Duolingo And The Big Bet Loop

One of the most under-appreciated outlier product stories there is

Out of the top 5,000 most valuable companies in the world, only one has its main revenue source as an educational product direct to the learner (D2L).

While I could point out the erroneous way that people spend money (you’d think spending money on your own education should be a higher priority than say…Candy Crush), I won’t. I’ll go down the path of how amazing the one product is that’s innovated and made a big impact in the learning space.

The product is Duolingo. It’s an outlier of the highest order and is a phenomenally under-appreciated product success.

I’m going to introduce the product broadly and why I believe they’ve been so successful, talk about the impact they’ve had and then highlight examples of why their success is no luck. Finally, I’m going to put my PM hat on and look at what I’d do to keep the momentum going in the future.

A quick aside before getting into it. Out of the top 1,000 most valuable companies in the world, there are around 130 technology companies. Not one of those is in Education Technology (EdTech). It’s astonishing how few EdTech companies have cut-through in the learning space. There’s no better way for an individual to spend money than on their education, yet such little is actually spent on it. In America, spending on supplementary education is 2% of a person’s budget, while spending on transport is 18%. Does society have an issue with priorities, or are there a lack of great products in the space?

The product

Duolingo is a language learning app whose mission is to “develop the best education in the world and make it universally available.” Their primary business is the language learning app, although they have recently started testing a reading app for kids and a maths app.

They’ve managed to integrate the best of ancient learning principles, human psychology and game design to build an app that is addictive for all the right reasons. No learning app has ever come close to the impact they’ve managed.

The reason I say it’s ‘for all the right reasons’ is because they have built their business on a freemium model (plus in-app purchases). Millions have been educated on the free version, and the subscription model means that only the people who can afford it are charged. On top of this, as opposed to an advertising revenue model, they’re not monetising your attention. They only get paid when the app delivers value to the user who pays for it. So, how have they managed to do this?

Taking a broad view, they’ve managed to build a faultless user experience with incredibly engaging user design and a learning experience that is better than any online experience out there. They have fun, they make fun of you (see below) and somehow and you feel emotionally connected to these little icons. In the end, you learn more of a language than you ever thought you would.

None of this is a fluke either. It’s the result of an organisation that’s been set up around product-first principles where they have a long-term vision and methodically solve problem by problem to get to this vision. They develop many hypotheses, test them relentlessly and only when they’re confident it’s worth perfecting will they take the time and resources to do so. Below is The Big Bet Loop that explains this process.

The impact Duolingo has had and how they’ve done it

Duolingo is valued at close to $5B, in a market that isn’t currently looking favourably upon tech stocks. It’s no surprise though when you look at some of the impact they’ve had. These are some numbers from the recent earnings release:

• Over 16m DAUs (daily active users) growing at 62%
• Over 60m MAUs (monthly active users) growing at 43%
• 4.2m paid subscribers growing at 67%
• Nearly $370m in total revenue
• Over 3 million people with a streak of 365 days or longer (using it every day for 365 days or longer).

How have they done this?

Constant, repetitive, relentless iteration

All tech companies these days know they should be testing and iterating, thanks to the build → measure → learn loop popularised by Eric Ries in The Lean Startup. 90% of companies that believe this iteration is at their core are wrong. When you look at how those 90% work, it’s actually very waterfall-y.

How can we tell from the outside that Duolingo doesn’t fit into the 90%? Well, first I suggest reading their blog. If you’re a product manager, this will immediately give you tips on how to do your job to a much higher level.

But it’s not just this, it’s something you can tell from using the app. Everything is intuitive. It doesn’t give you a chance to get frustrated and it leads you to places you want to be, even if you didn’t know you wanted to be there. Use this for 10 minutes and compare it to an app like LinkedIn or Twitter.

The 10% true product focused organisations are characterised by extreme long-term thinking, are centred around solving customer problems that benefit the business, make great use of data at their disposal and have a maniacal focus on outcomes over output.

