How Duolingo Grows: If Angry Birds Taught You French
Insights on crowdsourcing, product principles, gamification, personalization, storytelling, ruthless A/B testing, duel flywheels, community, finding new ways to reaccelerate growth, and more.
👋Welcome to How They Grow, my newsletter’s main series. Bringing you in-depth analyses on the growth of popular companies, including their early strategies, current tactics, and actionable business-building lessons we can learn from them.
Bonjour, friends 👋
Put your hand up if you’ve ever wanted to learn a new language. 🙋
Now, keep it up if you started learning one and stuck with it to a point where you’re able to have a conversation in it. 👀
I imagine a lot of hands went up, and a lot went down. Mine certainly included. And for the simple reason that learning to speak another language is one of the most aspirational new skills people have, but one of the hardest to follow through with. We dream of ordering crepes and croissants in the streets of Paris in lovely flowing French, or touring the coast of Italy on a Vespa and navigating with locals using our Italian.
There’s no shortage of people dreaming about developing this linguistic skill. The will to learn is certainly there. The problem is—as is often the case with learning something new—we suck at sticking with hobbies like this. Just take the drop-off rate for online courses as proof (a stat I’ve contributed to several times) — an abysmal 90%.
This can be boiled down to the way we learn new things, which is usually rife with too much friction. And if the path to getting somewhere is tedious, overwhelming, or time-consuming…we probably won’t make it. That’s the same reason we abandon things like our new year's resolutions with such reliability (8% completion rate)…running a marathon sounds great, but getting out of bed every morning sucks. 🛌
So, the waytrumps the will.
And today, we’re going to get into the weeds of an extraordinary EdTech product that relentlessly optimizes the way we learn new languages.
Founded in 2011 in Pittsburgh, Pennsylvania, Duolingo makes learning languages fun and even slightly addictive. Instead of building a boring classroom-style product in an app, Duolingo took inspiration from the world of mobile games (like Clash Royale and Angry Birds) to create a new type of learning experience that keeps people engaged. Their app is beautifully designed, and through an intense focus on A/B testing and deep data and analytics, nothing is there by accident and everything is continuously optimized.
Duolingo has built a product-obsessed culture, and after spending 10 days learning French, I can see and feel it every day I pick up the app to maintain my learning streak. Their approach to building and growing—coupled with their mission of making learning languages free and universally accessible—has bought them over 500M users (40M+ Daily Active), a public valuation of $5.73B (DUOL 0.00%↑) , and has made Duolingo a leading consumer brand.
Honestly, this has been one of the most fun and interesting companies to learn about, and in today’s analysis, we’ll cover a ton of highly actionable advice for founders, product managers, and operators.
I’m excited. Or should I say, Je suis surexcité. 🤌
Here’s what we’ll be covering in our Duolingo analysis:
How Duolingo Started: The best education, universally available
Using the crowd to drive profits and reduce costs
The Duolingo learning experience
Money: The only language business speaks
How Duolingo Grows
Duolingo’s product philosophy
Duolingo’s technical advantages
Duolingo’s core flywheels
Duolingo’s community-led growth engine
Teaching High Valyrian: A lesson on experimental viral marketing
Reaccelerating growth with a top-down growth model
Small ask: If you enjoy today’s post and learn something new, I’d be incredibly grateful if you helped others discover my writing. Hitting ❤️ in the header, or sharing/restacking, are all a huge help. Thank you!
NB: If you’re reading this in your email right now, click to read the full thing. 👇
First time reading? Join 5K+ others to get all my deeply researched content. Free for now, not forever.
How Duolingo Started: The best education, universally available
You almost certainly don’t know the name, Luis von Ahn. But, you absolutely have used his technology. And I’m not talking about Duolingo…
How many times have you had to prove you’re not a robot before? 👇
Too many to count. 🤦 And you can thank Luis, because he’s the guy who invented this now widely-deployed reCAPTCHA service back in 2007 to verify real humans from bots, selling it to Google in 2009.
He also invented its more basic predecessor during his PhD in computer science while at Carnegie Mellon in 2003 — the old-school CAPTCHA with squiggly text.
As annoying as these tools are, they are in fact ingenious. In the early 2000s, people were writing programs to create millions of email accounts per day on Yahoo! so they could bypass the sending limits, blasting people with millions of spam messages a day. Intrigued, he picked this problem to work on for his thesis, and in his research figured out that computers could not read distorted characters. So, he built CAPTCHA V1 and ended up giving it away to Yahoo! for free.
Now, going from reCAPTCHA software to building one of the most consumer-centric, meticulously designed, and fun education apps seems like quite the leap. But, there’s a common thread. Both the idea for reCAPTCHA and Duolingo came from the concept of harnessing the power of crowdsourcing to solve problems.
Every time someone completes a human verification (about 200M a day), they also contribute to training computers in image recognition or digitizing a tiny portion of a book. In large part, the dual purpose of CAPTCHA tools is their brilliance. And the same was true for early Duolingo.
Using the crowd to drive profits and reduce costs
Coming from Guatemala City, Luis saw first-hand the inequalities of traditional education, and how expensive it was for people in his community to learn English.
He felt that, particularly in poorer countries, education was broken. Those who have money can afford the best education, while a much larger portion of people who have nothing barely know how to read and write. And because of that, they struggle to break through and find economic opportunity.
Just one example of the brutal cycle of inequality.
Inspired by this, in 2011 he turned his attention to language learning. He saw there was language learning software out there, like Rosetta Stone, but it was prohibitively expensive, and therefore, it didn’t do much in closing the education gap he was interested in.
If you look at language learning in the world, there are 1.2 billion people learning a foreign language and two-thirds of those people are learning English so they can get a better job and earn more. The problem is that they don't have equity and most language courses cost a lot of money.
