Cem Kansu
Chief Product Officer at Duolingo
Where to hear and read about Cem Kansu1 episode or test
Cem Kansu is Chief Product Officer at Duolingo, who joined in 2016 and was told his sole job was to work out how to monetize a free learning app. The lesson most worth learning from him is to keep the free product strong because free users drive growth, and to monetize only what makes the experience better.
if you only always go by positive revenue metrics, you will burn yourself to the ground.
Why Your Free Users Are Your Real Growth Engine – Cem Kansu (Duolingo CPO)
Cem Kansu: the story and the lessons
1How Cem Kansu started monetizing Duolingo
Kansu joined in 2016 after Duolingo's first business model had failed. Duolingo launched as a free website that earned money from learners translating sentences for news outlets like CNN and BuzzFeed. By 2016 it had dropped the model and was searching for the next one.
Translation was a race to the bottom on price, learners translated poorly, and fixing quality turned the company into a translation business. Resources drifted from teaching.
Luis von Ahn told Kansu his only job was to find revenue. Duolingo already had tens of millions of users. A Capital G board member told Luis the next round should be the last before he had a monetization plan.
Consumer companies like Facebook were told to grow audience first and monetize later, and Duolingo followed that until around 2016.
Streak repair was the first thing users could buy. Fixing a broken streak cost about $1 to $3, and a good share of users paid. It showed users would pay for extras and started an experiment mode: what works with the audience while retention stays high.
Luis had said publicly there would never be ads, in-app purchases or subscriptions. Duolingo built those three, in that order.
Kansu had to convince a mission-driven team that revenue serves the mission. About 60 employees were there when he joined, and many saw the company as almost a nonprofit. The message became that revenue funds the mission and lets Duolingo be more impactful.
Testing ads was fairly easy. The harder work was maximizing ad revenue while retention stayed strong.
Duolingo monetizes only the extras and keeps learning free. The mission is access to education. Rosetta Stone locked content behind a paywall, so Duolingo chose to charge only for the bells and whistles.
2How Kansu introduced ads without hurting learners
Kansu uses the app daily and measures everything before shipping ads. An app-open ad might look fine in the metrics but feel awful in use. The team tracks retention, app closes and reviews, and accepts that some curves may dip when going from zero ads to some.
He compared the anger to Uber starting to charge after being free. Complaints were expected, so the question was how many and whether retention held.
He counted complaints per million daily users to judge backlash. In 2016, complaints about the missing Finnish course were 3x the complaints about ads. Retention was neutral to slightly down in most ad tests.
Without this view, every negative comment feels like the world is falling apart. He ran the analysis at the moment of deciding whether to launch ads.
Ads appear only after a lesson, and risky categories are blocked. Duolingo blocks gambling, sketchy-looking games and anything unsafe for young audiences or brands, even at a revenue cost. It asks networks to drop formats such as ads with three exit buttons.
It started with AdMob, then added Facebook ads, Flurry and later Unity to broaden demand. Kansu says the lost revenue does not matter next to user experience and ad quality.
Duolingo's own ads sell the subscription, chosen by a machine learning model. After a lesson, a model picks between a programmatic ad and an internal ad for Super, Max or the English Test. It optimizes lifetime value, with likelihood to purchase as the goal.
If someone has seen the Super ad three times, the system backs off. Kansu would drop programmatic ads if he could.
Direct character ads are replacing programmatic ads. Direct ads run only in the US so far, with one of eight characters introducing the ad. The 2026 goal is to raise direct volume and cut programmatic by a good amount.
Scaling is slow because each advertiser needs the custom format explained.
Kansu puts more effort into the subscription than into ads. A learning app can't raise ad load the way a social feed can, because every user is there to learn. Better paid features and clearer explanation of them return more than ads do.
Communication levers include videos, purchase page design and how users enter the flow.
3How Kansu built Duolingo's subscription
Kansu read every review after ads launched and found users asking to pay. He read reviews across the App Store, Reddit and the forums. Fans wrote that they would happily give a few dollars to remove ads. Apple opened subscriptions to all apps in 2016, and Duolingo launched its own in 2017.
