Duolingo | Investment Thesis

The one who will teach you everything.

We are living through another disruptive period made possible by new technologies—namely AI. The last time something comparable happened was when the internet emerged & created countless opportunities, revolutionizing our ways for more efficiency in every sector.

What can seem surprising is that, decades after the internet’s creation & millions of transformative innovations later, some sectors have barely evolved or only slightly because of innovation’s impacts, not by choice.

A prime example of this is education.

I believe many have already seen this video as it made lots of noise when released & accumulated 28M views. It is meant to be entertaining but certainly raised many very valid points & ironically, since its release 8 years ago… nothing has changed.

“If we can customize healthcare, cars & Facebook pages, then it is our duty to do the same with education.”

This is going to be today’s subject. It won’t be about language learning, no, but about learning at large, using new & tailored methods, as this is what Duolingo is building - a global learning platform.

Artificial Intelligence & Education.

As with many other sectors, education will be disrupted by AI, whether our education system follows up or not, and many will turn toward better educational systems if it doesn’t upgrade.

AI models are available at any hour, always patient, capable of explaining a concept or the reasons for a mistake as many times as necessary through different angles. It will never get angry nor frustrated with students & be more knowledgeable than many teachers - without disrespect.

AI will be a better pedagogue than 99% of teachers.

It won’t replace schools, but we will have very efficient educational AI models, capable of helping students in their weak areas. We’re not there, but as infrastructures scale, models get better, and we’ll reach that point. Here’s a really quick example of what Grok can already do.

It is pretty easy to ask him to hide it, give him your answer, and ask for a correction by providing pictures of your work. Is that really different from any class? Everyone has a different opinion, but objectively, in terms of teaching, the data is the same, although it sure is a bit dull to learn through an LLM.

That is why we have Duolingo.

Education | The Market.

I am European & part of those who believe that our education system is now terrible. I wouldn’t go as far as to say that school didn’t teach me anything, but I would say that most of the important lessons or skills I acquired through the years aren’t thanks to it. I went through it as most do: bored, enjoying the social aspect of it, and only studying to pass my tests until I finally got the paper I needed to move on.

It doesn’t mean school is useless, but it should be adapted to nowaday’s technologies. We see many videos on socials media platforms of teachers losing their minds at their students for doing their homework with LLMs, despite their hallucinations… But those tools are here to stay & ignoring them is a waste of potential.

I share the video’s conclusion that we expect students to behave as if we were in 1950, which makes no sense considering the amount of technology we rely on nowadays & how impactful it could be if positively leveraged.

But we are not here to talk about me or my opinion of school & education; we’re here to talk about Duolingo & what they aim to become, which isn’t about languages but about education.

I asked Grok to provide some numbers with a focus on official sources, though these are certainly not 100% accurate. The goal is to get a rough idea of the figures.

I won’t focus on the schooling or university costs for now - this will be a subject for a bit later, but only on the extra classes as those are students who need help in specific subjects after school. Based on this data, we’d be talking about a $1.88B/hour market.

There aren’t extra classes all day; Grok estimated that students have more or less two sessions per week during the 36 weeks of the school year, for a total of 72 sessions & a grand total market of $128B per year, only for extra tutoring, only in three regions. A pretty lucrative market.

Duolingo is not ready to address it all, but this is what management is looking at and they confirmed it rapidly; this is what they shared in their IPO filing 4 years ago.

“According to HolonIQ, approximately $6 trillion was spent on education globally in 2019. And GSV Ventures reports that $160 billion was spent on digital learning, with digital spend expected to grow at a 26% CAGR from 2019 to 2026. We believe we can expand our addressable market by extending our scalable platform to other segments of learning such as literacy and math.”

If those figures are correct, we’d be talking about $850B spent in 2024 on many types of tutoring Duolingo isn’t addressing, but the conclusion is the same: the market for tutoring & online tutoring is huge, and this is what Duolingo is looking to conquer.

For now, Duolingo is focused on languages & has 130.2M monthly users. Demand for learning a new language is estimated at around 1.65B worldwide, meaning Duolingo covers around 7.8% of it as of its last quarterly data.

The online tutoring market is massive. And Duolingo is coming for it.

The Company.