We can also look at specific stories through the history of Duolingo to confirm they are in the 10%, starting with how they selected the right revenue model for them.

Early growth and the start of the test and iterate culture

In the early days, the app was always incredibly popular with the users, but it was free. A free product that users love is delightful, but (unless you’re WhatsApp) you’ll eventually need to make some money.

What they did next is a perfect example of the Big Bet Loop. They had a big problem that was holding them back from getting to their vision (no revenue). So they formed multiple hypotheses of how they could monetise the app, before persistently testing them all. Once they had enough data to support a winning hypothesis (freemium and in-app purchasing) that was still in line with the long-term vision, they exited the loop and started perfecting it.

Some of the hypotheses they tested to solve this problem included selling translations to news organisations and advertising. These both attracted significant revenue but they decided to sunset both of them. In the long-term, it was decided that both led to a worse user experience which would not help “develop the best education in the world and make it universally available.” How many companies would have the confidence and ability to think in the long-term to give up attractive revenue sources in the present?

Long-term thinking is at the core of all strong product organisations and this is a wonderful example of it.

The signup process

Another problem they were having according to their VP of Growth, Gina Gotthilf, was the conversion rate (people signing up once downloading the app). So as is their way, they start hypothesising on how to solve this problem. As you can see in the Big Bet Loop, n is ‘much greater than’ 1, and it’s clear this is what Duolingo does. Because it’s only when n is much greater than 1 that you start thinking counter-intuitively to a problem like this and think ‘maybe we should delay when we ask people to sign up.’

Low and behold, the tests of this hypothesis showed the best results and it’s now the way people are introduced to the app. Download it, use it, take a lesson, look at all the features – all without having to give your personal information away. The data was the data though, and it showed this was the most successful way of eventually converting users. Not only was it increasing the conversion rate, it was improving the long-term performance of users staying with the app.

A quick look at what happens after breaking out of this initial loop

To quickly look at what can happen when a feature breaks out of this loop and has been confirmed to have a permanent part of the Duolingo design, we can look at Streaks. This is the counting of days in a row you’ve used the app – the leader has over 3,000 days.

As an education product, Streaks made complete sense. Study after study proves that learning is most effective when it’s done consistently. For example, you’re far better off studying for 20 minutes two days in a row compared to 40 minutes on one day, and nothing the next. But does this necessarily solve the customer problem in the best possible way?

The early versions of Streaks were basic, so they could confirm whether this would become a permanent fixture of the app. But once they had the data to confirm it was going to be a part of the product long-term, they were able to start perfecting it. This is what led them to adding in steps like committing to a goal.

Humans have a nearly obsessive desire to appear consistent with what we commit to (read Influence for more on this). Duolingo worked this out and knew it wasn’t enough to encourage users to hit landmark numbers that they’d set for the user, they needed the user themselves to commit. Once a user hit the commit, the chances they’d become a daily user went through the roof.

In addition to this type of big change to the flow, they constantly test every minor detail. From the timing of notifications, where to show the streaks, the wording on the buttons, the facial expressions on Duo and so much more. It’s now one of the main success stories of the Duolingo product.

What would I do as a PM at Duolingo?

After everything I’ve written, it might appear Duolingo is the perfect app and will naturally just grow up and to the right. But that would be missing the point of how they got to where they are – relentless testing and iterating.

The world changes, preferences change and the best companies adapt with the time. Plus, with a language learning market size of 2B people and the potential of the entire education market spanning the globe, they have a long way to grow. With this in mind, I’m going to put my PM hat on and try to predict some future moves. Broadly speaking, the three main problems they’ll be continuing to look at solving are:

1. Increasing the new user pool for the free product
2. Increasing the conversion of free to paid
3. Decreasing the churn rate / increase retention

I’d like to focus on the main objective I see based on their current position – increasing new users.

