Growing Duolingo to 200m users, one habit at a time

A talk from Canvas 2017 on gamification, iteration and forming habits.

Language learning tool Duolingo is the most downloaded education app of all time. At Canvas 2017, Product Manager Zan Gilani shared his insider stories of how they’ve grown to 200 million users through constant experimentation and borrowing the best from game mechanics.

Zan kicked off with a general overview of Duolingo; beyond the impressive download stat and the fact they offer lessons in 25 languages (and counting), their mission is equally ambitious to provide free language education for all’. They believe in the wider ability of this to reduce educational disparity as well the app is currently being used by both Bill Gates, and public schools in Guatemala.

Besides being free to use, it works so well because it actively uses billions of data points to refine how languages are best learned, but more importantly because the app helps to solve a key problem for most; staying motivated with self-directed learning.

Zan Gilani, Duolingo

Staying motivated

Self-motivated study is notoriously hard just look at the completion rates of most MOOCs. It’s even harder online, where distractions are everywhere and the isolation is felt even more. 

So the product challenge for Duolingo then, is to help people stay motivated and form a daily habit around learning a language.

They want to set their users up to succeed, and they do so by borrowing from the gaming world. Games like CandyCrush have been designed to be addictive, and Duolingo has followed suit; its interactive lessons and progress tracking can feel extremely game-like. They’ve taken these features and tested/iterated on them regularly over the past 5 years to really figure out what truly helps language learners form daily habits.

One of the most crucial of these is streaks – in other words, being rewarding for consecutively reaching the daily goal you set when you signed up for an account. Zan took the Canvas Conference attendees through how they use streaks at Duolingo, and how they reinforce the four key criteria that need to be in place to form a habit:

Create small, concrete goals

Streaks have become pretty prevalent in product design; they’re used in Snapchat, Headspace and Facebook, just to name a few, and that’s because they work. Rather than your goal be “become fluent in Spanish’, it becomes ‘complete two lessons a day on Duolingo’ – which is a concrete goal that can be completed, rather than the nebulous goal of fluency.

As well as this, it rewards consistent practice rather than bingeing (a common pitfall when learning something new), and the significance of it accumulates over time, as users care if they lose something that’s been built up over days or weeks. Amazingly, people caught up in Hurricane Irma actually wrote to Duolingo asking them to reinstate their streaks after losing them as a result – just showing their power to motivate.

Visible progress

Zan used a useful analogy of flossing your teeth; people struggle to keep this as a habit, as it’s not something where the benefits are that obvious, at least not over a short period of time. Duolingo has a game-like ‘progress bar’ in the app, but they’ve iterated on top of this basic feature to demonstrate improvement more regularly and prominently.

The first experiment with the visual progress indicator of the streak displayed the streak clearly on the homepage, to a good boost of Duolingo’s key metrics; Daily Active Users (+3%) and Daily Retention after 14 days (+1%). Zan pointed out that although these don’t sound like a lot, they have an impact at scale and they’re running hundreds of these tests at once – 200 A/B tests at the time of the talk.

The second experiment showed how the designers emphasised the streak appearing visually after every single lesson – this increased the D14 retention goal by a further 3%.

The second visual experimentation for Streaks.

External triggers

Features like streaks help set users up for success, but how do you get them to keep coming back? The most common triggers are emails and notifications; they’re potent, but can be spammy. However, Streaks give you a more compelling reason to contact users. Duolingo sends notifications 23.5 hours after your last activity, to remind you not to break your streak – plus, habits are more likely to stick when done at the same time each day.

They also have used what’s been called a passive-aggressive reminder to people who aren’t coming back – ‘These reminders aren’t working, we’ll stop sending them’. Counterintuitively, these messages about ending communication boosted retention and helped people stick around longer.

A notification for a user who hasn't opened the app in a while.

User investment

Habit-forming products also have to create opportunities for people to invest in their own goals. Zan shared how the streak wager’ concept has helped them retain users people can bet a certain amount of their in-app currency lingots that they can stick to a streak for 7 days, for example.

They could see that this worked, most likely as people feel like they’re investing something of their own into their learning but not many users were taking it up. They surfaced the offering a different place in the flow (exposing it to users at the right time) and found that it boosted usage massively when shown in the right place (5% in D14). This sort of iteration and experimentation in small steps is all part of the Duolingo culture.

But if streaks don't work?

Streaks won’t work for every user, or there’s always a valid reason they get broken. Zan explained that their approach to this is to design for weakness and create multiple ways to win.

They know that weekends are a weakness (9% less Daily Active Users), so they launched a ‘weekend amulet’(again, borrowing from games) to allow users to protect their streaks over the weekend, leading to a 4% increase in the 14 day retention metric.

 

Duolingo also know that losing a streak feels horrible, so it’s important to have other ways to ‘win’ on top of streaks. Users can unlock ‘achievement badges’; although these are often maligned as a lazy way of doing gamification, they offer a lot when done right as they’re permanent and there are more interesting ways to complete them compared to streaks.

Interestingly, Zan shared an anecdote about the first experiment with badges which had no impact whatsoever. He explained that Duolingo’s enthusiasm for ‘minimum viable testing’ had gone a bit too far, and missed out the ‘viable’ bit. They initially granted a badge for users who had signed up, but nothing else, quickly learning that signing up is not enough of an accomplishment, and then you need to see the full range of badges up for grabs – one in isolation means nothing.

Zan’s talk was a fascinating insight into a product company that’s doing gamification right, and puts testing and iteration at the forefront of its culture.

Zan’s talk was a real highlight of Canvas this year. You can watch his talk in full below, alongside a Q&A with Zan.

Watch all of Zan's talk, as well as an exclusive Q&A.