— Luis van Ahn
So, along with a fellow graduate student, Severin Hacker, they set out to solve the education gap by promising freeforever language learning — making it accessible to everyone who wants to learn. Together, they believed that "free education will really change the world".
What a great mission.
Severin Hacker (left) and Luis von Ahn (right)
However, if you’re building a private tech company, someone’s going to ask how you make money. And you’ll need to have an answer.
This is where Luis piggybacked on his experience with his two CAPTCHA tools, as well as his thesis research: crowdsourcing.
The original idea for Duolingo was that as users learned a language and translated sentences, they were simultaneously being crowdsourced to translate English articles into their native tongues.
So, to keep Duolingo completely free, this business model of selling translations was how Luis could do it. Students would receive high-quality free language education, and businesses, in turn, got translation services powered by the students. Simply, Duolingo used an algorithm to combine the multiple efforts of students translating each phrase to produce a crowdsourced translation as accurate as those of a professional translator. Each article was translated by 30 to 40 people.
What’s more, crowdsourcing wasn’t just used to make money, but, also to save it. A key element of Duolingo is that the course content on the app is largely crowd/community generated — bringing down the cost of labor. This crowdsourcing model was key to their early success.
So in 2011, Luis and Severin launched Duolingo with just two languages: Spanish and German.
The novel hook of learning a language for free helped them quickly amass a following, and users began to contribute content and help each other learn new languages, all while helping keep the lights on by playing the role of translation service on the backend. This early traction caught the attention of investors, and that same year they raised a $3.3M seed round. This allowed them to add support for more languages and evolve their platform into what’s now one of the most effective ways to learn a new language, and the most popular, on the planet.
The Duolingo learning experience
The Duolingo learning experience sits at the rare intersection of fun and self-improvement.
It’s an app built for short bursts of on-the-go usage, nudging people towards 5 to 15-minute daily commitments. All the coursework is bite-sized, on-demand, crafted to feel like a game, and suitable to be done while taking the subway.
There are tons of gamification features that we’ll get more into, all designed to motivate and retain learners, and all constantly tested using the data generated by users to optimize each feature for maximum engagement.
But, while the app looks and feels like an entertainment product in many ways, the science behind the learning process on Duolingo is meticulous. Designed by language experts, the courses are made up of Skills, each divided into five Levels that comprise of a series of Lessons made up of short multimedia Exercises that help learners practice reading, writing, speaking, and listening skills. Plus, Duolingo’s core courses are supplemented with other in-app learning features like Stories and Audio Lessons, giving people more free comprehension practice.
Here’s a peek:
And Duolingo’s playful and gamified approach works really well. According to a study, learners who complete five Skills learn as much as students taking four university semesters of language education—and they do so in half the time.
Now, from its inception through to today, Duolingo remains free. However, in 2015 they dropped the crowdsourced translation business model, instead, finding a new route to monetization. 👇
Free may not have paid the bills, but it did come with a valuable upside: growth.
And according to Bing Gordon, partner at Kleiner Perkins who led Duolingo’s $20M Series C in 2014, “In Silicon Valley, there was this notion that if you had users, you could turn anything into money.”
So, while Duolingo had fairly lenient investors in the early years when it came to revenue on the balance sheet, patience started to run thin as their VC checks got bigger. As Luis recalls, “This was not very controversial back then, at least with investors. But this became controversial for us once we raised a ton of money, and we still weren’t making more money.”
From the beginning of Duolingo the culture was always anti-monetization, and they launched declaring they’d never do ads, subscriptions, or in-app purchases. But, in 2015 a change was somewhat pushed on them.
When a startup chooses to raise venture capital, it sets itself on a heavily-prescribed course. Suddenly, success isn’t defined merely as cash-flow breakeven with a long-term sustainable business. It has to be an exit of some sorts, and a big one at that. While Duolingo used venture as a lifeline to fund its product development, venture also came with pressure to become a billion-dollar company, or more. And that meant making revenue, not just growing engagement.
Today, they have all three of those things, with their freemium subscription model driving ~75% of their revenue. Although, the way they rolled this model out was unconventional, and kept them true to their core mission of giving away education for free.
Simply, the free version has everything you need to learn a language — nothing is paywalled in the path to learning. By upgrading to a paid plan (of which they now have two variations), you get perks, such as no ads and a few extra gamification features.
Most recently, they just launched Duolingo Max, a new subscription tier above Super Duolingo that gives learners access to two new chat-based features and exercises: Explain My Answer and Roleplay.
Of course, using AI built on GPT-4.
In many ways—just like Netflix, Peloton, and Spotify— Duolingo is a subscription-based content company.
And when looking at it through that lens, we can see Duolingo is doing an excellent job in this arena:
Spotify has a ~25% gross margin as it pays out >65% of its revenues to labels
Netflix has ~39% gross margins and spends ~$22B per year on content
Peloton’s subscription offering has ~67% gross margins
Meanwhile, Duolingo has ~73% gross margins (with app store fees being the main cost), and for reasons we’ll shortly get into (hint: crowdsourcing again), they had extremely low costs in creating their course content.
As of December 2022, Duolingo has ~2.5M subscribers, tallying them up $369.5M in annual revenue. And there is a ton of room for growth as the shift towards online learning continues to accelerate. The global E-learning market currently drives $245B in spending each year (CAGR 17.5%, of which, direct-to-consumer (D2C) language learning is large ($95B), growing (18.3%), and shifting online quickly (online CAGR 26%).
So, let’s get into how Duolingo is winning in this D2C language learning market. Along the way, we’ll uncover the key practical lessons, hopefully leaving you with a ton of insights to help build/grow your own products.
