Andrew Chen's growth frameworks, tested against one real app
the cold start problem and the law of shitty clickthroughs, checked against scaling ZuAI to 2 million users. what held, what did not, and what he never claimed.
the short version: Andrew Chen’s two best known ideas are the cold start problem and the law of shitty clickthroughs. one of them predicted almost everything I ran into scaling an app to 2 million users. the other is frequently misused by people quoting it, including as an excuse.
I am not a critic of his work, I am a user of it. this is what it looked like from inside an ad account rather than from a book.
The two ideas, briefly
Chen spent years writing about growth before becoming a general partner at a16z, and wrote The Cold Start Problem in 2021. its argument is that network products are worthless until a small part of the network is dense enough to be useful, and that the entire job early on is manufacturing that first pocket of density. he breaks the arc into five stages: the cold start, the tipping point, escape velocity, hitting the ceiling, and the moat.
the law of shitty clickthroughs is older and simpler: every marketing channel decays. response rates fall over time, on every channel, forever, because the channel gets crowded.
What held up exactly as described
the decay law. it is the most reliably true thing anyone has written about growth.
scaling ZuAI from 10,000 to 2 million users at a $0.02 blended CAC, nothing we found stayed cheap. a creative angle that worked in month one was tired by month three, not because we got worse at making them but because the audience had seen it. reddit worked until the obvious subreddits were saturated. every single channel followed the same curve down.
the practical consequence is not “find better channels”, it is build the machine that finds the next one before the current one dies. that is why creative volume matters more than creative quality at seed stage, and why I test in the dozens rather than agonising over three. the creative testing planner is the arithmetic version of that argument.
What did not apply, and Chen never said it would
the cold start problem is about network products. things that get better as more people use them: marketplaces, social apps, messaging, anything with a two-sided dynamic. Slack, Uber, Tinder, Airbnb.
most consumer AI apps are not network products. ZuAI got no more useful to you because someone else used it. the product worked identically for user ten and user two million.
this matters because founders read the book and apply the framework to a product that has no network effect, then spend months trying to manufacture density that would not help them if they got it. if your product does not get better with more users, you do not have a cold start problem. you have a distribution problem, which is a different and frankly easier thing.
the test is one question: does user 1,000 have a better experience than user 10 because of the other 990? if not, skip the network chapters.
Where I would push back gently
the ceiling stage in Chen’s model is described as something that happens when the network saturates and channels stop working. in my experience the ceiling usually arrives long before saturation, and it arrives because of retention, not acquisition.
at 8% month one retention, your active users flatten out at roughly your monthly installs divided by 0.92, and no amount of extra spend moves that much. you can run the numbers yourself in the retention ceiling calculator. most teams who think they have hit an acquisition ceiling have hit a retention ceiling and are trying to solve it by buying more users, which is the most expensive way to fail.
that is not a contradiction of his framework so much as a different failure mode being mistaken for it.
The part people quote wrongly
the law of shitty clickthroughs is often used to argue that marketing is futile, or that a channel dying was inevitable and therefore nobody’s fault.
that is not the claim. the claim is that channel efficiency decays, not that growth is impossible. the operators who do well are the ones who accept the decay as a constant and build for it: always testing the next angle, never letting one channel become the whole business, and measuring honestly enough to notice the decline early rather than after a bad quarter.
used as an excuse it is fatalism. used as a design constraint it is one of the most useful ideas in the field.
If you want the shorter version
read Chen for the model. it is genuinely good and I would not have framed my own thinking as clearly without it. then check whether your product actually has a network effect before you use the network parts, and check your retention before you conclude that your channels are the problem.
I have written up what actually worked in the ZuAI case study, including what failed, and the current state of the discipline in growth hacking is still real in 2026.
Sources: The Cold Start Problem and its five stages are summarised by a16z; the decay principle is Chen’s own law of shitty clickthroughs.