At the time most apps sold in-app purchases, and few outside streaming had digital subscriptions.
He proved demand with a clumsy one-off ad-free month first. It was tested at $3, $5 and $10, and everyone expected it to fail. It sold far better than expected, which justified building Duolingo Plus.
Ad-free alone felt thin, so offline learning was added. The purchase page needed a third line and nothing fit, so it said to support the mission. Kansu thinks about half of early demand came from supporters.
Unlimited hearts made the subscription worth more. Tying Plus to hearts gave users a choice about how to learn. Many users want only new content, so removing the limit lifted both conversion and retention.
Free users refill hearts by practicing old lessons, watching ads or paying. Hearts is now called energy. Kansu says a single silver bullet is rare and gains came from many feature improvements.
Kansu added free trials even though the product was already free. He resisted trials for years, sure they would flop. They worked because they remove payment hesitation, and nothing behind the paywall needed trying out.
"I think it just removes the payment hesitation."
Duolingo's purchase page spells out how the trial works. The page explains a 7-day trial with a notification on day 5. This won A/B tests and eases subscription aversion and, in many countries, fear of subscription scams.
Changing button copy to 'Try for $0.00' moved revenue more than expected. Copy went from 'subscribe' to 'start my free trial', then 'try for free', then 'try for $0.00'. It lifted both trial starts and trial-to-paid conversion.
Some results looked like measurement errors because the graphs were so green. Kansu says zero and free mean the same thing but create different emotions. About 400 tests run at once, and a good share sit on the purchase flow.
Kansu says copying Duolingo's purchase flow is safe only because each part was tested. The same testing ran across every piece of the flow, not just the copy. He cautions that copying Duolingo probably won't work for most apps.
His logic is that anything copied from Duolingo has at least been tested.
4Why Kansu protects the free product
Kansu treats free users as Duolingo's marketing engine. Word of mouth drives most of the growth, and it works whether or not the friend ever pays. About 10% of users subscribe, and subscriptions bring in more than 80% of revenue.
"free users are our marketing engine"
He calls paywalling free features the premium trap. Moving a free feature behind a paywall turns every metric green for about 6 months to a year, then stops working. It also leaves room for a rival to offer the feature free.
Kansu says charging for the great features would likely lead another app to offer them free within two years. No numbers in the source show this outweighs the revenue gained, so it rests on his judgment.
Duolingo moved Practice Hub and Explain My Answer from paid to free. Explain My Answer sat in the top tier, Max, because LLM costs were high. When costs fell and caching improved, Duolingo tested free against Max, revenue dropped, and it chose free anyway.
Kansu calls both features really additive to the experience.
Persuading teams to accept a revenue loss is harder than deciding. Leaders find it easy, but teams dislike dropping their own metric. The benefit has no metric while the cost has a very real one.
Kansu says subscription businesses that nickel-and-dime users keep revenue growing for some years.
Kansu asks whether the company is built for two years or 100. Heavy monetization earns well for two years but hurts free users. His operating principle is to take the long view.
"we cannot build a 100-year business by taking more and more from users."
5How Kansu runs experiments and settles trade-offs
Kansu prefers 3 to 6 month holdouts and avoids year-long ones. Leaderboards take 10 weeks to climb from the lowest league to the top, and a new push notification can turn negative after about 6 weeks. Duolingo once ran a one to two year ads holdout and no longer does.
Over that long, bugs get introduced and pollute the holdout group.
Big trade-offs go to senior leaders, small ones stay with teams. When a change helps learning but costs revenue or daily users, no single metric settles it. Large strategic hits go to senior people including the CEO for sign-off.
Duolingo has only proxy metrics for learning efficacy, so trade-offs need judgment.
6How Duolingo added math, music and chess under Kansu
New subjects had to pass four tests. The subject must teach well on mobile, have a large existing market, suit bite-sized lessons, and have a tenured internal builder who wants to build it. Coding was rejected as too hard today, and running as too offline.
The new subjects arrived about three to four years ago.
Kansu moved math from a separate app into the main app. Duolingo first assumed math learners would barely overlap with language learners and planned separate apps. It found users want several subjects in one place, and a separate app costs far more to grow.