How Duolingo Grows: Product, testing, data, and mission-obsessed
As of early 2023, ~80% of Duolingo’s users were acquired organically.That’s huge.
Whether through social media, from a friend, or reading a newsletter like this one, Duolingo maintains this organic growth by building a remarkable product and giving it away for free. That’s how they drive their core growth engine of word-of-mouth virality.
And since we’re all interested in creating incredible products we want people to remark about, let’s start our analysis there. Put your hard hat on, because there is a goldmine of tactical advice here. 👷
Duolingo’s product philosophy
Duolingo’s flagship product is their learning app, and everything they put in front of customers is carefully crafted, and defined, by a few key product principles.
Product principles are the core DNA of the product. They’re the fundamental values that underpin every action, decision, or move the product team makes.
Principles are never reached. They’re not the vision, strategy, or KPIs. Rather, they’re like a North Star that guides customer-centric product development, helping create consistency, alignment, and focus.Yikes, that sentence had a lot of buzzwords. 🙃
According to Duolingo’s S-1 SEC filing, they have a few principles that have been instrumental to their growth since the beginning. Let’s go through them — I’ll expand on each and leave you with an actionable tip per principle.
Low friction 🛝
Of all the apps I’ve used (besides maybe TikTok), beginning the learning journey on Duolingo is one of the easiest onboarding experiences I’ve gone through. And since an education app is more complicated than TikTok, with many more opportunities for people to get overwhelmed, Duolingo is a stellar example of nailing low-friction onboarding.
After downloading the app, self-service signup is all led by Duolingo’s friendly and rebellious owl mascot, Duo. Starting off with what language you want to learn, Duo then talks you through a few more steps, like your proficiency and goals. With these three vectors, Duolingo can craft you a personalized course and user experience.
Importantly, Duolingo communicates the value you are getting during onboarding (see screen 4 below), and also, gives us a great example of how to use value-centric copy to explain why a user should provide some information/permission during onboarding. (See screen 5 for how they ask for permission to send notifications).
So far, that’s really good product-led onboarding. But, where things get excellent, is that Duolingo doesn’t require you to sign up before you start using the app and learning.
We found that by allowing users to experience Duolingo without signing up — do a lesson, see the set of skills that you can run through — we could increase those sign-up metrics significantly. Simply moving the sign-up screen back a few steps led to about a 20% increase in DAUs.
— Gina Gotthilf (Ex-VP of Marketing/Growth)
Users flow directly into a personalized course without having to hand over any personal details (see screen 6). This shows them the promised land almost immediately and offers users a chance to experience value firsthand before sending any registration prompts.
Only once a user has completed their first language lesson will they be asked (via a soft wall) to signup/create a profile, as well as make a small commitment:choosing a (brief) daily learning goal. Asking users to set a goal after already investing time in taking a lesson not only impacts the types of notifications they get, but most importantly, it’s a commitment. Which, no doubt, has psychological benefits for retention.
The first soft wall sign-up prompt a user gets can be dismissed.It’s optional. Only after a series of lessons is there a hard wall that requires sign-up. Of course, this was tested in great detail, and Duolingo found that the performance of this hard wall was actually improved by the presence of the nudges beforehand, as users were already primed to sign up.
Just looking at their low friction onboarding, it’s easy to see why Duolingo is often recognized as a gold standard for user experience design.
🛠️ What you can do with this: Test whether it is better for early users to sign up before or after using your product. If it is better to sign up after, form a hypothesis about when users will discover the value of your product. Estimate that point in your sign-up funnel and try a soft wall right after it. Do this as early as possible because, at every step, you have the potential to lose users. Then, try another soft wall after a few instances of engagement. Finally, try a hard wall and analyze the leaks at each wall. Keep testing where you place them until you find, to use some astronomy jargon, your Goldilocks Zone.
Motivating game mechanics 🎮
Duolingo believes that staying motivated is the hardest part of learning something new. So, they relentlessly focus on keeping people engaged/retained by molding the app as close as possible to one of the most sticky categories of products out there: video games.
Through gamification, instead of feeling like you’re taking a course, the whole experience is more playful — pulling from some of the most successful elements and dopamine hits from the world of gaming.
Here’s a short overview of what game mechanics Duolingo has going on:
ExperiencePoints (XP): The longer or more challenging a learning activity is that a user completes, the more XP they earn.
Motivator: People love to gain things.
Streaks: The more consecutive days someone uses Duolingo and hits their self-defined goal, the longer their streak will be. This is one of their most powerful engagement mechanics. Each day a user comes to Duolingo, they care a bit more about coming back the next day than they did the day before, increasing retention and DAU.
Motivator: People don’t like to lose things, and they love bragging rights.
Crowns: Learners collect crowns upon completion of a Level of a Skill, and a certain number of crowns are needed to unlock additional features.
Motivator: People love to work towards things.
Gems: Gems are virtual currency, which rewards learners for accomplishments like leveling up a Skill. Gems earned can be used in the Duolingo “Shop” to buy things like a “Streak Freeze”, premium subscription discounts, and merch.
Motivator: People love variable rewards and getting free stuff.
Leaderboards: Broken up into different leagues (i.e. Bronze to Diamond), leaderboards are weekly cohorts of 30 learners who compete to earn the most XP. Each week, the top 10 learners in each Leaderboard advance to the next League.
Motivator: People love competition (and bragging rights again)
Quests and Badges: Each day, new quests unlock (i.e. “Score 80% or higher in 4 lessons”) which allows users to earn rewards and badges. There are also social quests, which drive referrals and social dynamics.
Motivator: People love playing with friends.
To come up with these ideas, the Duolingo team takes inspiration from the games they play and love which have proven mechanisms. And one failed test they ran based on a game mechanic from Gardenscapes—counter moves—brings us all a valuable lesson.
