Why leaderboards kill retention (and how to fix it)

Your newest user is always last

Picture Dana. She installs a running app on Monday, jogs 3 km, feels great, and taps the leaderboard to see how she did. She is 48,213th. Rank one has run 1,912 km this year.

It took the app four seconds to tell her: you are not a runner here. She never opens it again. Your dashboard shows nothing wrong, because people who feel out of place don't complain. They just stop coming back.

This is the shape of every open leaderboard. There is one winner, and everyone else gets a reminder of where they are not. The people most exposed are the ones you can least afford to lose: new users in their first week, with no history to hide behind.

Comparison is fuel, until the gap gets too wide

In 1954 the psychologist Leon Festinger argued that people judge their own ability by looking at other people, and that they pick people who are close to them. A rival slightly ahead of you is fuel: you can see how to catch up. A rival 10,000 places ahead is not a goal. It is a wall.

There is a second layer. Self-determination theory says people stay with an activity when it satisfies three needs: competence, autonomy and belonging. A rank is a public verdict on competence. When the verdict is "48,213th", the need isn't just unmet, it's graded in front of you, every time you open the app.

Four studies that should make you nervous

Add one cautionary tale about what happens when the rank is all that matters. In 2016, US regulators fined Wells Fargo $185 million after finding that staff had opened more than 2 million bank and credit card accounts that may not have been authorised, under aggressive sales targets and bonuses. About 5,300 employees were fired. That wasn't a leaderboard in an app, but it's the same law: what you rank is what you get, and people will optimise the number, not the thing the number was meant to stand for.

But Duolingo has a leaderboard. And it works.

Here is the twist, and the reason the title of this tip is a little unfair. Jorge Mazal, Duolingo's former chief product officer, wrote that leaderboards were the retention team's first breakthrough. They raised overall learning time by 17% and tripled the number of highly engaged learners, people studying an hour a day, five days a week. Standard retention metrics improved with statistical significance.

Figures from Duolingo's own account of the launch. Not independently audited.

So why didn't it blow up Dana's first week? Read how they built it. You are matched with people whose engagement last week was similar to yours. You are opted in automatically, no setup. Each week the top finishers move up a league and the board resets. Mazal said they deliberately made it "as casual and frictionless as possible". Dana would never see a 1,912 km runner. She would see about thirty people who are roughly as casual as she is.

Leaderboards don't kill retention. Unfair comparisons do.

That's the real finding, and the MovieLens experiment backs it up with a controlled test: when researchers randomly changed who users were compared against, larger groups lowered effort, and seeing peers who contributed less raised it. They also noted that earlier claims that "low rank demotivates" were only correlational. Who sits next to you on the board matters more than the board itself.

Six fixes, in the order we'd try them

1. Shrink the arena

   Group users into small cohorts of similar recent activity, around thirty, not everyone. The N-effect says a smaller crowd makes every name matter more; the Duolingo design says similar activity keeps the gap catchable.

2. Show the neighbourhood, not the summit

   Two people above you, two below, and a message with a number in it: "4 km to pass Ben". Keep the full table one tap away for those who want it. Most users should never have to look at the top.

3. Reset often

   A weekly board gives everyone a fresh start every Monday. An all-time board gets worse for new users every day it stays on.

4. Rank consistency, not volume

   A volume board rewards whoever has the most free time. In 2020 Strava added Local Legends for exactly this reason: the title goes to whoever completes a segment most often in the last 90 days, regardless of speed. A commuter with a steady habit can win; a pro cyclist who rode it once can't.

5. Let people compete with yesterday's self

   Show a ghost line of last week's you. It is the one rival who is always the right distance away. And it works with an effect called endowed progress: in a car-wash loyalty experiment, cards that came with 2 free stamps pre-filled (10 slots) were completed by 34% of customers; blank 8-slot cards by 19%. People push harder on a journey that has already started.

6. Give users an off-switch

   Autonomy is one of the three needs from earlier. Duolingo lets learners opt out. A board you chose to look at feels like a game; a board you can't escape feels like surveillance.

A ten-minute audit

1. Open the board as a day-one user. Make a fresh account, do one small thing, and look at what you see. If the first number is bigger than your phone screen, you have your answer.

2. Split retention by rank. Pull day-7 and day-30 retention for each rank quartile instead of one average (Figure 5). If the bottom quartile falls off a cliff, the board is the suspect.

3. Ask what the bottom half sees. More than half of your users are, by definition, below the median. If the board has nothing useful for them, it's only talking to the top.

4. Test, don't believe us. Hide the board for a random 10% of new users for two weeks and compare their retention.

Where the evidence is thinner than the headline

Honesty check. The literature is mixed. One study (Mekler and colleagues) found points, levels and leaderboards raised how much people did without hurting their intrinsic motivation. The famous "rewards undermine motivation" result (Deci and colleagues) is still argued about. The Hanus and Fox study isn't randomized, the furniture study is a working paper, and Duolingo's numbers come from Duolingo. And a board can be great for the motivated top slice of your users.

What survives all of it is a narrow, practical rule: a leaderboard is a feature for the people near the top, and you have to design it on purpose for everyone else.

Sources

1. Mazal, J., "How Duolingo reignited user growth", Lenny's Newsletter.
2. Leung et al., "How do one's peers on a leaderboard affect oneself?", CHI 2019. ACM Digital Library.
3. Brown, J., "Quitters Never Win: The (Adverse) Incentive Effects of Competing with Superstars", Journal of Political Economy 119(5), 2011.
4. Garcia, S. M. & Tor, A., "The N-effect: More competitors, less competition", Psychological Science, 2009. Abstract.
5. Barankay, I., "Rank Incentives: Evidence from a Randomized Workplace Experiment". Penn repository · Wharton summary.
6. Hanus, M. D. & Fox, J., "Assessing the effects of gamification in the classroom", Computers & Education 80, 2015.
7. Mekler et al., "Towards understanding the effects of individual gamification elements on intrinsic motivation and performance", Computers in Human Behavior 71, 2017.
8. Deci, Koestner & Ryan, "A meta-analytic review of experiments examining the effects of extrinsic rewards on intrinsic motivation", Psychological Bulletin 125, 1999.
9. Nunes, J. & Drèze, X., "The endowed progress effect", Journal of Consumer Research 32(4), 2006.
10. Festinger, L., "A theory of social comparison processes", Human Relations, 1954.
11. Strava, Local Legends.
12. Wells Fargo fine, September 2016: WAMC / NPR.