Why high growth, high churn products never seem to work

yes, D30 ends up dominating over random social media spikes

First, a meme. We all love memes. We live in a world of memes. Usually these memes are expressed as photos and videos (like the one below, that is just so relatable):

Increasingly, there are now meme apps — these are simple, single action, single use experiences that are shared widely on social media, they explode virally, and ton of people try them. We see an amazing demo video of a cool visualization, and we must try it!! But after that a try or two, we also leave, never to return again. Easy come, easy go.

We’ve all seen them, so I won’t point fingers. But increasingly, we see them in the ecosystem of AI apps since this is the more dynamic field right now. I find myself following hundreds of “AI influencers” to see the flow of new products, and every week there are tons of product launches, cool video/photo effects. Sometimes I’ll share a video or two, just as I’d share a meme. But I find that I’m only sticking with a few products, leaving many abandoned. I’m sure your behavior is similar.

As I’ve written about with the Rise of Dopamine Culture, this is the natural evolution of software development becoming more accessible and more powerful:

• If apps are easier to develop — and they are, just think open source, cloud, plus copilot, LLM-driven app dev, etc — there will be a lot more apps. People can build them faster, and build many more of them. (Note there’s over 4 million apps in the various mobile app stores! And countless websites.) Inevitably they become quasi-disposable projects, delivered to the market quickly and then abandoned for the next idea. Just like a meme.

• Launching and growing apps via social media taps into meme culture. In a world where Discord, X, TikTok, etc are important marketing channels, the best apps are highly visual, engaging, a little funny, and tap into our collective mimetic nature. We subconsciously get hyped to try things that other people are trying, the digital reflection of the societal dynamics that philosopher Rene Girard argued. But it’s a double-edged sword, because memes are also driven by novelty and die off quickly.

• Customers who come in via social feeds have lower intent than those who are referred word-of-mouth. They are dopamine addicts, looking for their next rush, and that means they want to hit the magic moment in your product immediately. There’s no time for onboarding tutorials, long signup flows, or upfront account setup. They want their hit, and then to go back to social media where they can try their next new thing. Thus, the product experiences naturally become simpler, shorter, with an emphasis on sharing the output.

Thus, meme apps.

But there are dangers. Even in a world where every meme app gets their 15 minutes of mimetically-driven fame, there’s been a sordid history of apps that hit #1 and can’t ultimately sustain/retain.

The math behind meme apps
A while back I had a few charts in The red flags and magic numbers that investors look for in your startup’s metrics – 80 slide deck included! where I talked through the underlying math.

Here’s the first one, which is the way scaled customer acquisition tends to work. You might get a spike at first, but eventually it settles down into a linear (or S curve). There’s only a few channels that scale, and as a result, it’s natural that you can only grow your channels incrementally:

On the other hand, churn lags behind acquisition, but operates as a percentage of active users. This naturally means, more users = more churn. It’s generally very hard to move.

Combine these two graphs and you end up seeing a cross-over point where products naturally grow, then hit a peak, and start to fall. If you have very strong retention, and have a solid reactivation strategy, you can maybe keep the lines going for a while. But that is the exceptional case.

Meme apps have it worse, in many ways.

• The acquisition curve (what I call new+reactivated) ends up being spiky rather than a solid up-and-to-the-right curve. That means it’s even harder to generate meaningful separation

• The inactives end up churning more quickly, if it’s a single-use experience. Thus, the peak MAUs end up low

• Yet we see, of course, that sometimes these meme apps go way up into the charts. Why is that? Well, it’s simply because the underlying social media platforms they tap into now number into the billions of DAUs.

And as a result, they can spike so hard in their first few days that they look like they’re working. In fact, these spikes are dangerous.

The dangers of spiky customer acquisition
The dominant strategy for much of the marketing discipline over the past hundred years has been to drive big spiky launches. Consider the past century of centralized media ecosystem (think, 3 national TV channels, a major newspaper monopoly per city etc) — it meant that a great launch was just about getting a journalist to pay attention (organic) or to buy a big ad campaign (paid).

Today we use very different language, but the desire is similar. A decade+ ago, the startup ecosystem was obsessed with the “Techcrunch Effect” of being on the front page of Techcrunch. That might be surprising, but this was back when TC was highly relevant, under its founder, and one of only a few sources of startup news. Before that, there was slashdotting. In recent years, the media landscape for startups has wildly fragmented, particularly with the advent of social media, but generating a big launch spike still remains relevant.