In hindsight, it became clear why the Gardenscapes moves counter was not a good fit for our product. When you are playing Gardenscapes, each move feels like a strategic decision, because you have to outmaneuver dynamic obstacles to find a path to victory. But strategic decision-making isn’t required to complete a Duolingo lesson—you mostly either know the answer to a question or you don’t. Because there wasn’t any strategy to it, the Duolingo moves counter was simply a boring, tacked-on nuisance. It was the wrong gamification mechanic to adopt into Duolingo. I realized that I had been so focused on the similarities between Gardenscapes and Duolingo that I had failed to account for the importance of the underlying differences.
We had failed to account for how a change in context can impact the success of a feature. I came away from these attempts realizing that I needed a better understanding of how to borrow ideas from other products intelligently. Now when looking to adopt a feature, I ask myself:
Why is this feature working in that product?
Why might this feature succeed or fail in our context, i.e. will it translate well?
What adaptations are necessary to make this feature succeed in our context?
In other words, we needed to use better judgment in adapting when adopting.
🛠️ What you can do with this: Gamification can be a powerful strategy if used correctly, specifically on the consumer side. If you’re considering implementing game mechanics into your product, here’s some advice:
Understand your audience: Before implementing gamification, it is essential to understand your audience and their behavior. Analyze their motivations and what drives them to engage with your product. This information can help you create a more effective gamification strategy that resonates with your target audience.
Set clear goals: Define clear goals for your gamification strategy. What do you want to achieve with this strategy? Whether it's increasing user engagement, improving retention, or driving revenue, having clear goals will help you measure the success of your gamification efforts.
Keep it simple: Gamification should be simple and easy to understand for your users. Avoid creating complex rules or mechanics that could confuse people or turn them off from engaging with your product. Keep the experience fun and enjoyable, and don't make it feel like a chore.
Leverage user data: Use user data to personalize the gamification experience for each person. By tracking user behavior and preferences, you can tailor the gamification mechanics to each user, making the experience more engaging and rewarding.
Reward meaningful behavior: Reward users for meaningful behavior that aligns with your product goals. For example, if your goal is to increase user engagement, reward users for completing certain actions that encourage engagement, such as sharing content or inviting friends.
Test and iterate: Like any product feature, gamification should be tested and iterated upon based on user feedback and data. Continuously monitor the performance of your gamification strategy and adjust it as necessary to optimize for your goals.
For the curious, Rahul Vohra (founder/CEO of Superhuman) has some awesome thoughts on this:
Duolingo focuses obsessively on every detail of their design. From the precise shape and color of each button to the mood of the celebratory animations that congratulate learners upon finishing each Lesson, every corner of the app feels like it’s been calibrated to maximize user delight.
Their owl mascot (Duo) has become a popular brand icon and a marketing asset that shows up in social campaigns. Duo is also the face of most user-facing communication. Parlaying on the success of this cheeky owl, in 2019 they introduced an additional cast of characters to the app, each with their own personality and backstory. These characters now feature prominently across the app as people learn, and as the Duolingo team noted in their S-1:
We believe that recognizable characters and character-driven storytelling increase learner engagement and stickiness. As our product offering grows to include additional types of learning, our characters will become a common asset unifying the different learning experiences, such as Duolingo ABC.
Overall, their design choices and use of characters are excellent lessons in product storytelling, which is a way of communicating the value of a product by focusing on the user’s problem instead of the product’s features or technical data. It’s a conversational narrative woven into the whole product experience (i.e. across all user touch points) and, according to Stanford research, using product stories helps the customer develop an emotional connection with the product and increases the chances of a product being remembered by 2200%.
People don’t buy what you do. They buy why you do it. And in the more poetic words of Maya Angelou:
I’ve learned that people will forget what you said, people will forget what you did, but people will never forget how you made them feel.
🛠️ What you can do with this: Think about the why and the how of your product, and look to craft a product story that identifies who your target customer is and how they can get the most out of your product. Here’s some advice:
Speak to customers, ask for feedback, and figure out why people buy your product. Remember, people don’t buy 1-inch drill bits…they buy 1-inch holes.
Define your ideal customer profile and the core message for them. Understanding who this audience is will define your storytelling process and shape both the content and tone of your messaging. Don’t make it too professional — keep it in natural language, it’s a story.
Be the problem solver for your target audience and drive urgency. Address the pain points in your story (like frustration over a mistake in French), and highlight the positive impact of your product on the user’s problem. Create an urgency that drives anticipation for the professional and personal rewards your product promises.
Make the product story relatable. Good stories deepen understanding of a subject and make an idea more relatable. The more relatable the story is, the deeper the emotional response invoked. So, like Duolingo, try to include emotional designs and human interactions.
Choose the story medium and craft your messaging. I.e are you telling your story in-app (modals, banners, tooltips, mascots), across other channels (like blogs, emails, newsletters, video)…or like Duolingo, a mix of channels?
Diverse learning experiences leveraging data and AI 🤖
There’s a lot we could unpack here in terms of how Duolingo teaches, but here are Cliff Notes.
The Duolingo learning experience has been developed by learning science and language experts. From this, they’ve developed the most effective way for people to learn, and stick to learning, a language.
Courses are also personalized to each learner. For instance, a Spanish speaker learning English will get a different course than a Chinese speaker learning the same language. This was found to increase success and make a more relevant experience for users.
Duolingo leverages their large scale of learner data to inform continuous improvement of their learning content. For example, each time a learner finishes a Unit, they complete a brief assessment before unlocking Skills in the next Unit. This “Checkpoint Quiz” provides Duolingo with a way to assess the efficacy of each course, and enables targeted improvements if learners are not achieving mastery over a specific concept or grammatical structure.
