Duolingo Found the One Number That Moved Everything

Case study · Marketing 465 · Measurement: KPIs and Dashboards

Duolingo restarted its growth by modeling how users move between engagement states and picking the one retention rate that drove daily active users, instead of chasing whatever looked promising.

The situation

By 2018 Duolingo was already the most downloaded education app in the world. Downloads were not the problem. Keeping people coming back was. According to Jorge Mazal, who joined as head of product in late 2017 and later became chief product officer, growth in daily active users (DAU) had slowed to a single-digit rate year over year by mid-2018.

Duolingo’s business depends on daily use. A language takes months of practice, and free users who show up every day are the pool that subscribers come from.

The challenge

The team’s first instinct was to copy tactics that had worked elsewhere. Mazal has described two early bets. One borrowed a limited-moves mechanic from the mobile game Gardenscapes. The result on retention and DAU was neutral, and the team behind it was disbanded. The other was an Uber-style referral program offering a free month of Super Duolingo. It lifted new users by only 3%, partly because the most loyal users already paid and had no use for the reward.

Both ideas were reasonable. The problem was that the team had no shared model of which number, if moved, would move DAU the most. Without that, every idea looked equally worth trying.

What they did

The team built a growth model. Every user was placed in exactly one bucket each day based on recent activity. Four buckets added up to DAU: new users, current users (active today and at least once in the prior six days), reactivated users (back after 7 to 29 days away), and resurrected users (back after 30 or more days away). Other buckets tracked at-risk weekly and monthly users and dormant users.

Each bucket had a retention rate. Current User Retention Rate (CURR) was the chance that a current user stayed current the next day. The team ran a sensitivity analysis: move each rate by 2% per quarter for three years, hold the others still, and see what happens to DAU. CURR came out with roughly five times the DAU impact of the next-best lever. That made it the team’s North Star.

With one number chosen, the work became focused:

• Leaderboards: weekly leagues that grouped learners with similar activity levels and let them move up or down.
• Streaks: a streak-saver reminder, then streak freezes, calendar views, milestone animations and rewards.
• Notifications: steady tests of timing and copy, under a rule to ‘protect the channel.’ Raising notification volume required CEO approval.

What happened

Mazal reported the results in a 2023 essay for Lenny’s Newsletter. Duolingo’s own blog published separate data on streaks. Duolingo went public in 2021, and Mazal credits the DAU turnaround as key to that IPO.

• Over four years, CURR rose 21% and DAU grew 4.5 times (Mazal, Lenny’s Newsletter, 2023).
• Daily churn among the best users fell by more than 40% over the same period (Mazal, 2023).
• Leaderboards raised overall learning time by 17% and tripled the number of highly engaged learners (Mazal, 2023).
• Learners who reach a 7-day streak are 3.6 times more likely to complete their course, the company said in January 2022 (Duolingo blog).
• Letting learners equip up to two streak freezes raised daily active learners by 0.38% (Duolingo blog, 2022).

Why it worked

The model turned DAU, a lagging number, into parts the team could act on. DAU tells you how many people showed up. CURR tells you whether your most valuable users are building a habit, and it moves first. That is the leading and lagging pairing in practice.

It also made the team say no. A good idea that would mostly help new or resurrected users ranked below a smaller idea that kept current users current. The notification rule did the same thing: it stopped the team from buying short-term DAU by spamming people, which would have hurt the channel over time.

Notice the scale of individual wins. A 0.38% lift sounds small. Applied to a compounding retention rate across millions of daily users, it adds up. Only a team that knows which number matters will chase gains that small.

Where it gets complicated

A North Star can pull against other goals. In its Q3 2025 shareholder letter, Duolingo said it was ‘prioritizing user growth over monetization in the A/B tests that get launched.’ On the Q4 2025 earnings call in February 2026, the company said adding friction to the free tier had been its fastest way to raise revenue, but it hurt DAU growth. It chose DAU and said revenue growth would slow as a result.

DAU also responds to forces a growth model does not capture. DAU growth stayed at or above 40% year over year from Q2 2022 through Q2 2025, then fell to 30% in Q4 2025. Coverage tied part of the earlier slowdown to consumer backlash over CEO Luis von Ahn’s 2025 comments about becoming an ‘AI-first’ company. Brand trust shows up in the dashboard, but only after the fact.

Finally, a metric built on daily habit can reward features that make people feel guilty for missing a day. Streak mechanics drive retention. Whether they always serve the learner is a fair question, and Mazal himself noted that some of the link between streaks and retention is correlation, not cause.

The marketing principle

Measure what matters to the goal, not what is easy to count. Duolingo did not add more numbers to its dashboard. It found the one number that, when it moved, moved the goal, and it organized the team’s work around it. The test for every KPI is the same one Duolingo applied: if this number changed, what would we do differently?

Sources

• How Duolingo reignited user growth, Lenny’s Newsletter (Jorge Mazal) (2023)
• How the Duolingo streak builds habit, Duolingo Blog (2022)
• Duolingo Q3 2025 Shareholder Letter (Form 8-K exhibit), Duolingo, Inc. via SEC (2025)
• Duolingo will limit friction as growth slows, CX Dive (2026)
• Duolingo, Inc. Form 10-K for fiscal year 2021, SEC (2022)

---

Back to Measurement: KPIs and Dashboards · All case studies

Brand names, images and trademarks belong to their owners and appear here for educational purposes only.