I went #1 on the AppStore in December with @yusuff_hamza. We beat ChatGPT, TikTok, and hundreds of other apps.
We ran 0 paid ads. Our growth was entirely through word of mouth, seeded originally through UGC.
This is every product and growth hack we ran.
2025 Wrapped, #1 on the AppStore
About me: I've built multiple #1 apps, 2025 Wrapped being the first that hit #1 overall. I'm currently building @lightreelai, an AI marketing tool that helps you find creators, analyze your Instagram / TikTok accounts for feedback, and surface viral hooks right as they're exploding. however, everything below can be done without using Lightreel
Becoming a creative data analyst
I mainly adopted this philosophy from @nikitabier and @rjvir, both also founders of multiple #1 apps.
There are two parts to building a great, viral product:
**1) Optimizing local-product maxima
2) Optimizing for macro trends
**
Most founders do one or the other really well, but rarely both.
**Local-product maxima
**
Local product maxima can only be discovered with intense data analysis. Not just % completion rates of onboarding, but combinations of different events to answer:
"What product behaviors stem from causation and not correlation"
As a default, you want to track all the basics. Screenshot rates of every screen, sharing to instagram, drop offs on certain screen, # of posts created, etc.
Just a fraction of the events we were tracking for 2025 Wrapped. Imagine this 10x
However, you also want to focus on unique cohorts.
For example, you might slice your screenshot rates on age group. Perhaps you track which apps a user has installed and detect how that impacts their onboarding completion rate. Maybe you detect that people with gambling apps complete your onboarding 3x as often as those without.
These will make developing product insights 10x easier.
For 2025 Wrapped, we realized the onboarding completion rate was significantly higher for newer iPhones.
People with older phones could barely get through the onboarding as the photo-processing steps were so computationally intense.
The loading screen went from a huge drop off experience to something that actually IMPROVED our viral-ness
After discovering this, we completely redesigned the loading screens, employing caching tricks, UI designs, and subtle UX polish (like surfacing your friend's names during the loading screen, without contact-book access) to drastically increase the onboarding completion rate.
Early onboarding improvements. All of these preceded our US explosion to #1
These cohorts allowed us to refine the app and really hone in on our thematic messaging. Some of our default assumptions (like # of screenshots as related to age, gender, region, etc) were completely off.
However, these creative analytics alone could never lead to a product that beat ChatGPT.
Building towards macro-trends
No amount of micro-optimizations can ever force a successful product
Too many founders treat product development as a mish-mash of analytical improvements. "Perhaps if our Mixpanel stats were marginally better, we'd finally take off"
There is some truth to this. Often times, a viral product is ~almost viral for the entirety of its duration until one small onboarding tweak changes everything (the difference between a 0.95 k-factor and 1.01 k-factor, for example)
As you improve your metrics, the new cohorts might not necessarily convert / complete / activate at the same rate
Optimizing for macro-trendsIt is very likely, however, that you are in micro-optimization hell.
Try describing your product without any adjectives.
Many "authentic, friend's only music sharing apps about how you feel" reduce to "we rebuilt Airbuds, except now it's 2026"
There is almost always some technological or cultural shift that enables really great products, especially in consumer.
Locket
-> home screen widgets
Cursor
-> Launch of GPT-4 (plus willingness of devs to try new tools)
NGL
-> Instagram enabled sharing on stories
CharacterAI
-> Signals from Replika + guardrails of ChatGPT
2025 Wrapped took a few signals:
1) Spotify Wrapped + similar Wrapped products had been exploding every year
2) Camera roll had insights that local-models were just now powerful enough to detect in real time, on device
3) Hamza and I had already made other consumer apps explode
4) It was late November, when Wrapped apps usually blew up
0:00 / 0:15
Spotify Wrapped video with 3 million views. Hundreds (thousands?) of Spotify Wrapped videos would get millions of views every year
These macro-cultural trends, combined with an insane obsession with slicing our data (moreso than this article lets on) is what made 2025 Wrapped explode.
How to find a distribution arbitrage
I'll keep this section short as I've already written extensively about running a UGC program. However, most Twitter-guru app founders miss out on this:
**Your distribution arbitrage is usually in plain sight.
**
It's not so much about finding some diamond in the rough, but instead about focusing on doubling down on one ~easy~ distribution angle.
Cal AI
-> Medium sized fitness influencers (shoutout
@zach_yadegari , and
2025 Wrapped
-> Sorority girls who had never done UGC (for context, we did have some of the best UGC creators, like
@carlynorthmedia , originally go viral before we scaled)
S
ocial apps today
-> Creators in Spain, Eastern Europe, Mexico
https://x.com/jakecastilloooo/status/2067602533799874630
None of these creators are ~that~ unique. Everyone knows fitness influencers do paid collabs. Same with creators in Spain. What's unique is all these companies scaled and doubled down on their particular angle.
All of these startups (including 2025 Wrapped) spent thousands of dollars / month on finding their unique distribution.
I've spent the last few months building out a lot of the creator discovery infrastructure into Lightreel for much cheaper. A LOT of consumer apps (like, all the ones you know) are using it to find their unique distribution arbitrage.
Asking Lightreel to find every creator that partnered with Anthropic.
Lightreel is really, really good at finding hyper-specific creators with qualitative descriptions. A VA in the Philippines often won't understand what a "DITL creator who doesn't have too many brand collabs in the last 2 weeks, has signs of being in a relationship, and under 10,000 followers" actually means. Normal creator databases don't, either.
https://x.com/Nick_5anchez/status/2070192926961938783
What to do now
- Add analytics to everything, even the things you don't think are important
- Shift your mentality. What am I building? Why? Am I optimizing towards a local maxima without considering broader cultural and technological trends?
- Focus on one distribution angle and scale that hard. What is the unique edge you have here? You can't outspend Higgsfield, but you can be more creative.
If you appreciate the way I think through hooks, script writing, and creator marketing, try out Lightreel for free, too. It has everything I've learned, all as a product you can just talk to.
Get feedback on your content with Lightreel
