breme.

Case study · Retention and loyalty · Duolingo

Duolingo and the streak: designing a reason to come back every day

By BREME · · 6 minute read

In short

Duolingo treats the return visit as the product's main job. Its published work shows a run of small, tested changes: making the streak easier to keep, building in forgiveness with streak freezes, choosing reminders carefully and measuring the retention of its most active users above everything else. Its former product chief reports daily active users grew 4.5 times over four years. The lesson for any organisation that needs customers between purchases is to design the return, not hope for it.

Who this is for: CX leads, product and operations managers and executives in subscription brands, membership bodies, multi-site clinics and any organisation that needs customers to come back between purchases or appointments.

The problem: growth had stalled

Duolingo is a US language-learning app with a global audience. In a 2023 article for Lenny's Newsletter, former chief product officer Jorge Mazal described joining as head of product in late 2017. By mid-2018, growth in daily active users had slowed to a single-digit rate year on year.

The first fixes did not work. A moves counter borrowed from a mobile game produced completely neutral results. A referral program lifted new users by only 3%, partly because the most loyal users already paid for the premium tier and were left out.

What changed the outcome was not a single feature. It was deciding which customer behaviour mattered most and then testing many small changes against it.

The measure that changed the work

Mazal's team built a growth model that split every user into one bucket: new, current, reactivated, resurrected, at risk or dormant. A current user was someone who had used the app today and at least once in the prior six days. The rate at which current users stayed current was called current user retention rate, or CURR.

When the team simulated small improvements in each rate, CURR had about five times the impact on daily active users of the next best lever. It had not moved in years. Mazal reports that the work that followed lifted CURR by 21% and cut daily churn among the best users by more than 40%.

The measure has stayed central. Duolingo's shareholder letter for the June 2026 quarter put CURR at an all-time high of 84%, up about one percentage point on the year before, with 58.7 million daily active users, up 23%.

For organisations outside software, the point is the choice of measure. Duolingo did not chase sign-ups. It chose the behaviour of its most engaged customers and protected it.

Making the streak easier to keep

The streak counts the days in a row a learner completes a lesson. In a 2020 post on the Duolingo blog, the company explained a problem it found: the streak was tied to each learner's daily goal, and learners who set the most demanding goal were the least likely to be on a streak. The goal had become a barrier to the habit.

Duolingo separated the two, so one lesson was enough to extend a streak. In its A/B test, the company reported relative lifts of 3.3% in day 14 retention, 1% in daily active learners and 10.5% in daily learners on a streak after 20 days, and 19% among new learners. It was open about the trade-off: fewer learners reached their daily goal. A year on, just over half of daily learners had a streak of at least seven days, up from about a third.

Jackson Shuttleworth, a group product manager on the retention team, told Lenny's Podcast in December 2024 that Duolingo had run more than 600 streak experiments. He described retention as most fragile in the first seven days, and said changing a button from 'continue' to 'commit to my goal' significantly increased retention.

Building in forgiveness

A habit that breaks the first time life gets busy is fragile. Duolingo's 2022 blog post on the research behind the streak describes the streak freeze, which pauses a streak for one day. Allowing learners to hold up to two freezes at once raised daily active learners by a relative 0.38%, and the company reports it did not encourage more days off. The post cites research suggesting a little slack can be more motivating than rigid rules.

The same post reported that learners who reach a seven-day streak are 3.6 times more likely to complete their course, and that a new celebration animation on the streak screen lifted the share of new learners still using the app seven days later by 1.7%.

In June 2026 Duolingo went further with a one-off Streak Revival event. Eligible learners could restore their longest streak by opting in and completing three lessons. The company reports 15.4 million learners took part, including nearly 8 million with no active streak when it began. Its shareholder letter names the event as one of three factors behind faster user growth that quarter.

That is a planned way back for lapsed customers, framed around their own past progress rather than a discount.

Reminders treated as a channel to protect

Notifications are where many organisations damage the return journey by sending too much. Mazal writes that the team's rule was to protect the channel: sending more notifications needed strong justification and the chief executive's approval. Even so, he describes a reminder that warns learners their streak is about to end as the streak team's first big win.

A 2020 Duolingo blog post explains how reminders are chosen. An algorithm picks from pre-written messages for each learner, counts success as a completed lesson, and demotes messages the learner has seen recently because novelty wears off. New messages are tested on a small group first. The post admits most people have swiped one of these reminders away.

Mazal also reports that leaderboards, which group learners into weekly competitions, increased overall learning time by 17% and tripled the number of highly engaged learners.

The criticism

Habit design has costs, and a fair reading includes them. A 2022 qualitative study accepted at the ACM Learning at Scale conference analysed nine years of Duolingo forum posts and interviewed 15 users. It found some users become fixated on game elements such as points and leaderboards and are distracted from learning, which the authors say can waste time and hurt learning performance. They link this to competitiveness, playfulness and following the crowd.

Duolingo's own writing touches on the tension. Its 2022 post names loss aversion as a motivator as streaks grow, and also reminds learners that losing a streak does not erase what they learned. In May 2026, CX Dive reported chief executive Luis von Ahn saying the company had overdone monetisation in the free experience, adding friction that worked against daily use, and had since made more features available to free users.

In our view, the lesson is that a return measure tracks the habit, not the outcome the customer came for. Both need watching.

Lessons for organisations that need customers to come back

Most organisations that rely on return visits are not apps. A clinic needs patients to rebook, a membership body needs members to engage before renewal, a subscription brand needs a second and third order. The same design questions apply. These are the ones we would put on the wall:

  • Choose one return measure for your most engaged customers, such as active patients who book their next visit before leaving, and give it an owner.
  • Lower the bar for the return action. Ask what the smallest meaningful step back is, and count that.
  • Design for the first seven days, or the first month, when the habit is weakest and handovers between teams are common.
  • Build in forgiveness. A missed appointment or a lapsed month should have an easy way back, not a penalty that ends the relationship.
  • Treat reminders as a channel to protect. Test messages, retire tired ones and make sure each one leads to a single clear action.
  • Check that the measure still serves the customer's real goal, and listen for signs it has become pressure.

Sources

See your journey from the customer's side

We map the end-to-end journey with you, find the moments that matter most, and build a strategy your teams and systems can deliver.

Start a conversation More insights