He calls separate apps hard mode and one app easy mode. "why are we playing hard mode when we can play easy mode?"
Each new subject took 6 to 12 months of design and testing. The bar was parity with language courses, which have improved for 13 years. Chess was easiest because it needs one visual element, the board. Music was hardest.
The piano keyboard was turned horizontal and tilted a few degrees for touch space.
Kansu says awareness of the new subjects is still unsolved. Many users don't know math, music and chess exist, though they sit in the same course switcher as languages. Next year's goal is to make daily users aware of them.
7Cem Kansu's product beliefs
Duolingo builds for the general population, not personas. People learn for jobs, moves abroad, heritage and more, and a persona cuts off that long tail. A billion-user ambition eventually requires building for everyone.
New PMs arrive wanting persona maps with names like 'the wanderer', and the team tells them to stop. Kansu says the common PM framework is how you succeed, but for Duolingo it's how you fail.
Kansu does not take user requests at face value. Users asked for years for on-demand human tutors. Duolingo built it, and no one used it because speaking a new language to a real person was scary.
The tutors were a costly higher tier, and the team gave up after many tries at easing users in. Video call with Lily replaced it, and Kansu says nine years of Duolingo never improved his Spanish as much as that feature.
He designs for people who barely read. Attention is short, and users focus on roughly the core 30% of a screen. If you need to write a tutorial, you've already lost.
Kansu calls 'people don't read' an exaggeration, meant to push teams toward design over labels.
Kansu sees screen time, not other education apps, as the competition. Lost users mostly go to Instagram, YouTube or other fun phone use. Language learning is brain intensive, so Duolingo works hard to make it fun.
Duolingo is already bigger than any education app.
He calls AI an accelerant for Duolingo, not a threat. Duolingo uses AI to make content at higher quality and volume and to build features like conversation practice. ChatGPT can teach Spanish, but it isn't fun enough to threaten the business.
People learn for varied reasons such as heritage, which translation can't serve, much as calculators didn't end math teaching.
Why Your Free Users Are Your Real Growth Engine – Cem Kansu (Duolingo CPO)
Duolingo's CPO explains how the app moved from fully free to ads, subscriptions and a multi-subject product without hurting its free-user growth engine. He covers button copy tests like 'Try for $0.00', measuring monetization backlash, and why protecting free value is both a mission and a business moat.
Kansu joined in 2016 after Duolingo's first business model had failed. Duolingo launched as a free website that earned money from learners translating sentences for news outlets like CNN and BuzzFeed. By 2016 it had dropped the model and was searching for the next one.
Translation was a race to the bottom on price, learners translated poorly, and fixing quality turned the company into a translation business. Resources drifted from teaching.
Luis von Ahn told Kansu his only job was to find revenue. Duolingo already had tens of millions of users. A Capital G board member told Luis the next round should be the last before he had a monetization plan.
Consumer companies like Facebook were told to grow audience first and monetize later, and Duolingo followed that until around 2016.
Streak repair was the first thing users could buy. Fixing a broken streak cost about $1 to $3, and a good share of users paid. It showed users would pay for extras and started an experiment mode: what works with the audience while retention stays high.
Luis had said publicly there would never be ads, in-app purchases or subscriptions. Duolingo built those three, in that order.
Kansu had to convince a mission-driven team that revenue serves the mission. About 60 employees were there when he joined, and many saw the company as almost a nonprofit. The message became that revenue funds the mission and lets Duolingo be more impactful.
Testing ads was fairly easy. The harder work was maximizing ad revenue while retention stayed strong.
Duolingo monetizes only the extras and keeps learning free. The mission is access to education. Rosetta Stone locked content behind a paywall, so Duolingo chose to charge only for the bells and whistles.
Kansu uses the app daily and measures everything before shipping ads. An app-open ad might look fine in the metrics but feel awful in use. The team tracks retention, app closes and reviews, and accepts that some curves may dip when going from zero ads to some.
He compared the anger to Uber starting to charge after being free. Complaints were expected, so the question was how many and whether retention held.