They use AI for real-time hyper-personalization and gains in learning efficiency. Using data from over half a billion exercises completed daily, they train their machine learning algorithms to improve learning efficacy as learners engage with content. For instance, they predict the probability that a given learner will get the next exercise right or wrong, then adaptively construct the next exercise so it is “just right” in terms of difficulty, increasing motivation and improving learning outcomes.
Learning with Duolingo goes beyond their core app. They also offer a diverse set of experiences to build language proficiency outside of the app, such as their various podcasts focusing on comprehension experiences. This is a great example of the Whole Product Concept: where you view your product as more than just a sum of its features, but rather as everything involved with the experience customers have with your product.👇
🛠️ What you can do with this: Use personalization where possible. From onboarding to product storytelling and in-app usage — customization drives conversions and builds loyalty. Here are some tips on implementing personalization:
Gather and analyze user data: Start by collecting data on your users, including their demographics, behavior, preferences, and past interactions with your product. Then, look for patterns and trends.
Define personalization goals: Determine what you want to achieve through personalization. Is it increased engagement, improved user satisfaction, or driving revenue?
Create user segments: Segment your audience based on what you found in #1. This allows you to tailor your product to unique customer needs and preferences.
Use personalization algorithms/tools: Implement a way to recommend personalized content and experiences to your newly define segments. This doesn’t mean you need to make your own algorithm, there are lots of tools (like Amplitude Recommend) that can help you get something to market sooner.
Test and iterate: Personalization isn’t a checklist, it’s a continuous process that should be iterated on to optimize your personalization strategy’s effectiveness.
As you can see, these principles have shaped a product philosophy for Duolingo that is easy for people to refer back to. They are simple, focused on the customer, align with the company’s mission, and ultimately help Duolingo build better products.
Next, let’s look at some of the technical advantages Duolingo has that are helping them acquire and retain customers, as well as how you can leverage them in your own product. ⚙️
Enjoying this analysis so far? This post is free, so feel free to share it.
Technology is at the core of everything we do. We utilize the latest in machine learning and data analytics, along with a relentless focus on A/B testing, to fuel our differentiated learning experience.
We are proud to consider ourselves a technology and product-driven company—70% of our employees work directly on improving our products.
— via Duolingo, S-1 filing
A large data moat 🏰
With over half a billion exercises completed every day on our platform, we believe we have built the world's largest collection of language-learning data. We leverage this data by developing novel AI models at the intersection of machine learning, natural language processing, and cognitive science, which enable personalized instruction and power new product features that drive both engagement and efficacy. The majority of the data we collect for analytics purposes is in the form of tracking events, which are pieces of data associated with a specific learner action, such as opening the app or completing a Lesson.
— via S-1
🛠️ What you can do with this: You want to be collecting data on users and their behavior as soon as possible. For a new product (or company), consider the following:
Define clear business objectives: Before you start instrumenting event tracking and analytics, it's essential to know your business objectives and what your data will be used for. What specific insights do you want to gain, and how will they be used help you achieve your goals?
Establish a clear data governance framework: In order to be useful, data needs to be accurate, consistent, and trustworthy. And having a clear data governance framework is how to make sure you have good and actionable data. This includes defining data ownership, data quality standards, and data retention policies.
Use a robust analytics platform: Choose an analytics platform that can handle large volumes of data, provide real-time insights, output visuals, and scale as you grow (switching is a nightmare). Pro tip: make sure the platform you use integrates easily with your product and other tools in your stack. Amplitude and Mixpanel are great options.
Instrument event tracking thoughtfully: Be strategic about which events you track and how you track them. Recording data for the simple sake of just having it can lead to unnecessary costs, as well as a confusing/overwhelming mountain of metrics down the line.
A robust testing framework and disciplined experimentation culture 🧪
The foundation of Duolingo’s product strategy is their relentless focus on improving learner engagement through experimentation. As Lavanya Aprameya, Senior Software Engineer at Duolingo, describes:
“Test everything.” This is one of the key operating principles that we follow at Duolingo in order to continuously improve the learning experience for our users. It means we rely heavily on experiments and data to help us make informed decisions about any updates or new features we launch.
Experimentation has always been core to how Duolingo operates. On a given week, it’s not uncommon for us to have a few hundred experiments running simultaneously. We conduct experiments for any changes we want to make to Duolingo – from seemingly small ones like updating a single button in the app to rolling out a major feature like Leaderboards.
And as noted in their S-1:
The velocity of our A/B testing capabilities is a core competency that allows us to optimize the Duolingo learning experience at a rapid pace. Through A/B testing, we have increased the fraction of learners who came back one day after starting Duolingo from 12% when we first launched in 2012, to over 40% today. A/B tests also provide us with the data to make decisions that positively impact paid subscriber conversion.
Gina Gotthilf (Ex-VP of Marketing/Growth) led over 70 different types of A/B tests over her career at Duolingo and was instrumental to their testing culture. In a First Round Review article, she shared what has become Duolingo’s “Tenets of Testing” — an integral element of Duolingo’s growth: 👇
Rake in ideas and rank continuously. Simply, collect as many ideas as possible from anyone who has them. At Duolingo, any idea for a test — from anywhere in the company — is added to a master list in JIRA. Then, about every 6 weeks, they comb through the list to pick which ideas to work on. To choose, they simply ask two questions to find the highest impact/ROI tests:
How many people will be affected by this change?
How many hours will go into running the test?
Refine and repeat. Every test Duolingo runs needs to be a learning opportunity, and most of the time once they get data in they run new variations based on what they discovered. I.e, they don’t stop at one win or loss.
Design and test with restraint. Given how many tests Duolingo runs, you might think at any moment there are tons of variations going on. However, they stick to no more than three arms per experiment: a control and two test conditions.
