Jackson Shuttleworth
Group Product Manager leading the retention team at Duolingo
Where to hear and read about Jackson Shuttleworth1 episode or test
Jackson Shuttleworth is the Group Product Manager who leads the retention team at Duolingo, which owns the streak, the company's biggest growth lever after the lessons themselves. The most useful thing to learn from him is how to make a daily habit simple to grasp, forgiving early and worth protecting, by testing small versions constantly.
Streaks work best when they're sitting on top of an app that users care about.
Behind the product: Duolingo streaks | Jackson Shuttleworth (Group PM, Retention Team)
Jackson Shuttleworth: the story and the lessons
1Jackson Shuttleworth and the Duolingo streak
Shuttleworth leads the retention team that owns Duolingo's streak. He is a group product manager at a company worth about $14 billion, which has roughly doubled in value in six months. Retention means getting a user to come back the next day.
When new team members join, he spends an hour on the streak and covers about 30% of what he wants to share.
The streak is Duolingo's biggest growth lever after the lessons. More than 9 million users hold a streak of a year or longer, and many notifications work because they reference the streak.
He calls it one of the company's "golden geese" and says it ships a disgusting amount of DAUs.
The streak began as an XP goal and became one lesson a day. Early streaks required hitting a personal goal, such as 10 or 50 XP, which suited the tech-forward audience that arrived after Luis von Ahn's TED Talk. Moving to one lesson a day was a huge DAU win, and users did not care less about the streak.
"The unit of use for Duolingo is doing a lesson."
The team has run over 600 streak experiments in four years. That is about one every other day, and about half of them lose. The streak is still far from optimized.
The prior retention PM said the team wasn't 30% of the way there. Four years on, Shuttleworth still agrees. He has a 400-day streak himself and lost and restarted streaks many times before it.
2How Shuttleworth designs a streak people understand and commit to
He rejected a streak that counted just one exercise. Extending the streak with a single exercise caught users who did one or two questions and left. DAUs moved not one bit and the test was shut down.
The test followed a funnel insight that many users started lessons and never finished. He would run it more carefully now, because the streak holds the best-retaining users.
He puts the most help into a user's first seven days. Each added day from one through seven gives a big retention jump, then the curve flattens as loss aversion kicks in. Duolingo runs more experiments than average on the zero to seven day experience.
"Once you get to seven days, loss aversion kicks in, and you retain."
His first Duolingo win was an eight-word line explaining the streak. The copy said, roughly, start a day to extend your streak, but miss a day and it resets. It was worth over 10,000 DAUs about four years ago and was a top three win on current user retention rate (CURR).
Many users do not understand how a streak works. One user's mother was surprised to return and find her streak intact. The calendar was also redesigned to look more like a calendar, so the daily rule is clear.
He added a goal screen where users say yes or no. Users were told they were seven times more likely to finish the course with a 30-day streak, then offered goal lengths and a no button with an easier fallback goal. The opt-out button alone was almost as big a win as the fallback goal.
Preselecting a harder goal lost, because choosing was where the engagement came from. The seven times line was borrowed from a monetization message about Plus subscribers.
He tests copy constantly because wording alone can win big. Changing the goal button from Continue to Commit To My Goal was a massive win. Translating a few strings is cheap, so one batch of ideas covers every interface language.
The reviewer often says he doesn't think a test will win and then lets it run. He is often right, but the cost is so low it doesn't matter.
He leads the streak design with the number, not the flame. The flame metaphor did not resonate in user research in India. The new screen uses an odometer animation that shows the count rising, then "seven day streak".
Delight such as the Phoenix Duo is kept for deeper streaks. A one-day user only needs to understand the feature.
He holds new streak concepts back until the streak is established. Perfect Streak appears only after seven days, because two streak numbers would compete. Showing it earlier was tested and pretty universally lost.
Too many concepts too early add cognitive load.
3How Shuttleworth tunes flexibility, rewards and reminders
He gives every new streak two free freezes. Early streaks are easy to let die, and users need about seven days for the habit to lock in. Granting two freezes at streak start was one of the biggest streak freeze wins.
