The Duolingo Streak: Why It Works and How to Copy It (Full Teardown)

In one paragraph

Watch on YouTube: The Duolingo Streak: Why It Works and How to Copy It (Full Teardown) (channel @arshsingh).

Key points

• In the video, Arsh says Duolingo did over $1 billion in revenue last year while spending only around 12% on marketing, against the 30 to 40% he says is typical at that scale.
• He says the real engine is retention: 37% to 41% of monthly users open the app daily, against 10% to 15% for typical apps.
• The streak is the core feature: one lesson a day keeps it alive, and he says over 50% of daily active users are on a streak of seven days or more.
• Loss aversion makes the streak work, and Duolingo kept the bar low, running over 600 A/B tests on the streak and simplifying it to one lesson.
• The streak saver notification, which nudges someone about to lose a live streak, is the one he says mattered most for retention.
• On top of that, social leagues of 30 people lifted learning by 17% and lesson completion by 25% and tripled the most engaged users, according to Duolingo's former product lead Jorge Mazal.
• He says 80% to 90% of Duolingo users arrive organically, which he treats as the payoff from retention.
• Free-to-paid conversion is 8.8%, up from 3% in 2020, helped by a two-week reverse trial, against a freemium norm of 2% to 4%.
• He notes that Duolingo does not publish clean day one, day seven and day 30 retention curves, so he does not invent them.
• The three levers for a small team are one daily action built into a streak, asynchronous social such as a weekly leaderboard, and a protected free tier with a reverse trial.

Transcript

Cleaned from YouTube auto-captions: caption errors fixed, paragraphs added, timestamps and channel subscribe prompts removed.

Now, last year, Duolingo did over $1 billion in revenue and they spent only around 12% of that on marketing. Now, most apps at that scale spend 30 to 40% to grow. So the obvious question a growth team asks is, what's Duolingo's marketing secret? Is it the owl on TikTok? Is it the unhinged social media theme? Or is it the 16 million followers? But here's the thing, that's not the engine. You know, the owl is just the amplification. If you simply copy the marketing and skip the main foundation machine underneath it, you'll end up with the costume without the body.

And you know what, here's why that should bother you. If you're pouring your capital into user acquisition before you build your retention system, you're not scaling, you're simply paying more every month to lose the same users just faster. But if you fix the one thing that Duolingo fixed first, the math flips. You know, your best growth channel stops costing you anything. That is why 80 to 90% of their growth is organic. Because the real number is this. 37 to 41% of everyone who opens Duolingo in a month comes back the same day.

And that's the DAU to MAU ratio, the stickiness number. Typical apps sit at 10 to 15%, and Duolingo is running two to three times the category. That one number is why they can grow the way they do. Today, I'm taking it apart. The real figures straight from their filings and the 3 pieces that you can copy with a small team of let's say 6 people and a seed round, not a billion in revenue. There's also one number in here that almost nobody talks about. And that single change that tripled their most engaged users. And I'll show you that one near the end.

For now, let me just pull up the numbers and help you walk through the entire teardown. Okay, so let's start with a stickiness number because everything else hangs off it. So here's the scale, 133 million monthly active users. 50.5 million of those users are daily active users. So that is the thing that gives you the 37 to 41% DAU to MAU ratio. And it's been climbing up more than 4 points year over year. Now think about that education for a second. Where the usual pattern is that you download an app in January, open it four times and never use it again.

Now that is not incremental. It's two to three times a normal consumer app and the trend here matters a lot. Because they have reported that the churn in Western markets have dropped from 47% to 28% over a few years. Now one honest caveat and because I know that this audience will check is that Duolingo does not publish clean day one day seven and day 30 retention curves. So I'm not going to invent them. What they do publish in their filings and what their own former product leaders have said on the record is more than enough to see the machine.

And guess what, it comes down to one feature that most people would call a gimmick and that one feature is called the streak. So it's as simple as that. One lesson a day keeps the streak alive. If you miss a day, it goes back to zero. Now it sounds trivial and that's exactly why most teams dismiss it. But it is the single most important retention feature in the app and the numbers behind it are all the tell. Now over 50% of their daily active users are on a streak of seven days or more. And more than 10 million people are carrying streaks over a year long.

And the most committed users have streaks running past a decade. Think about that for a second. Over 10 years of not missing a single day. That's how powerful this feature is to keep bringing the users back. Now obviously there are two things that make it work. The first one is basically your loss aversion. On day 200, you're not protecting a language goal. You're just protecting the 200. Breaking it feels like losing something that you already own. And while most teams get this backwards, Duolingo made the bar to keep it absurdly low.

Their retention team and this is Jackson Shuttleworth on the record. They ran over 600 A/B tests on the streak alone. And he said that one of the biggest wins was to simplify it. They stopped tying it to points and made it just one lesson. They even tested the button copy, changing it to continue and then to commit to my goal. And that also moved the number. So the lesson here is that lower the bar, more people will keep the streak. And then there's also the notification. And this is the whole discipline. Not one alert.

Company wide. They run 300 plus experiments a quarter, 20 to 80 live at any moment. But the one that mattered the most for retention was the streak saver. The nudge that fires when someone with a live streak is about to lose it. Their former product lead, Jorge Mazal, singled that one out as the notification that worked. You're not reminding a stranger to come back. It's simply reminding someone with something to lose that they're about to lose it. Now the streak and notification hold a user by themselves. But there was one more layer.