By testing nearly everything and focusing on whatever produces results, Duolingo has created a very meritocratic way of looking at things. Instead of leaving decisions to opinion or egos or background, Duolingo lets the metrics make the vast majority of decisions.
🛠️ What you can do with this: The value of A/B testing is well known. To earn— and defend— an edge, the best companies continuously test all aspects of their product. And with that in mind, here is a great insight, said perfectly by Gina:
More than 50% of experiments fail. You need to tighten your seatbelts and be ready for that. Don’t allow your team to dwell too much on any specific experiment. Designers and engineers will want to perfect the experiment, but you want to create a Minimum Viable Product (MVP) of each experiment intelligently so that it’s simple and you’re able to move quickly. Your focus should be on looking at results regularly, analyzing those results, and then making decisions. You have to have a portfolio of experiments running at any given time. The most important thing for experiments that fail is being able to learn from them. Some experiments may have a negative impact on user growth. In many cases, people think that didn’t work out. But you actually hit a lever that impacts user growth now what happens if you push the lever the other way?
Shared infrastructure 🌐
One last advantage here worth mentioning…
Products across the Duolingo platform, including the core language app, Duolingo ABC, Duolingo for Schools, Math, and the Duolingo English Test, all share a single technology infrastructure. This creates huge operational efficiencies in implementing new features for each and allows Duolingo to innovate at a much higher velocity.
🛠️ What you can do with this: It’s tempting, for speed, to build separate technology for every new product you launch. But, there are huge benefits to abstracting services, like building a platform advantage in that technology, instead of just in one product.
Now, if you’ve been following all the deep dives in this newsletter, you’ll know how much I love talking about flywheels. Duolingo has two that virtuously work together. 👇
Duolingo’s core flywheels
Duolingo’s growth is driven by two mutually reinforcing flywheels: the learning flywheel and the investment flywheel.
The Learning Flywheel: The more people that use Duolingo, the more data and insights they collect, which in turn is used to improve engagement and efficacy. With more engagement and effectiveness, the more people speak about Duolingo, increasing the size of their user base.
TheInvestment Flywheel: Learner scale and word-of-mouth growth allow Duolingo to invest more time and money into product innovation, learning experience, and data analytics instead of brand or performance marketing. And, as more people use Duolingo and convert to paid subscribers, the more Duolingo invests in creating an engaging and practical learning experience. In turn, this increases their popularity and user scale, as well as the effectiveness of their data analytics, further widening Duolingo’s competitive moat.
Together, they look like this:
🛠️ Finding your own magical flywheel: A growth loop (AKA flywheel) is a self-reinforcing cycle that drives the growth and development of a business. It works by taking an input, action, and output and creating a cycle of growth, where each sprint of the cycle leads to the next sprint of the cycle.
In practice, if you’re a PM or founder trying to find and determine your product’s growth loop, here’s some stuff to think about:
Map out your customer journey: Start by understanding how people flow through your product, from the moment they become aware of you to when they become loyal customers. Map out the touchpoints and interactions they have with your product and identify where you can create opportunities for growth.
Find your activation point: Identify the point in the customer journey where users become activated and start engaging with your product more frequently. This is typically where you can create the most leverage for growth.
Test and iterate: Experiment with different growth strategies and tactics, and measure the impact on your key metrics. Use data to identify what's working and what's not, and iterate your growth loop/flywheel accordingly.
Leverage network effects: Look for ways to leverage network effects to drive growth. This could include encouraging user referrals, creating a sense of community around your product, or integrating with other products in your ecosystem.
Focus on retention: Retention is a critical component of any flywheel. Make sure you're providing a great user experience that keeps users coming back and encourages them to invite others to use your product.
Now, we already know Duolingo was founded on the premise of crowdsourcing. While the initial business model of leveraging users to deliver translations was switched out for a freemium model, the power of the crowd still lives on in Duolingo’s genes through community. And community is a powerful force in keeping The Learning Flywheel in motion by driving greater efficacy. 👨👩👧👦
Duolingo’s community-led growth engine
From course creation partnerships between volunteer contributors to user-generated content moderation systems, Duolingo is a great example of how valuable community involvement can be in helping grow a company. Especiallially when it’s regarded as a top-level business priority.
Let’s take a look at the different things Duolingo has done around community. 🔬
Crowdsourcing content moderation
High-quality content is essential to Duolingo’s success. If language courses get local nuance wrong or make any mistakes, that could destroy user trust. And without trust in the learning process or material, that core flywheel will get stuck.
But, while each course is reviewed by different language professionals, there are still gaps as courses are made up of millions of pieces of personalized content (including sentences, translations, hints, audio, and images). No amount of testing and review will ever leave a course completely error-free.
So, to ensure they maintain a high bar, Duolingo taps into the crowd and relies on translation review processes and feedback mechanisms from users who help identify areas that need improvement or refinement. This collaborative approach to QA guarantees continuous progress while maintaining consistent educational standards.
Similar to how Wikipedia beat Microsoft’s Encarta, this open model that relies on (1) intrinsically motivated contributors from a micro-community, and (2) the community self-regulating itself, clearly works really well.
A prime example of this feedback-driven model is Duolingo’s Report Quality Estimation Tool. Wherever users are in the learning journey, they have the opportunity to report concerns and issues — and over 200,000 reports are registered in this way every single day.
This tool, powered by machine learning, ensures that high-priority issues rise to the top and can be given the attention they require.
Crowdsourcing course/content creation
In October 2013, The Duolingo Incubator was launched. This was a platform where bilingual volunteers from around the world could connect to build new language courses for Duolingo (within their learning framework/methodology).