Freezes used to be bought with gems so they would feel earned. Flexibility should shrink as the streak gets longer, because long-streak users rarely need days off.
He caps freezes at two because more trains users to take days off. Going from one freeze to two raised DAU and weekly active user return rate but hurt CURR. Three freezes was no better than two.
Longer breaks make a return less likely. After three days off, even day four becomes unlikely.
He replaced paid streak repair with earning a streak back. Earn Back asks for a few lessons soon after the loss, where Streak Repair charged gems. It was a big retention winner and now exists as a free option, at a small cost to revenue.
The streak feels earned, so it isn't cheapened. Cheapening the streak almost always lifts engagement up to a point, and where that point is isn't clear.
He wants a keeper guarding the sanctity of the streak. More freezes and easier extension nearly always win engagement, but past some point users decide the streak means nothing. That would hit the 9 million users on year-long streaks.
A co-lead and the founder act as guardians because this is hard to set as policy. "You kind of got to hold the line at some point."
He balances flexibility with a purely visual reward for perfection. After a few days without a freeze, the streak turns gold and the calendar bar looks nicer. There is no prize, yet users care about it deeply.
An employee was upset after losing a four-month perfect streak to a date-line backend quirk.
He uses animation and haptics to make users pause on the streak. Haptic patterns and animations as the streak extends keep winning tests, because users enjoy the moment and care more.
Pauses should be kept to a few big screens. Haptic work moved from engineers to product designers to an animation studio in Detroit.
He sends the practice reminder 23.5 hours after the last lesson. Users used to pick a reminder time, but plans change. When someone practiced the day before predicted best when they would practice next, and more complex logic was tried many times and always lost.
Revealed behavior beats stated behavior.
He sends a last-chance streak saver at 10 PM. Users who have not extended their streak are warned it is their last chance. It reads as helpful, not spammy, because people attach positive feelings to the streak.
Notification load should follow how users perceive the feature the message is tied to.
4How Shuttleworth's team runs hundreds of experiments
His team owns a metric, not a feature. Duolingo focuses on CURR and DAU. Retention owns the streak only because it drives CURR best, and other teams, such as leaderboards, pitch ideas into a multi-quarter streak roadmap.
Jira automation and detailed roadmaps plan engineering, design and product bandwidth months out, because lost cycles are lost opportunity.
He ships the lightest first version that tests the core hypothesis. Streak goal started as a single message and grew over many iterations. Simple versions ship, get designed and get approved faster, and bells and whistles can make a test win for the wrong reason.
"We really resist the urge to do the big V1."
He shuts down neutral experiments that add complexity. Added UI brings cognitive load, so a neutral result usually means shutting the test down. The exception is real conviction that it opens a platform for later wins, and then he advises building that into V1.
The team also periodically resets a feature's UI to something just as good but simpler, to avoid a kitchen sink product.
He killed a winning test that cluttered the lesson screen. Showing XP tick up during a lesson won, but the CEO asked where the idea would go next and there was no answer. The lesson stays a learning sanctuary.
The team judged it an easy engagement win that would get in the way over time. Wins are weighed against product direction, not only the metric.
He starts experiments from hypotheses borrowed from past wins and games. Monetization and retention teams share what worked, and mobile games such as Royal Match and the Pokemon trading card game are studied for ideas. Belief in a hypothesis sets how big the experiment is.
Tests need a hypothesis and intent, or the team gets stuck in local maxima. Senior reviews, including the CEO, hold the bar on coherence.
He says a streak only works on an app users already want. Streaks use loss aversion, but only multiply an experience people like. Without that, a streak distracts from the real problem.
Shuttleworth thinks almost any app can use one if the streak fits its natural rhythm. Peloton uses weekly streaks. His rule: figure out your usage pattern as a user, then build the streak around it.
He makes the streak highly visible so users learn it matters. Board member Bing Gordon observed that users care because Duolingo cares: a big animated streak screen follows each session and messages reference it.
A streak hidden in a corner of the app likely won't matter to users.
