Growth
Duolingo — Flock
Learning alone is lonely — and lonely churns. Flock is an async friend-league: small groups compete across any subject on normalized weekly scores, turning friends into the acquisition engine.
Try the prototypeDuolingo's North Star
100 million daily users by 2028.
58M to 100M by 2028. Free DAU is the lever. A bigger top of funnel and a better free experience drive more paid conversions. The fastest path is virality.
Personas · Three Duolingo users
Who we build for. And who we don’t.
All three are already on Duolingo. We prioritize Maya, the power user. If we make her invite-worthy, the rest of the funnel follows.
The Problem
Learning, quietly, gets lonely.
Maya is putting in the work. No one sees it. The streak counts, but no one’s counting with her. Meanwhile her friends keep saying they want to pick up Spanish, chess, music. They never quite start. The push to act isn’t there.
Insight
Duolingo’s own data: a group of friends is the biggest lever.
Learners with friends finish courses 5.6× more often. Shared streaks lift daily completion +22%. The streak already works. What’s missing is the study group wrapped around it. The motivation Maya is losing is the exact gap a cohort closes.
Idea · Mental model
Group accountability for the actual learning habit.
Duolingo already has a primitive: Chess invites.But it’s synchronous, 1 on 1, chess only, single match. Flock is the async, group, all week, any subject version of the same instinct.
User Journey
Two paths into the same Flock.
Maya creates. Marcus joins from outside the app. Both paths have to feel obvious.
Flock's Nest · The centerpiece
The Nest visualizes your Flock learning together.
Prototype
From entry to recap. Both sides of the loop.
Interactive prototype for illustrative purposes. How a Flock could work.
Interactive prototype
Maya
maya_learns · joined 2022
127
day streak
24,310
total XP
Streaks & Flock
Learn with up to 5 friends
Step 1 · Entry: profile
Flock lives where the streak lives — the emotional center of the app, one row below Maya's 127-day flame.
The acquisition loop
Every Flock member is a distribution node.
Marcus joined for Maya. Now he’s pulling his own friends in for chess. Flock makes Duolingo social. And turns Duolingo kids into a growth engine.
Loop metrics · The funnel
1. Creation rate
% of DAU creating a Flock
2. Viral conversion
Invites sent → Link clicks → New Installs
These two steps prove the loop closes. We track new user attribution directly back to the invite link.
Success metrics
The metric that only moves when all three win.
Only moves when a learner stays, a friend joins, or an inactive friend reactivates. Tied directly to Duolingo’s 100M-by-2028 goal.
Social pull retains the bottom half of the funnel that solo loses.
First release & experimentation
Ship the smallest thing that proves the loop.
- Create-a-Flock from profile + feed
- Invite link · any share channel
- Daily shared ring on Home
- Milestones + emoji reactions
- ≥10% of cohort create a Flock + add a friend
- D30 retention ≥ solo control
- Time spent / DAU flat or up
- Track: invite → install (directional)
Risk register
Six risks I’d name. Six mitigations.
Why Duolingo
The pieces are already on the home screen. Duolingo turns them into delight.
Built with AI
AI in the product. AI in the pitch.
The case asked how generative AI shaped this. Short answer: load-bearing in the product (Lily-led practice, cross-subject scoring, weekly recap), and the entire interactive deck you just scrolled was AI-coded in 8 hours.
The result isn’t a static slide deck. It’s a fully coded, interactive React application using real Duolingo design tokens, components, and animations.
FAQ
Questions I’d ask myself. So you don’t have to.
›Why hasn't Duolingo built this already?
Friends Quests and Shared Streak exist but assume aligned subject and treat social as accessory. Flock inverts that. Social is the entry point, subject is the variable.
›Can users be part of multiple Flocks?
To start, users can only be part of one Flock. Focus is key. On a weekly basis, that Flock has an opportunity to disperse unless the creator sets it to auto-continue. Users can also choose to gracefully exit a Flock if they'd like.
›Who can be invited to a Flock?
Creators can directly invite users they mutually follow on the Duolingo social graph. For external friends, the creator can share an invite link that drops them straight into the Flock after install.
›Are AI-powered insights available to everyone?
Flock itself is free, but we can treat the AI-powered cross-subject weekly recap and advanced Nest insights as a showcase feature exclusive to Flock creators who have Duolingo Max.
›Won't the Flock card clutter the Feed?
Only during the first entry before a Flock is created. Once created, the Flock stays cleanly on top of the Feed (and Home) so it's ambient but not obtrusive.
Feed
You + 4 friends
Sarah · 22h
Reached 100 day streak!
›What's the smallest thing you'd ship in 6 weeks?
The MVP: a daily shared ring on Home + milestone reactions, gated to an alpha cohort of >100-day streak users. No full Nest, no AI report, no Practice Together. We test the two beliefs the whole idea rests on. That a shared ring pulls people back daily, and that reactions close the recognition gap.
›What if I don't want my Flock to see when I'm online?
When We Overlap is opt-out per Flock from Settings. Turn it off and your active hours stay private. Your Flock still sees the daily ring and focus minutes contributions, just not when you're logged in or studying. The default leans social because that's the point, but no one is forced to share their schedule.
👉 Click a question to expand
Appendix · Data I'd want before scaling
The numbers that would tell me if Flock can land.
Before committing to MVP, I’d want a week with the data team digging into these. The Chess invite is the closest existing primitive. Its baseline is the single most informative input we don’t yet have.
Chess invite baseline
- Of users in a Chess course, what % send at least one match invite per week?
- Of those invites, what % open the link? What % of openers were not already on Duolingo?
- What does the onboarding funnel look like for an invitee whose first surface is a Chess invite vs the standard app-store install?
- D1 / D7 / D30 retention of invite-driven installs vs paid-acquisition installs.
Friend Streaks · social graph
- Median number of mutual-on-Duolingo friends per DAU. Tail of the distribution matters most. How many users have zero?
- Of users with ≥2 Friend Streaks, what's their D30 vs users with 0?
- Drop-off curve when one party of a Friend Streak lapses.
Feed · what gets reacted to
- Reaction-per-post rate on the existing Feed. Bumps from streak milestones vs lesson completions vs leaderboard wins.
- What % of Feed views convert to any in-app action (open lesson, react, message)?
Power users · Maya check
- Among 100+ day streak holders, what share send invites of any type today? Self-reported reason for not inviting more.
- Cross-subject behavior: how many DAU touch ≥2 of Duolingo's four subjects in a month?
Appendix · Future vision
The endgame: learn anything on Duolingo.
Four subjects today. Many more tomorrow. AI generates a personalized, gamified course on any topic. And the spark for what to learn next? Your Flock.Friends already know things you don’t.
Speculative. Years out, not roadmap. AI opens the door to any subject, and your Flock becomes the most relevant recommendation engine Duolingo has. You learn what your friends already love.
The problem
Solo streaks are fragile — learning alone is lonely, and lonely churns. Duolingo's social hooks are too shallow for the friends who'd actually keep you coming back.
Metric that matters
Primary: D30 retention of users in an active flock vs. control. Leading: friend-invite conversion (friends as the acquisition engine). Guardrail: pressure-driven churn from competition.
Key tradeoff
Async normalized-score leagues across any subject vs. real-time head-to-head. Chose async + normalization so mixed-level friends compete fairly — cut live battles to keep it low-pressure and inclusive.
Biggest risk
Competition can demotivate as easily as it motivates, and small/empty flocks feel dead. Cold-start: you need 2–3 active friends before the loop works.