This was a brilliant way to (1) engage the community, and (2) ramp up the number of languages offered on Duolingo, assisting in new market growth with relatively lower costs.
Duolingo’s community team ultimately saw an opportunity to leverage the creativity and drive of their community members – and in doing so, the entire ecosystem was enriched. This UGC-like program ran for 8 years (up to their IPO) and was invaluable in helping Duolingo scale up their content to over 90 courses, and bring 300M people access to free language.
Crowdsourcing knowledge sharing
An undeniable drawcard that strengthened the Duolingo community was their forums. Here, people could start threads to discuss the content, the platform, the languages and their nuances, and any other relevant topic with the largest language learning community available. This was a huge value add to Duolingo’s user base — another example of building a whole product.
Offering this community-led support was also super helpful in sustaining the platform's autonomy while reducing the costs that come with traditional customer service models. Although the official forums were sunsetted in March 2022 due to logistical challenges, a strong community had already formed on other forums like Reddit. And also, in the real world…
Bringing the crowd together through offline events
Recognizing that (1) language learning is inherently social, (2) people are learning a language to actually speak it, and (3) there was an opportunity to expand their whole product language learning ecosystem — Duolingo launched offline meet-ups and events.
This bought learners together, gave people a chance to practice talking with others, and helped build brand loyalty.
Source: Duolingo
🛠️ Building and leveraging community like Duolingo: In a great talk by Laura Nestler (Ex-Global Head of Community at Duolingo, and now VP of Community at Reddit), she describes the core framework for how Duolingo builds community.
With a community-driven model, the hardest question you’re going to face is how to tap into the intrinsic motivations that get people to participate and engage week after week.
And she goes on to describe that in order to fuel community growth, you must invoke the right intrinsic motivations. Namely:
Autonomy. People need to feel like they’re in control and not being micromanaged. Them adding to the community needs to fit into their lives in a way that they want to contribute.
Mastery. People need to have a desire, and means, to get better at, or improve something, that matters to them.
Purpose: People need to feel like they are serving something bigger than themselves.
If you can deliver on these three things, start asking yourself how you can empower your community to add value.
And once you have an idea of where the crowd can be leveraged, go and validate early, cheaply test your concept (focus on high-effort, low-tech), and only then start to systemize.
Teaching High Valyrian: A lesson on experimental viral marketing
As mentioned, almost 80% of Duolingo’s users are acquired organically through word-of-mouth virality. Still, they have made some investments in marketing to supplement this model and amplify their brand. In turn, driving more chatter without needing to spend much on paid acquisition.
And given how excellent their brand campaigns are of creating experimental viral content that’s both entertaining and educational, let’s give them some brief attention.
The bottom line:Tapping into trends/the cultural zeitgeist allows Duolingo to become part of the conversation, and focus on organic marketing over paid ads.
For example, Duolingo partnered with HBO to create a fictional language course (High Valyrian) given the popularity of Game of Thrones. This tapped into the fandom of GOT, and through tactical placements, drove waves of new users to Duolingo to learn the language of the dragon.
And this wasn’t the only time…
In 2018, the company collaborated with CBS to develop a Klingon course for “Star Trek” fans. Last year, it offered aspiring French speakers a free month of classes to help co-market Season 2 of Netflix’s “Emily In Paris.” DuoLingo has also seen upticks in learning certain languages even when there’s not an official partnership. For example, it found that more people started learning Korean on the app after the Netflix debut of “Squid Game.”
When people sign up for DuoLingo to learn a fictional language—whether it’s High Valyrian or languages from other shows such as Klingon in “Star Trek”—43% of them also begin to study another language, according to DuoLingo.
They also tested a partnership with the widely popular gaming company, Roblox, by creating an experience that lets people test their Spanish skills in various virtual environments. This experience has been visited more than 10 million times.
We try and find moments in culture, where there are these cult audiences — these cohorts of people — who are hardcore fans about something. What we do is really try and resonate with them because we feel like when we can inspire identification with them and our brand and what we do as an education language brand, they almost become ambassadors for us. That in and of itself is a piece of content we can share and go back and engage with that community.
— James Kuczynski (Creative Director of Brand and Marketing at Duolingo)
🛠️ Creating experimental, timely, content for your own brand: Going viral is not a strategy in and of itself…it’s an extremely hard goal to attain. That being said, you can still set yourself up for capitalizing on viral moment like Duolingo does, by:
Identifying cultural trends and conversations: Start by making sure you stay up-to-date with what’s going on in internet culture (using social media listening tools and market research), specifically in the world your target audience lives in.
Aligning your brand with the trend: Once you’ve found a relevant topic (i.e a new series premiere) think about how your brand can align with it in a meaningful way. This might involve creating content that leverages the trend in a humorous, informative or emotional way, while also highlighting your brand's unique value prop.
Creating shareable content: Then, create content that is easily shareable and spreadable, ideally optimized for the different platforms your target audience is using. Importantly, make sure it’s content that is genuine, original, and rooted in your brand’s identity.
Leveraging partnerships: Collaborations can help amplify your content's reach into new audiences and increase its virality. Look for opportunities to work with other brands, influencers, or content creators that share your target audience, align with the trend you’re leaning into, and align with your band.
Now, in 2018, after a good run of strong year-over-year growth, Duolingo’s Daily Active Users (DAUs) metric was hitting a plateau. In this last section, let’s look at the brilliant way Duolingo reignited their growth using a data-driven growth model. 👇
Reaccelerating growth with a top-down growth model
To maintain a healthy ecosystem of Daily Active Users (DAUs), all of who are at various points in their lifecycle, is a delicate balance. The more learners Duolingo has, the more diverse their needs become.
To maintain organizational focus while serving an expanding, and evolving, population, we orient teams around movable metrics that matter, and then run hundreds of A/B tests to optimize for those metrics!