Behind the product: Duolingo streaks | Jackson Shuttleworth (Group PM, Retention Team)
Duolingo's retention lead walks through four years and over 600 experiments on the streak feature, covering unit of use, goal setting, streak freezes, copy, notifications and how the team runs experiments. The streak is described as Duolingo's biggest growth lever for daily active users, with over 9 million users holding a year plus streak.
Shuttleworth leads the retention team that owns Duolingo's streak. He is a group product manager at a company worth about $14 billion, which has roughly doubled in value in six months. Retention means getting a user to come back the next day.
When new team members join, he spends an hour on the streak and covers about 30% of what he wants to share.
The streak is Duolingo's biggest growth lever after the lessons. More than 9 million users hold a streak of a year or longer, and many notifications work because they reference the streak.
He calls it one of the company's "golden geese" and says it ships a disgusting amount of DAUs.
The streak began as an XP goal and became one lesson a day. Early streaks required hitting a personal goal, such as 10 or 50 XP, which suited the tech-forward audience that arrived after Luis von Ahn's TED Talk. Moving to one lesson a day was a huge DAU win, and users did not care less about the streak.
"The unit of use for Duolingo is doing a lesson."
The team has run over 600 streak experiments in four years. That is about one every other day, and about half of them lose. The streak is still far from optimized.
The prior retention PM said the team wasn't 30% of the way there. Four years on, Shuttleworth still agrees. He has a 400-day streak himself and lost and restarted streaks many times before it.
He rejected a streak that counted just one exercise. Extending the streak with a single exercise caught users who did one or two questions and left. DAUs moved not one bit and the test was shut down.
The test followed a funnel insight that many users started lessons and never finished. He would run it more carefully now, because the streak holds the best-retaining users.
He puts the most help into a user's first seven days. Each added day from one through seven gives a big retention jump, then the curve flattens as loss aversion kicks in. Duolingo runs more experiments than average on the zero to seven day experience.
"Once you get to seven days, loss aversion kicks in, and you retain."
His first Duolingo win was an eight-word line explaining the streak. The copy said, roughly, start a day to extend your streak, but miss a day and it resets. It was worth over 10,000 DAUs about four years ago and was a top three win on current user retention rate (CURR).
Many users do not understand how a streak works. One user's mother was surprised to return and find her streak intact. The calendar was also redesigned to look more like a calendar, so the daily rule is clear.
He added a goal screen where users say yes or no. Users were told they were seven times more likely to finish the course with a 30-day streak, then offered goal lengths and a no button with an easier fallback goal. The opt-out button alone was almost as big a win as the fallback goal.
Preselecting a harder goal lost, because choosing was where the engagement came from. The seven times line was borrowed from a monetization message about Plus subscribers.
He tests copy constantly because wording alone can win big. Changing the goal button from Continue to Commit To My Goal was a massive win. Translating a few strings is cheap, so one batch of ideas covers every interface language.
The reviewer often says he doesn't think a test will win and then lets it run. He is often right, but the cost is so low it doesn't matter.
He leads the streak design with the number, not the flame. The flame metaphor did not resonate in user research in India. The new screen uses an odometer animation that shows the count rising, then "seven day streak".
Delight such as the Phoenix Duo is kept for deeper streaks. A one-day user only needs to understand the feature.
He holds new streak concepts back until the streak is established. Perfect Streak appears only after seven days, because two streak numbers would compete. Showing it earlier was tested and pretty universally lost.
Too many concepts too early add cognitive load.
He gives every new streak two free freezes. Early streaks are easy to let die, and users need about seven days for the habit to lock in. Granting two freezes at streak start was one of the biggest streak freeze wins.
Freezes used to be bought with gems so they would feel earned. Flexibility should shrink as the streak gets longer, because long-streak users rarely need days off.
He caps freezes at two because more trains users to take days off. Going from one freeze to two raised DAU and weekly active user return rate but hurt CURR. Three freezes was no better than two.
Longer breaks make a return less likely. After three days off, even day four becomes unlikely.