Of course, there are major problems with a big bang launch:

• It comes and goes. If it’s the front page of a news publication, naturally it will be front page for ~24 hours (even shorter these days!) and then go away. If it’s driven by viral growth, then people will share it as long as it feels novel. A day or two in, once everyone has absorbed the killer demo video, or all the speculation and discussion has died down, it’ll go away.

• Big bangs are hard to repeat. Naturally you end up saving news, video content, and features, and then releasing it all at once. You might save up months of marketing effort so that it lands in one go. But once it’s out, how soon can you repeat a similar-sized beat? It’s just very hard

• You get a spigot of unqualified customers. Naturally the more viral the content you spread, the more appealing the content has to be to a wider audience. There’s a very big top of funnel, but how many end up being relevant as actual value-generating customers? This is obv even more tricky for AI products where each use case costs real dollars in compute

• For networked products, like marketplaces or social apps, a big spike of users that are unrelated to each other does not build a strong foundation. As I’ve discussed in my book, The Cold Start Problem, there’s a reason why the best network effects-driven products have started college-by-college, or team-by-team, rather than a big launch

Of course meme apps have all of the above issues, but even more so. Naturally they are memes so they try to tap into the cultural zeitgeist, are broadly appealing, and naturally have a very short shelf-life.

The biggest danger is simply the Illusion of Success. If you have a few days (or a few weeks) of viral growth, it can hide poor retention. It can make you feel like you have product/market fit, simply because the top-level numbers look like they’re going up. But as I argue mathematically above, eventually they will peak and come down, if retention isn’t there.

The growth treadmill
Real products can’t just spike and go away. Instead, they need to grow an active customer base, start doubling and tripling in growth, and continue year over year. I’ve written previously about the framework of T2D3 — triple, triple, double, double, double — the rate of growth needed to go from $1-2M in revenue to a scale where you could exit — and upon a big launch a real product has to follow that.

It gets very hard when you churn nearly all your users each year. Here’s a thought experiment — let’s say you have ~100% churn annually — to triple your overall active users, you must then triple your customer acquisition. This means you have to triple your ad spend, or triple the number of influencers you’re working with, or whatever. It’s very hard to triple SEO, or word-of-mouth, etc. Even if you can do that, imagine the next year — now you have to be 6-9x growth in acquisition, because you are replacing everyone who’s burned off in addition to hitting the new growth number. It’s very hard.

Products simply need to have great retention in order to make the T2D3 framework happen. If you retain nearly all your customers (and the best products do!) then you can double simply by acquiring the same number of users in one year as the next. And then you have a slower ramp after that. Much easier.

So what?
You might read this and ultimately think it a diatribe against AI products with poor retention. And in a way, it is. But of course all the founders should ride the novelty effect if it creates free users. After all, we’re at the beginning of a new tech S-curve (AI) and the close of an older one (mobile) and of course we should all take advantage, but also go in eyes wide open on the growth strategy and what it entails. Thus, a few conclusions from this timing in the market:

• Figure out what’s sticky (and who’s sticky). The products who are utilizing the novelty effect well have figured out that they have a particular segment of paying customer is actually sticky. So even if top-of-funnel comes in and out, they can build on this base of high(er) retention subscriber

• The novelty window is closing. Not that long ago, we were wowed by AI-generated photos. Now that content isn’t getting much mimetic love, and only highly sophisticated genAI videos will do. Eventually those will get boring too, as we perfect the technology. Not that long from now, consumers will mostly ignore new AI tech, and companies will have to actually market their products, not just ride viral waves on social media

• Learn what’s worked before. Maybe I’m old fashioned, but if there’s a new gen AI product that’s useful in work/productivity settings, likely what will cause it to spread within a company are network effects-driven. Collaboration features, ways to invite your team, integration into existing workflows, etc. Right now the many of the AI products are headed by highly technical teams that aren’t building these kinds of feature set, but within 2-3 years these will be the major drives. The AI tech will be the cherry on top

• Embrace meme culture. If meme apps are going to be a thing, it’s also natural to think about what it means to build thousands or millions of these fun apps that capture the zeitgeist. In a way, that’s the difference between Roblox and a gaming company, or YouTube and a traditional streaming service. Perhaps we’ll see many of these tools turn more into platforms, as we’ve seen face filters do.