But how do you decide on the metrics that matter? And how do you advocate for an organization to adopt new metrics? And what happens if existing metrics stop moving? Our Data Science team developed a growth framework that helped to grow DAUs by 4x since 2019
— Erin Gustafson (Lead Data Scientist at Duolingo)
For a later-stage company, that’s a super impressive lift in engagement.
So, let’s take a look at the path that led Duolingo to their framework (the Growth Model), the impact it’s had on their growth, and how they’re thinking of evolving the framework to take them into their next phase of growth. Hopefully, this leaves you with some inspiration. 🤓
The Growth Model
Up until growth was stagnating, the team focused on improving DAUs was struggling to find meaningful, needle-moving, A/B tests. So, they rethought their approach. 💡
As Erin describes, “Could we refine our focus by optimizing metrics that drive DAU indirectly? In other words, how could we break up the DAU monolith into smaller, more meaningful (and hopefully, easier to optimize) segments?”.
This led them to the Growth Model: a series of metrics they created to helped identify new avenues for kickstarting growth.
It is a Markov Model that breaks down topline metrics (like DAU) into smaller user segments that are still meaningful to our business. To do this, we classify all Duolingo learners (past or present) into an activity state each day, and monitor rates of transition between states. These transition probabilities are monitored as retention rates (e.g., NURR or New User Retention Rate), “deactivation” rates (e.g., Monthly Active User, or MAU, loss rate), and “activation” rates (e.g., reactivation rate).
Illustration of the Duolingo Growth Model: Technical Details
The model above classifies users into 7 mutually-exclusive user states:
New users: learners who are experiencing Duolingo for the first time ever
Current users: learners active today, who were also active in the past week
Reactivated users: learners active today, who were also active in the past month (but not the past week)
Resurrected users: learners active today, who were last active >30 days ago
At-risk Weekly Active Users: learners who have been active within the past week, but not today
At-risk Monthly Active Users: learners who were active within the past month, but not the past week
Dormant Users: learners who have been inactive for at least 30 days
As the arrows in the chart indicate, we also monitor the % of users moving between states (although we watch some arrows more closely than others).
— Erin Gustafson (Lead Data Scientist at Duolingo)
Identifying their new movable metrics
With their newly minted Growth Model, and trained on Duolingo’s goldmine of historical data, the DAU team began to run growth simulations. The goal of these simulations was simple: find new metrics that, when optimized, were likely to increase DAUs the most.
In a great post by Jorge Mazal (ex-Duolingo Chief Product Officer), he describes how they started using this new secret weapon:
With the model created, we started taking daily snapshots of data to create a history of how all of these user buckets and retention rates had evolved on a day-by-day basis over the past several years. With this data, we could create a forward-looking model and then perform a sensitivity analysis to predict which levers would have the biggest impact on DAU growth. We ran a simulation for each rate, where we moved a single rate 2% every quarter for three years, holding all the other rates constant.
Below are the results of our first simulation. It shows how those small 2% movements on each lever impacted forecasted MAU and DAU.
So, by systematically pulling each lever in the model to forecast what the downstream impact would be, they figured out that Current User Retention Rate was the lever to invest the most time in optimizing for.
They spun up a new team, the Retention Team, and CURR became their North Star metric. This new team quickly set out to validate that (1) CURR was a metric they could move, and (2) moving it actually impacted DAUs (correlation ≠ causation).
As you can tell, it did. And this guiding metric helped Duolingo prioritize the right things and find new vectors of growth, pushing them past a period of stagnation. 📈
Source: Jorge Mazal
I’ll leave you with this insightful parting thought on the growth model.
A careful reader will note that the Growth Model is calculated on an aggregate basis. This leads us to our second question: What opportunity are we leaving on the table by reducing a diverse learner base to a simple average? Averages are convenient and scalable, but we’ve found that because of our strong growth in CURR over the years our bases of Current Users has grown into a new monolith of users (90% of our DAU fall into this state!).
Why does this matter? Well, we’ve found that our aggregate metrics aren’t allowing us to see all of the distinct, diverse learners in each state. This means that CURR is an increasingly imprecise measure of Current User behavior, and we run the risk of this metric becoming hard to move (just like the 2018 problem we faced with DAUs!). It’s also harder to set reasonable goals and forecast accurately, which is increasingly important now that we are a public company.
With these questions in mind, we’ve begun exploring “bottom-up” methods for user segmentation as a complement to (and perhaps as an eventual replacement for) our current “top-down” method from the Growth Model. By turning to unsupervised learning techniques to allow unexpected patterns to emerge in the data, we’re also moving the organization away from analytical frameworks that can foster confirmation bias. The “top-down” nature of the Growth Model bakes in a lot of our preconceived notions about what matters for our business, while the “bottom-up” nature of our new approach will unlock new insights beyond “the path most taken.”
If you enjoyed this analysis, the best way you can support my work is by hitting the like button in the footer below, giving this post a share, or better yet, just telling some friends or co-workers about How They Grow. 🙏
Super interesting to see how they show an onboarding without a paywall, because I feel that this is usually what most experts currently recommend. What I definitely struggle with is to put the sign up so much later because it's so much easier to subscribe to a paid version of the app when you are signed up already. Thank you as always for the great input, I really appreciate it.
This is a fascinating piece. After using Duolingo (20 day streak 😇) I can see the relevance behind the article. Thanks for sharing, great listen.
Super interesting to see how they show an onboarding without a paywall, because I feel that this is usually what most experts currently recommend. What I definitely struggle with is to put the sign up so much later because it's so much easier to subscribe to a paid version of the app when you are signed up already. Thank you as always for the great input, I really appreciate it.