How to increase new users to the free product

With a market mentioned above of 2B people and possibly more, the current position of 16m DAUs and 60m MAUs means there’s lots of room to grow. But what will add fire to this growth?

Without know the specific customer problems people are having, and without defining specific target users, let’s look at a few options of bigger bets they can place:

• Build out teaching products for high school and higher ed teachers to be able to implement with their students.
• Improve the social aspect and encourage more connections between users.
• Start a jobs marketplace for people to hire learners based on their Duolingo record.

I’m excluding AI here which I think is the biggest opportunity in education currently, because it’s too obvious – and as of a couple of days ago…they’ve actually launched.

Prioritising these big bets

Many users are learning a language to get a job, so if you follow this flow, you could make a case that it’s a fit for hirer and potential employee to meet here. It would make the process quicker, cheaper and more transparent which is valued in the HR world. However, just like the translation work, it’s off-brand for the D2L language learning app.

Building out teaching products for language learning is a valuable market. While D2L has been a successful user generator for them, they could escalate this user growth with a B2B model in an antiquated market. Teaching products in general lack any sort of innovation and genuine product principles we’ve outlined that Duolingo excel at. The difficulty about this is it would require a sales motion they likely don’t have. This is expensive and difficult to get going. It’s not uncommon for businesses to change sales motions from product-led to sales-led, but I don’t think they’re at this point yet.

I think the huge opportunity is in building out the social network they have started.

The big bet on socialising the experience

I believe that social media will become unbundled in the next decade. So rather than everyone congregating on Facebook or Twitter, there will be more ‘interest-specific’ social networks.

One of the most unifying experiences between two humans is doing something extremely difficult together, failing, making a fool of yourself, working harder, failing again and then gradually improving. It’s why there’s so much love shown in sporting teams. Learning a language fits all these points as well, and would form a solid place to solidify friendships and meet new people.

Focusing on the social aspect would be a way for Duolingo to bring in strong network effects to an already sticky product (network effects are when the app increases in value to a user when someone else joins). Initially, this will be a great driver of new users and long-term, it will greatly improve the retention levels.

It also aligns with the mission to “develop the best education in the world and make it universally available” as studies have proven time and again that learning in a group significantly improves learning outcomes.

Goals for the coming twelve months

Looking at the mission and strategy, we can now put in place some goals (I won’t put specific numbers down, but typically they should have measurable goals):

• Increase the number of users who sign up via referral.
• Increase the number of connections people are making.
• Increase the interaction between connections.

Tests to run on what will lead to these goals

Hypothesis 1: Users aren’t encouraged to connect enough in the current app.
Test: Add a button after finishing some lessons saying that ‘learning is more fun together, invite your friends.’ The measure will be the sign up rate of new user, plus whether they connect on the profile.

Hypothesis 2: Do people actually want to use Duolingo to interact with others?
Test: The current friend profile page has a comparison of lesson points between friends. I would allow this to be shareable and I’d add a like button to notify your friend that you’ve liked the competition. We’ll be looking for number of shares and likes.

Hypothesis 3: Will users share the app with non-users for more points?
Test: In the store, a user should be able to earn points for inviting referrals who sign up. Again, I’d be tracking to see the penetration of sign up from referral. I’d also want to ensure this didn’t negatively impact the current monetisation of the store.

These three hypotheses are aimed at the very next steps in trying to achieve the goals outlined above. They’re a way to test high level outcomes before diving in and perfecting the ones that show the most promise.

---

There’s no doubt that Duolingo has the potential to do as much for education globally as any other business in the EdTech space. They have the right business model, the right product and the right culture.

The final deeper question that we need to think about is how much learning can be achieved online, and how much should be. There’s no doubt that human connection and teaching is essential to the learning process, but how much is needed? It would be disastrous if anyone suggested eradicating the human element in education. However, there’s no doubt the world would be a much better place if more EdTech products can improve the learning process like Duolingo does.