He counted complaints per million daily users to judge backlash. In 2016, complaints about the missing Finnish course were 3x the complaints about ads. Retention was neutral to slightly down in most ad tests.
Without this view, every negative comment feels like the world is falling apart. He ran the analysis at the moment of deciding whether to launch ads.
Ads appear only after a lesson, and risky categories are blocked. Duolingo blocks gambling, sketchy-looking games and anything unsafe for young audiences or brands, even at a revenue cost. It asks networks to drop formats such as ads with three exit buttons.
It started with AdMob, then added Facebook ads, Flurry and later Unity to broaden demand. Kansu says the lost revenue does not matter next to user experience and ad quality.
Duolingo's own ads sell the subscription, chosen by a machine learning model. After a lesson, a model picks between a programmatic ad and an internal ad for Super, Max or the English Test. It optimizes lifetime value, with likelihood to purchase as the goal.
If someone has seen the Super ad three times, the system backs off. Kansu would drop programmatic ads if he could.
Direct character ads are replacing programmatic ads. Direct ads run only in the US so far, with one of eight characters introducing the ad. The 2026 goal is to raise direct volume and cut programmatic by a good amount.
Scaling is slow because each advertiser needs the custom format explained.
Kansu puts more effort into the subscription than into ads. A learning app can't raise ad load the way a social feed can, because every user is there to learn. Better paid features and clearer explanation of them return more than ads do.
Communication levers include videos, purchase page design and how users enter the flow.
Kansu read every review after ads launched and found users asking to pay. He read reviews across the App Store, Reddit and the forums. Fans wrote that they would happily give a few dollars to remove ads. Apple opened subscriptions to all apps in 2016, and Duolingo launched its own in 2017.
At the time most apps sold in-app purchases, and few outside streaming had digital subscriptions.
He proved demand with a clumsy one-off ad-free month first. It was tested at $3, $5 and $10, and everyone expected it to fail. It sold far better than expected, which justified building Duolingo Plus.
Ad-free alone felt thin, so offline learning was added. The purchase page needed a third line and nothing fit, so it said to support the mission. Kansu thinks about half of early demand came from supporters.
Unlimited hearts made the subscription worth more. Tying Plus to hearts gave users a choice about how to learn. Many users want only new content, so removing the limit lifted both conversion and retention.
Free users refill hearts by practicing old lessons, watching ads or paying. Hearts is now called energy. Kansu says a single silver bullet is rare and gains came from many feature improvements.
Kansu added free trials even though the product was already free. He resisted trials for years, sure they would flop. They worked because they remove payment hesitation, and nothing behind the paywall needed trying out.
"I think it just removes the payment hesitation."
Duolingo's purchase page spells out how the trial works. The page explains a 7-day trial with a notification on day 5. This won A/B tests and eases subscription aversion and, in many countries, fear of subscription scams.
Changing button copy to 'Try for $0.00' moved revenue more than expected. Copy went from 'subscribe' to 'start my free trial', then 'try for free', then 'try for $0.00'. It lifted both trial starts and trial-to-paid conversion.
Some results looked like measurement errors because the graphs were so green. Kansu says zero and free mean the same thing but create different emotions. About 400 tests run at once, and a good share sit on the purchase flow.
Kansu says copying Duolingo's purchase flow is safe only because each part was tested. The same testing ran across every piece of the flow, not just the copy. He cautions that copying Duolingo probably won't work for most apps.
His logic is that anything copied from Duolingo has at least been tested.
Kansu treats free users as Duolingo's marketing engine. Word of mouth drives most of the growth, and it works whether or not the friend ever pays. About 10% of users subscribe, and subscriptions bring in more than 80% of revenue.
"free users are our marketing engine"
He calls paywalling free features the premium trap. Moving a free feature behind a paywall turns every metric green for about 6 months to a year, then stops working. It also leaves room for a rival to offer the feature free.
Kansu says charging for the great features would likely lead another app to offer them free within two years. No numbers in the source show this outweighs the revenue gained, so it rests on his judgment.