He replaced paid streak repair with earning a streak back. Earn Back asks for a few lessons soon after the loss, where Streak Repair charged gems. It was a big retention winner and now exists as a free option, at a small cost to revenue.
The streak feels earned, so it isn't cheapened. Cheapening the streak almost always lifts engagement up to a point, and where that point is isn't clear.
He wants a keeper guarding the sanctity of the streak. More freezes and easier extension nearly always win engagement, but past some point users decide the streak means nothing. That would hit the 9 million users on year-long streaks.
A co-lead and the founder act as guardians because this is hard to set as policy. "You kind of got to hold the line at some point."
He balances flexibility with a purely visual reward for perfection. After a few days without a freeze, the streak turns gold and the calendar bar looks nicer. There is no prize, yet users care about it deeply.
An employee was upset after losing a four-month perfect streak to a date-line backend quirk.
He uses animation and haptics to make users pause on the streak. Haptic patterns and animations as the streak extends keep winning tests, because users enjoy the moment and care more.
Pauses should be kept to a few big screens. Haptic work moved from engineers to product designers to an animation studio in Detroit.
He sends the practice reminder 23.5 hours after the last lesson. Users used to pick a reminder time, but plans change. When someone practiced the day before predicted best when they would practice next, and more complex logic was tried many times and always lost.
Revealed behavior beats stated behavior.
He sends a last-chance streak saver at 10 PM. Users who have not extended their streak are warned it is their last chance. It reads as helpful, not spammy, because people attach positive feelings to the streak.
Notification load should follow how users perceive the feature the message is tied to.
His team owns a metric, not a feature. Duolingo focuses on CURR and DAU. Retention owns the streak only because it drives CURR best, and other teams, such as leaderboards, pitch ideas into a multi-quarter streak roadmap.
Jira automation and detailed roadmaps plan engineering, design and product bandwidth months out, because lost cycles are lost opportunity.
He ships the lightest first version that tests the core hypothesis. Streak goal started as a single message and grew over many iterations. Simple versions ship, get designed and get approved faster, and bells and whistles can make a test win for the wrong reason.
"We really resist the urge to do the big V1."
He shuts down neutral experiments that add complexity. Added UI brings cognitive load, so a neutral result usually means shutting the test down. The exception is real conviction that it opens a platform for later wins, and then he advises building that into V1.
The team also periodically resets a feature's UI to something just as good but simpler, to avoid a kitchen sink product.
He killed a winning test that cluttered the lesson screen. Showing XP tick up during a lesson won, but the CEO asked where the idea would go next and there was no answer. The lesson stays a learning sanctuary.
The team judged it an easy engagement win that would get in the way over time. Wins are weighed against product direction, not only the metric.
He starts experiments from hypotheses borrowed from past wins and games. Monetization and retention teams share what worked, and mobile games such as Royal Match and the Pokemon trading card game are studied for ideas. Belief in a hypothesis sets how big the experiment is.
Tests need a hypothesis and intent, or the team gets stuck in local maxima. Senior reviews, including the CEO, hold the bar on coherence.
He says a streak only works on an app users already want. Streaks use loss aversion, but only multiply an experience people like. Without that, a streak distracts from the real problem.
Shuttleworth thinks almost any app can use one if the streak fits its natural rhythm. Peloton uses weekly streaks. His rule: figure out your usage pattern as a user, then build the streak around it.
He makes the streak highly visible so users learn it matters. Board member Bing Gordon observed that users care because Duolingo cares: a big animated streak screen follows each session and messages reference it.
A streak hidden in a corner of the app likely won't matter to users.
More from Duolingo
GuestAnmol Tiwari
Director of Product at Duolingo
Duolingo
1 source · 15 tips
GuestCem Kansu
Chief Product Officer at Duolingo
Duolingo1 source · 19 tips
GuestGina Gotthilf
Co-founder and COO of Latitud, former head of growth and marketing at Duolingo
Duolingo1 source · 18 tips
GuestNatalia Castillejo
Director of Product, leads monetization at Duolingo
Duolingo
1 source · 16 tips