I believe you already know Duolingo either because you used it yourself to learn - or try to learn, a new language or because someone around you used it. It went online a bit more than a decade ago and rapidly took the biggest chunk of the market.

The second-best language learning platform, Babbel, kind of falls short with “only” 16M subscribers - not monthly users, since its launch, around the same period.

The company and its mascot, Duo, took over the market thanks to their visuals, social & gamification features, marketing and shipping pace of new classes. Content quality also played its part competition could compare in quality. The difference came down to Duolingo’s users’ retention capacity.

They started as a freemium, offering free & limited classes, giving the opportunity to their users to discover the content & incite them to come back with features like day streaks, rewards or literal threats from Duo in the form of notifications. They also built content to be diversified & let their users explore it at their pace to avoid boredom or repetitions while sharing comforting, encouraging, or proud words on your progress. All of it combined with gamified visuals.

It is hard to keep people motivated in general & harder when the subject is learning a language, as it is a really slow process. Duolingo was, and still is, the best at it. It is fun to use & makes you want to go back to your lessons because of the gamification, the encouragements, your daily strike or the guilt Duo plays with.

No one wants to disappoint an owl.

Lessons.

One of the early difficulty for Duolingo was the volume of lessons available as most had to be done manually, which forced management to compromise and to either spend to have enough man power & deliver regularly or not and deliver slowly. It forced them to focus on one domain & a few languages at the beginning, the most demanded one without much room for expansion. This changed with AI.

Duolingo’s core business is to offer interesting, fun & interactive methods to learn, languages at the beginning but anything you’d like to learn eventually. As of today, the Owl can teach you 43 different languages, math, music, reading & writing basics and continues to expand to new domains, with chess classes announced during the last quaterly earning.

Some comments for the earning call to illustrate the shift between pre & post AI.

“So we’re very happy, we just added chess. It’s not quite yet live in the app, but it’s going to be live very soon. I should mention something amazing about chess is that it really started with a team of two people, neither of whom knew how to program, so they were not programmers. And they basically made prototypes and did the whole curriculum of chess by just using AI.

Also, neither of them knew how to play chess. And we started that for several months. And eventually, when they had a really good prototype, we had a whole team to professionalize it and put it in the app. So we’re very happy with it. This is, at the moment, mainly going to be to increase users. We think there as a huge demand for chess. There is hundreds of millions of people who want to play chess. And the same is true for math and music.”

To be entirely fair, I had a few talks on people who tried it and were not convinced but even if imperfect, the take away here is the pace & methodologies of improvement. Management will leverage AI and find the best ways to work with it to create more engaging content, grow their suject coverage & propose more advanced lessons for students who need to go further in any subject.

“We’ll also use GenAI to scale content across our language, math and music courses even faster.”

They already started.

This is why I started by talking about the education market at large. Duolingo helps to learn languages but the company will continue to expand & leverage their interactive teaching methods to teach everything, in time.

Conversion & Monetization.

This was about acquisition & retention, but more importantly is about conversion as free users don’t really bring money. Again, they execute perfectly on the freemium model & propose different subscription plans with some free trials, no commitment, different payment methods etc… Everything to propose every users a model which will please them & make them subscribe.

A free plan to discover, an individual plan for yourself with all the perks you need & a family or group plan to share with a community of up to 6, reducing your personal bill from $60 per year with Super to $120 per year for the family - so $20 per user.

The company scales on volume, as the investments are made by lesson & they won’t change whether there are 1M or 50M users. Attracting volume is what really matters, coupled with a strong conversion to their classic subscriptions and to their new AI subscription - Duolingo Max, afterward.

Duolingo Max adds new AI features: Explain My Answer, Roleplay, & Call Lily. The first one will detail your errors with personalized explanations, the second will enable the generation of real-world interactions which differ from the programmed ones in the lessons, and the last one will help you improve conversational skills with a written & recorded script to review it. Conversations are certainly the hardest part to simulate, so I believe this functionality to be pretty powerful.

Those new features aren’t yet available in many languages as they require powerful models & inference - only French, Italian, Spanish, German, Portuguese & Japanese for now. But it shows what will be possible in the future.

“We’re prioritizing using the latest AI models to deliver a high-quality experience.”