Duolingo moved Practice Hub and Explain My Answer from paid to free. Explain My Answer sat in the top tier, Max, because LLM costs were high. When costs fell and caching improved, Duolingo tested free against Max, revenue dropped, and it chose free anyway.
Kansu calls both features really additive to the experience.
Persuading teams to accept a revenue loss is harder than deciding. Leaders find it easy, but teams dislike dropping their own metric. The benefit has no metric while the cost has a very real one.
Kansu says subscription businesses that nickel-and-dime users keep revenue growing for some years.
Kansu asks whether the company is built for two years or 100. Heavy monetization earns well for two years but hurts free users. His operating principle is to take the long view.
"we cannot build a 100-year business by taking more and more from users."
Kansu prefers 3 to 6 month holdouts and avoids year-long ones. Leaderboards take 10 weeks to climb from the lowest league to the top, and a new push notification can turn negative after about 6 weeks. Duolingo once ran a one to two year ads holdout and no longer does.
Over that long, bugs get introduced and pollute the holdout group.
Big trade-offs go to senior leaders, small ones stay with teams. When a change helps learning but costs revenue or daily users, no single metric settles it. Large strategic hits go to senior people including the CEO for sign-off.
Duolingo has only proxy metrics for learning efficacy, so trade-offs need judgment.
New subjects had to pass four tests. The subject must teach well on mobile, have a large existing market, suit bite-sized lessons, and have a tenured internal builder who wants to build it. Coding was rejected as too hard today, and running as too offline.
The new subjects arrived about three to four years ago.
Kansu moved math from a separate app into the main app. Duolingo first assumed math learners would barely overlap with language learners and planned separate apps. It found users want several subjects in one place, and a separate app costs far more to grow.
He calls separate apps hard mode and one app easy mode. "why are we playing hard mode when we can play easy mode?"
Each new subject took 6 to 12 months of design and testing. The bar was parity with language courses, which have improved for 13 years. Chess was easiest because it needs one visual element, the board. Music was hardest.
The piano keyboard was turned horizontal and tilted a few degrees for touch space.
Kansu says awareness of the new subjects is still unsolved. Many users don't know math, music and chess exist, though they sit in the same course switcher as languages. Next year's goal is to make daily users aware of them.
Duolingo builds for the general population, not personas. People learn for jobs, moves abroad, heritage and more, and a persona cuts off that long tail. A billion-user ambition eventually requires building for everyone.
New PMs arrive wanting persona maps with names like 'the wanderer', and the team tells them to stop. Kansu says the common PM framework is how you succeed, but for Duolingo it's how you fail.
Kansu does not take user requests at face value. Users asked for years for on-demand human tutors. Duolingo built it, and no one used it because speaking a new language to a real person was scary.
The tutors were a costly higher tier, and the team gave up after many tries at easing users in. Video call with Lily replaced it, and Kansu says nine years of Duolingo never improved his Spanish as much as that feature.
He designs for people who barely read. Attention is short, and users focus on roughly the core 30% of a screen. If you need to write a tutorial, you've already lost.
Kansu calls 'people don't read' an exaggeration, meant to push teams toward design over labels.
Kansu sees screen time, not other education apps, as the competition. Lost users mostly go to Instagram, YouTube or other fun phone use. Language learning is brain intensive, so Duolingo works hard to make it fun.
Duolingo is already bigger than any education app.
He calls AI an accelerant for Duolingo, not a threat. Duolingo uses AI to make content at higher quality and volume and to build features like conversation practice. ChatGPT can teach Spanish, but it isn't fun enough to threaten the business.
People learn for varied reasons such as heritage, which translation can't serve, much as calculators didn't end math teaching.
More from Duolingo
GuestAnmol Tiwari
Director of Product at Duolingo
Duolingo1 source · 15 tips
GuestGina Gotthilf
Co-founder and COO of Latitud, former head of growth and marketing at Duolingo
Duolingo
1 source · 18 tips
GuestJackson Shuttleworth
Group Product Manager leading the retention team at Duolingo
Duolingo1 source · 21 tips
GuestNatalia Castillejo
Director of Product, leads monetization at Duolingo
Duolingo
1 source · 16 tips