We grew a consumer social app to 600k users without paid ads

The hardest period was Q2 2024. We had crossed 100k users. The inevitable consumer-app dynamic kicked in: D30 retention, which had been around 38% in the first cohort, drifted down to 22% over two months. Investors started asking pointed questions about whether we were a fad.

I didn't know. Honestly. The data was ambiguous. Wren and I spent three weeks doing user interviews — about forty of them, with both retained users and churned users. The signal in the interviews was stark and surprising. Churned users almost universally said the same thing: "My friends weren't on it." Retained users said the opposite: "All my friends are on it."

§Pausing engineering for ten weeks

Which meant retention was a network density problem, not a product problem. A user only stuck if at least three of their existing close contacts were also active. Below that threshold, the app felt empty regardless of how good the product was.

This insight forced an organizational decision I didn't want to make. We'd been planning to ramp engineering and ship faster. Instead we paused new feature work for ten weeks and rebuilt onboarding to optimize for getting at least three friends in within the first 48 hours of signup. The team hated it. Two engineers were openly skeptical. One quit.

The rebuild moved D30 retention from 22% back to 41% within sixty days. The engineer who quit, to his credit, emailed me later and said it had been the right call.

The structural breakthrough was an org-design decision, not a feature. In November 2024 we capped the engineering team at eight people. Permanently. We agreed as a team that growth in headcount would happen in operations, content moderation, and support — not in product engineering.

The reasoning: every consumer social app I'd ever watched die from feature bloat had done so because a growing engineering team needed something to ship every quarter. Capped headcount means capped feature surface area. Capped feature surface area means the product stays the thing users originally fell in love with.

Within a year of that cap, our shipped feature count went down 70%. Our D7 retention went up 14 percentage points. Wren and I became more selective about what we built. Engineers we hired specifically wanted the constraint — we lost some candidates because of it, but the ones we kept were better.

As of April 2026 we're at 600k users, ~$1.4M ARR in subscriptions, gross margin 87%, and the team is 14 people total. We're profitable on a contribution basis since November 2025. We've turned down two acquisition conversations.

§From 600k to 2.1M and the things that nearly broke us

The fifteen months after we crossed 600k MAU — January 2025 to April 2026 — were when consumer social got real. We hit 2.1M MAU. We did it without TikTok ads, without Meta ads, and without growth-loop tricks like fake social proof. The mechanic was unglamorous: every two-week sprint, we shipped one feature whose only job was to make a user invite a real friend within ninety seconds of signup. The "ninety-second activation" became our north-star metric, replacing DAU.

We hired Mei as Head of Trust & Safety in March 2025 — she'd run policy at Discord before us — and her existence prevented at least four near-platform-death events I can name (a coordinated impersonation campaign, a brigading incident around a public figure, a CSAM detection escalation, and a regulatory inquiry from Ofcom under the UK Online Safety Act). Without her hire we would have been Clubhouse: a hot summer and then a long winter. Our content-moderation stack runs on Hive, Sift Trust (no relation to us), and a custom toxicity classifier on top of an open weights Llama 3.1 fine-tune. The "narrow your channels, deepen retention" pattern overlaps with the open-source database story in this cohort — both are organic-only growth bets that worked because we refused the easy paid path.

1. 1
   Consumer social retention is a network density problem first.

   No amount of feature work fixes a user whose three closest friends aren't on the app. Optimize onboarding for friend-density before anything else.

2. 2
   Cap engineering headcount on consumer social.

   Big teams ship features users didn't ask for. Small teams keep the original product intact. The product is the moat; protect it with org design.

3. 3
   Group invite virality is the only consumer-acquisition channel that doesn't decay.

   Paid ads bring users who churn. Group invites bring users who arrive in a network they already trust.

4. 4
   D7, not DAU.

   A daily-active count is a vanity metric in a small-social app. The single number that mattered to us was: of users who joined this week, how many are still using it on day seven.

5. 5
   Choose not to add ads.

   It is a strategic choice with long compounding consequences. We chose subscription. The choice locks us out of certain growth paths and into others. Make the choice deliberately, not by drift.

What this looked like in practice.

The thing nobody tells you about *consumer social* is that the death of a network is invisible until it is irreversible. Clubhouse, Path, Peach, Vero — the deaths look identical from the outside (DAU collapses) and are very different inside. We instrumented our network density weekly and held a no-launch meeting any quarter the median user's connection count fell.

The thing nobody tells you about *organic growth* is that it is not free. It is paid for in product surface area. Every feature has to do double duty as an acquisition loop, which constrains the design space hard.

The thing nobody tells you about *Trust & Safety hires* is that the cost of getting this wrong is existential, and the cost of getting it right is invisible. Mei's salary is the highest non-engineering compensation on the team and it is the single best capital allocation we have made.

The thing nobody tells you about *content moderation classifiers* is that the open weights ones — Llama 3.1, Mistral, Gemma — are now competitive with OpenAI Moderation and Google Perspective on most categories, and they can be fine-tuned without sending user content to a third party. We saved roughly $90k a month and improved precision by switching.

The thing nobody tells you about *consumer DAU* is that it is a vanity metric the day after you raise. We track invitation-acceptance rate and seven-day return-with-a-friend instead.