And it worked. Within months, day-one retention jumped something like twenty percent. That's the Duolingo playbook in a nutshell — they treat language learning as a behavioural design problem, not a pedagogical one. Luna: And we should mention, if you're enjoying this kind of deep dive into how companies really build, it's worth knowing that listener support is what keeps this show ad-free and independent. If today's conversation moved your thinking forward in some small way, you can say thanks at buy me a coffee dot com slash fexingo. No pressure, just a coffee. Anyway — Lucas, you were saying about the behavioural design? Lucas: Right. So the streak is just one piece. The bigger insight is that Duolingo doesn't just teach you a language — it runs an endless series of micro-experiments on you. Every icon color, every button placement, every notification time is A/B tested. They have something like two hundred experiments running at any given moment. Luna: Two hundred? That's more than most tech companies run in a year. Lucas: It's insane. And the reason they can do that is because they have massive scale — as of early this year, over five hundred million registered users. But critically, the team is still relatively small. Duolingo's core product team is maybe a hundred people. So they're not throwing bodies at problems; they're running tight experiments and letting the data decide. Luna: Give me a concrete example. What's a weird experiment that actually shipped? Lucas: One of my favorites: they tested whether showing a progress bar during a lesson increased completion rates. It did — by about five percent. But then they tested a progress bar that moved slightly faster than actual progress, and that increased completion by another three percent. It's a tiny psychological trick, but across millions of users, that's millions more lessons finished. Luna: So they're basically gaming the dopamine system. Lucas: Totally. And the founder, Luis von Ahn, is very open about this. He's the guy who created CAPTCHA and reCAPTCHA, so he understands gamification and human behavior at a deep level. He's said that Duolingo is essentially 'a slot machine that teaches you Spanish.' He means it half-jokingly, but the mechanics are real — variable rewards, streaks, leaderboards, all of it. Luna: And yet it works. I mean, the science of language acquisition says spaced repetition and regular practice are key. So if a gamified interface gets people to practice daily, that's actually good for learning. Lucas: That's the tension. Critics say Duolingo optimizes for engagement, not proficiency. But von Ahn's counter is: if you don't stay engaged, you learn nothing. A perfect curriculum that people quit after a week is worse than an imperfect one they stick with for a year. Luna: Fair point. But they also have to make money. The freemium model is tricky. Lucas: Right, and that's where the business gets really interesting. Duolingo has about six million paying subscribers for Super Duolingo, their ad-free version. But the real money — about seventy percent of revenue — still comes from ads on the free tier. The freemium model works because the free experience is good enough that most people never upgrade, but the ads are non-intrusive enough that they don't drive users away. Luna: And they just launched Duolingo Max, their ai powered tier. What's behind that? Lucas: That's the big bet for 2026. Duolingo Max uses generative AI to create personalised explanations, role-playing conversations, and instant feedback on writing. It's essentially a tutoring layer on top of the existing curriculum. They started rolling it out late last year, and early data suggests it's boosting retention for advanced learners who used to plateau. Luna: So they're betting AI can solve the 'intermediate plateau' problem. Lucas: Exactly. And they're in a good position to do it because they have so much data on how people learn. Every mistake, every hesitation, every skip — they capture it all. That data lets them train models to predict exactly where a user will struggle next. Luna: But isn't that expensive? Generative AI isn't cheap. Lucas: It is, but they've been smart about it. They don't run a model for every single user action. They cache common responses, and they only use AI for high-value moments — like when a user gets a question wrong for the third time. The cost per interaction is manageable, and they're betting that the improved retention will more than pay for it. Luna: So where does Duolingo go from here? Five hundred million users is a lot, but growth has to slow eventually. Lucas: They're pushing into new verticals. The Duolingo English Test is now accepted by over four thousand universities as an alternative to TOEFL. That's a real revenue stream. They also launched Duolingo Math last year, and they're experimenting with music lessons. The idea is to become a general-purpose learning platform, not just a language app. Luna: But can the same gamification formula work for math? I don't know that people have the same emotional motivation to learn calculus. Lucas: That's the open question. Duolingo Math uses the same streak and leaderboard mechanics, but early reviews are mixed. Some users love it, others say it's too shallow. The company is iterating fast, though — they've already shipped big updates based on user feedback. Luna: Let's go back to the core business for a second. The stock has had a wild ride since their IPO in 2021. What's the market saying about them as of mid-2026? Lucas: The stock is up about forty percent from the IPO price, but it's been volatile. The big concern is that they're trading at a high price to earnings ratio — around sixty times forward earnings. That's expensive for a company that's not growing as fast as it was in 2020 and 2021. But the bulls argue that the AI tier and new verticals will reaccelerate growth. Luna: So it's a bet on execution. Either they nail the expansion, or the multiple contracts. Lucas: Right. And the CEO knows that. Von Ahn has said publicly that they're aiming for a billion users. That's audacious, but if any company can do it, it's the one that turned language learning into a daily habit for half a billion people. Luna: One thing I find fascinating is that they've managed to build a genuinely global user base. It's not just English speakers learning Spanish — it's Vietnamese learning Korean, Arabic learning English, all of it. Lucas: Yeah, they offer over forty languages now, including fictional ones like High Valyrian and Klingon. The Klingon course actually has something like a hundred thousand active learners, which tells you a lot about their audience. Luna: That's bizarre and wonderful. But it also highlights their challenge: most of those courses aren't profitable. They're loss leaders that drive brand awareness. Lucas: Absolutely. The real money comes from the big courses — English, Spanish, French. But the long tail of courses builds the network effect. The more languages they offer, the more users come, and the more data they collect. Luna: Let's talk about the competition. Babbel and Rosetta Stone have been around forever. What makes Duolingo different? Lucas: Price and personality. Duolingo is free with ads. Babbel costs about fifteen dollars a month. And the Duolingo mascot, the green owl, has become a cultural phenomenon. They've turned the app into a meme machine — the owl sends passive-aggressive notifications, users make jokes about it, and that organic virality is worth millions in marketing. Luna: So the owl is essentially a brand asset that drives free distribution. Lucas: Exactly. They've leaned into the 'unhinged owl' persona on social media, especially TikTok, where they have millions of followers. It's a brilliant way to stay top of mind without spending on ads. Luna: Okay, so what's the biggest risk for Duolingo right now? Lucas: I'd say two things. One: the AI arms race. Big tech companies like Google and Microsoft are investing heavily in language learning AI. If Google integrates a tutoring feature into Google Translate or Assistant, that could be a threat. Two: the regulatory environment around data privacy. Duolingo collects a lot of personal data — especially from kids — and if regulations tighten, their business model could take a hit. Luna: Especially in Europe, where GDPR is strict. Lucas: Right. They've already had to adjust their data practices. But so far, they've managed to stay compliant without a major disruption. Luna: Final thought: if you were going to invest in Duolingo, what would you watch for over the next year? Lucas: I'd watch two metrics: daily active user growth and the conversion rate to Super Duolingo and Max. If DAUs are still growing at twenty percent year-over-year and the higher-priced tiers are gaining traction, the bulls are probably right. If growth stalls and the AI bet doesn't pay off, the multiple could compress fast. Luna: So it's a high-conviction bet on the team and their ability to keep innovating. Lucas: Exactly. And given how disciplined they've been about testing and data, I wouldn't bet against them.