Small parenthesis: This reinforces my bullish case on Nvidia & the fact that we have no idea how much compute will be needed when many models like this ones are widely available. This is a great example.

Duolingo Max is more expensive as it implies more expenses on the company’s side to run the inference on the AI models, with individual plans around $170 annually & family plans at $200 - still affordable.

The feedback I found for Duolingo Max isn’t all positive yet - although people tend to focus on negative reviews usually, but this is to be expected as the service is new, AI is new by itself and it will take some time to develop great models, but the opportunity & the potential is here without any doubts. We will talk about AI later in this write-up. This is the beginning for Duolingo despites it not being perfect, there demand for Max.

“Max is now available to the majority of our DAUs and represents about 5% of total subscribers”

Advertising.

Those kinds of businesses usually need strong marketing plans to attract, retain, and convert users. Everything nowadays is a competition from them and it usually costs lots of money.

It isn’t any different for Duolingo, which spent a lot early on but rapidly stopped doing so and found a better strategy: to rely on social medias and leverage its very engaged community built through its amazing branding strategies.

With this kind of campaign.

Duo died. Speculation, eulogies, & jokes went on on social media; stars & companies piled on to profit from the buzz until management launched a campaign to revive Duo, a feat possible with 50 billion XP points. It took 13 days to bring Duo back, healthier than ever, telling the world this was just a test.

The campaign generated 1.7B organic impressions on social medias, 2.1M Instagram likes, 185M TikTok views, a 25% jump in monthly active Android users, 450 articles (60% from top-tier outlets), and more queries than some Super Bowl ads. And as management said.

“And the cost of that campaign really was essentially nothing.“

This is the power of Duo’s brand & proof of management’s qualities.

Total Adressable Market & Duolingo’s Future.

As already shared, around 1.65B persons learn a new language. It is hard to know the market share potential of Duolingo, but it has room with its actual 7% of penetration, 30% YoY growth & its new AI services which should drive demand.

It is hard to say what would be a reasonable penetration, but it would take more than a decade growing at 20% CAGR to reach 50% penetration & around 7 years to reach 25% penetration, achievable numbers in time.

By then, the company will propose new subjects & more languages, growing its TAM. Grok estimates that around 2 billions people are learning music or young enough to learn basic math, read or write. Some overlap as it is possible to learn music, math & a new language at the same time while playing some chess games. My point being that as Duolingo expands, its TAM will shift toward the entire educational market which is a multi-trillion dollar market.

Once again, this isn’t my perception of Duolingo, it is management’s.

The solution is already online, allowing teachers to craft lessons, homework and have visibility on their students’ advancements - something that doesn’t really exist at the moment although things might have changed since I left school - I don’t believe so.

It is free for now as the company is focused on delivering a product & gaining market shares, not on monetizing it. But you can see the potential, the alternative being for each school to develop internal tools? Again, I don’t believe so.

This is the next next next step for Duolingo, but it is coming.

Risks.

Translations.

The biggest risk is automatic & instant translation - for Duolingo’s core business right now. We already have tools capable of voice translation and any LLM can translate your sentences into any language.

But these tools have existed for some time and they’ve never replaced conversations nor the need to speak more than one language. Why? If you’ve traveled, you know conversation isn’t word-by-word translation. Communication is muh more than this and includes expressions, mimics, accents & so much more.

Google released its instant & automatic translation for meet few weeks ago and you can see how uncomfortable it is, with a simple second of delay changing the entire dynamic of the conversation.

Translations work for ordering at a restaurant or any quick interactions. My opinion? They’ll never work for social interaction. They haven’t so far, despite years of hearing this argument.

I might change my stance if we have interconnected brain chips capable of translating not just words but intonations, expressions & more perfectly & instantly. Until then, nothing will replace speaking the language yourself.

You’ll never get laid speaking through a translator. But you will with your accent, your mistakes, your expressions & your jokes. That’s what relations are.

Artificial Intelligence.

Beyond translations, many assume AI LLMs can offer what Duolingo provides, for free. You can ask ChatGPT to create a lesson to learn any language and I was the first one saying that AI will revolutionize education eventually.