And this is the one I promised you at the start. The change that tripled the most engaged users. And that one thing is putting the social on top. So basically leagues of 30 people competing on points each week. Friend streaks where a third of all daily active users. And now keep us going with someone they know. And here are the impact numbers shared by Mazal publicly. After leaderboards launched learning went up 17%. Lesson completion went up 25% and the count of their most engaged users. The people doing an hour a day five days a week.

It tripled. He also cites a retention metric moving from 12% to 55%. I would treat that specific one as his internal framing. Not a clean textbook figure. But across all the metrics the direction is the same. Which means that the competition against real people on your own schedule locks people in. And that's the whole machine loss a version. A low daily bar. A relentless notification program and social pressure. Not one dollar of that is marketing budget. And that's the setup for the part that actually matters for your app.

So here's why I'm walking a growth team through a language app. Because the next part is the one that most founders get exactly backwards. Now because of that machine roughly 80 to 90% of the Duolingo users come in organically. Word of mouth, not paid. And that's the direct pay off of the retention. A product this sticky is a product that people talk about. So look at what it compounded into. At the 2021 IPO they had 40 million monthly users. Today it's 133 million revenue went from about 55 million per quarter. To over a billion in year.

That's roughly 19 times the revenue in four and a half years on only 12% marketing spend. And honestly speaking this is the part I want you to sit with because it's the whole thesis. See most apps do this in the long order. They raise around and the first thing that they do is to pour money into user acquisition. More installs. More spend scaled at the top of the funnel. But you know that if the product leaks all that spend does is push more people into a bucket with a hole in it. So if they don't come back after day one your customer acquisition cost keeps climbing while your retention stays flat.

You're simply paying to acquire users. You were always going to lose. Now Duolingo built the retention first. They earned the right to spend. And then they barely had to spend anything on marketing at all. And this is the same thing I say to almost every founder I talk to. You know that most apps don't have a demand problem. They have a leak problem between interest and activation. Spending more on user acquisition doesn't fix the leak. It only makes your cost per install more expensive and your payback period longer.

So there's one thing that you take away from this slide is that you need to fix the retention rate first. And every dollar of acquisition works harder. Now even the money follows retention. Duolingo's free to paid conversion is 8.8% up from 3% in 2020. That's a 176% jump and it put them at 12.2 million paying subscribers. You know in general the freemium norm is 2 to 4%. So they're running 2 to 4 times the best in class. They get there partially with a reverse trial. They give you the premium tier free for two weeks.

So when it ends you feel the loss instead of imagining the gain. And the mix keeps compounding. The family plan is now 23% of the subscribers. And the AI tier climbed from 5% to 9% of the paying base. But by now you understand that every bit of that works only because people are still there in two weeks. You cannot monetize a user who has already churned. And a user who sticks around long enough to convert is a user whose lifetime value finally Outruns what you paid to acquire them. See retention is upstream of everything.

Your LTV your acquisition cost all of it. So strip out the billion dollars and here's the part that you can actually use. Here are the three levers that you can transfer to an app at your stage. And none of them need a big budget. Number one is to find your single daily action and build a streak around it. Not points, not a dashboard, one action. For a fitness app it could be one workout. For a finance app it's one check-in. But make the bar to keep the action low and make the counter visible. And only notify the people who are engaging.

The whole trick is that one repeated action becomes a habit. And a habit is a retention that you don't have to keep buying. Second is to add social but asynchronously. You do not need a real time anything. A simple weekly leaderboard among friends or a small cohort. Tied to that same core action with a simple one tap invite. I think that's where Duolingo got both a retention lift and a free acquisition loop. Because if your friend has a streak going on with you, you show up. And number three is to protect the free tier and use a reverse trial.

You have to give the premium experience away for a couple of weeks. So the paywall is a loss not a guess. But you have to test it on your own cohorts measure conversion at the end of the trial. Not signups at the start. And honestly none of these three need a bigger budget. They need you to treat retention as the thing. That you build first before you scale spend. So Duolingo's edge isn't the owl. It's the 37 to 41% stickiness ratio built on a streak 600 experiments deep. A relentless notification program and social pressure on top of it.

Now that retention is what lets them acquire 80 to 90% organically and convert at almost 9%. It's why they grew 19 times in revenue on 12% marketing spend. Now you might be thinking sure, but that's Duolingo. They have hundreds of people and years of experiment fair. I'm with you on that one. You don't have their scale, but the three levers that I just showed you do not come from scale. They came before it. Every one of those was a small product change. Their team tested them when they were a lot smaller than they are now.

So here's the honest question to sit with you. If you turned off the marketing spend tomorrow, how many of your existing users will stay with you for 30 days and how many new users would come in? So if you don't like the answer, the problem isn't your acquisition. It's your activation and retention of course. That's exactly the kind of leak we pull apart when we teardown an apps funnel. If you want us to do that on your app, there's a link in the description to request a teardown. We find the biggest leak between install and activation and show you where exactly the retention is bleeding.

It's the same thing that I just did to Duolingo done to your funnel. And if you want the flip side of this, I break it down in my next video, which is why pouring more money into user acquisition stops working past a certain point. I'll see you in that one.