And I maintain it, but once again there is learning & learning, those aren’t competition for Duolingo. Learning a language - and many other things, is complex, boring, long. It’s about repetition, memorizing exceptions, building vocabulary. And that is what ChatGPT or any LLMs will offer.

Here’s the prompt.

“Create a lesson for me to learn French conjugaison in a fun way.”

Sounds fun? Feels like school to me, and I’m not sure it’ll help anyone learn. Duolingo works because it transforms this boring, repetitive process into something fun. The owl is your friend, here to help you achieve your goals. Everything is gamified.

Competition.

We mentioned Babbel earlier, but there are many ed-tech platforms, and there will be more now that AI can be leveraged for countless subjects. But this entire write-up details how and why Duolingo dominated the market with its gamification and teaching methods - and that’s exactly why it will continue to lead.

I did the mistake of overlooking many companies over the years for this reason: the competitive landscape & the fact that others could replicate their business’s strengths. I ignored Netflix, Uber, Spotify & others. But I missed what truly mattered then: users. The winner isn’t the best platform based on personal opinion; it’s the platform that attracts & retain the most users.

Competition has always existed - in ed-tech and every other sector. Yet Duolingo, like those other companies, captured and held the lion’s share. Why? Because they were able to leverage they user base & revenues to scale faster, to propose more value than others & fidelize their actual users while always attracting more new ones.

Duolingo offers 43 languages plus mathematics, reading, writing & music, with chess coming soon and surely more subjects in the pipeline. They’re starting to expand and have the resources to do so, unlike many other ed-tech platforms. They attract more users, offer more content, and scale faster. They do this because they atrract & will attract even more by scalling while other ed-tech platforms remain focused on consolidating their core business without the resources to expand.

If you could choose an ed-tech platform offering thousands of lessons across diverse subjects versus one focused on a single subject, for similar pricing, which would you pick?

Exactly.

Monetisation, expansion, acquisition & retention.

These are core concerns for any investment in a business like Duolingo which depends on its user base, methodologies & ability to improve & expand. The company remains founder led which should be a boost in term of motivation & focus.

Those are risks and should be monitored quaterly during earnings.

Financials.

I mentioned earlier that Duolingo built a robust financial foundation before expanding into new subjects & AI. Here’s the visual version of my argument.

Rapid adoption, retention & strong conversion have driven a 46% CAGR over the last five years. A really strong growth which could continue, given their total addressable market & expansions as we talked about earlier. The language TAM remains huge & they own a very small portion of it, while any added subject would multiply their TAM. Lots of room for the future.

In terms of margins, this is a software company that scales with volume. They invest in R&D, infrastructure & new lessons to propose the best & biggest volume of lessions & scale revenues on volume & their different monetization plans - like Duolingo Max, which offer higher margins but require substantial development & maintenance costs. We should see margin fluctuate over the next few years as they refine monetization, but those should always remain really high while net margins should stabilize at worst & potentially continues to improve as once again, management leverages AI.

In term of cash, Duolingo is now profitable and should need less financing in the next years, hence less shares dillution - management confirmed less than 1% in the future. The dillution was reasonable - 17% in 5Y, considering revenue growth, but it is always better not to have any.

Dillution was the prefered way of raising capital and allowed Duolingo to build a really strong cash position with $822M of net debt at the end of FY24 & $273M of free cash flow impacted by $110M of SBCs. The owl became a cash machine and as its business scale on volume, cash generation will only increase moving forward.

Conclusion.

Duolingo built an important & rapidly growing user base through the years thanks to its branding, built to be different - fun, serious & engaging. Its teaching methods are comparable to competitors, but those aren’t the heart of the war. Users are, and users like Duolingo’s fun approach as it helps them be consistent & achieve results.

It is focused on teaching languages for all levels, with less than 10% market share in this market. It is expanding to new subjects: beginner mathematics, music, reading and writing, chess, and much more to come in the future, rapidly expanding its TAM and always providing more value to its subscribers.

Leveraging AI will help management create content: more complex for advanced learners, more volume for those who want variety in their lessons or subjects, and more methods to teach efficiently - like Duolingo Max.

Duolingo could become a global ed-tech, teaching all kinds of subjects to everyone around the world, from geography to advanced mathematics, passing by languages and any kind of games, be implemented in schools to finally leverage AI and help students achieve better results & more.