The Cold Start Problem: How to Start and Scale Network Effects

by Andrew Chen

Of course, the killer product is just one ingredient in solving the Cold Start Problem. It was also important for Zoom to quickly figure out its first atomic networks, finding that organizations like the Stanford Continuing Education Program, and small Bay Area colleges could each adopt the product independently. Get one to adopt, and others would soon follow, as people using the product would naturally spread it to others. Once the killer product and the first atomic networks are built, then a company starts to create “Magic Moments.”

A product that hasn’t yet solved its Cold Start Problem will fail to deliver any magic in its early days. Often, the network will seem empty, like a ghost town. But once the network forms, the Magic Moments start to happen all the time—that’s when the product is ready to expand. That’s when you know the Cold Start Problem has been solved.

But there were little bursts of magic: sometimes I’d drop in and start having one of the most wonderful conversations with friends I hadn’t seen in months—after all, it was during the COVID pandemic year.

Clubhouse wasn’t recorded, so the expectation was more like a phone call and less like a podcast—this made it lower-pressure to talk.

Great products take time to figure out, and Clubhouse is no different. It was an overnight success that took years.

The Magic Moment is a nice concept, but it would be even more useful if you could measure it. The way to best do this might be surprising—you start with the opposite of magic, the moments where the network has broken down, and you start solving the problem from there. At Uber, we called these moments Zeroes. A zero at Uber was the worst experience you could have, when a rider opens the Uber app with the intent to pick an address and pick up a ride—but there aren’t any drivers in the area! This is a zero. When the point of the product is to interact with other participants in the network, a zero means that its value can’t be fulfilled, which means users will bounce and possibly never come back again. It’s obvious when there are not enough users, listings, or videos to make a product engaging.

This isn’t a problem just for Uber or for other marketplaces—each product category has its own version of a zero.

For a social network, a zero might be when a user joins and none of their friends or favorite content are yet on the service—causing them to spend their precious attention elsewhere.

A zero is a terrible experience, but even worse, it’s not easily solved. Adding just more one driver to the Uber network doesn’t ensure zeroes won’t happen, nor does adding one more colleague to a Slack workspace. To consistently ensure that people never experience zeroes, the network needs to be built out substantially, and it needs to be active, too!

The real cost of a zero is not just at the moment where it’s experienced, but rather, the lingering destructive effects afterward. Users who get “zeroed” often churn and worse, they come to believe the service isn’t reliable. It’s not possible to sustain a strong network when a large percentage of users are churning, but unfortunately, by definition new networks default to having many zeroes. Until this destructive force is dealt with, the network can’t get off the ground.

Thinking about zeroes and unfulfilled requests was such a useful concept at Uber that we baked it into many of our more common dashboards, split by city and region so we could understand how often it was happening.

Of course he was talking about getting to product/market fit more generally, but for networked products, I would take this description and infuse it with network goodness—users are inviting other users, and sharing content from your product across the internet. You search on Twitter, Reddit, and other social media and it’s chock full of your loyal users talking about how good your product is. Engagement goes up as people find more utility from it, and as more users join.

Around the world, this scene repeats itself, and when you sum up all the activity, it’s a big number: at the time of this writing, tens of millions of people use Tinder and they swipe more than 2 billion times per day, leading to a million dates per week. That’s what love looks like at scale.

Sean described how it worked: There was one catch with the party: First, you had to download the Tinder app to get in. We put a bouncer in the house to check that you had done it. The party was great—it was a success, and more importantly, the next day, everyone at the party woke up and remembered they had a new app on their phone. There were attractive people they hadn’t gotten to talk to, and this was their second chance.

Amazingly, 95 percent of this initial cohort started to use this app every day for three hours a day. The Tinder team built one atomic network, but soon figured out how to build the next one—just throw another party. And then another, by going to other schools, and throwing even more parties. Each network was successively easier to start. Tinder quickly reached 4,000 downloads, then 15,000 within a month, and then 500,000 just a month after that—first by replicating the campus launch, but then letting the organic viral growth take over.

In India, the focus was on call centers, whose dense communities were the equivalent of Greek life on college campuses. For Europe, people in the United States would invite their friends overseas, and connectivity got the product launched.

Sorry, you need an invite to sign up to this app.” No one will be happy to see a message like this. It might seem like a counterintuitive way to launch an app, particularly during the Cold Start phase, when new users are so desperately needed. Why turn down users who want to try your product? Yet this constraint is at the heart of the so called “invite-only” strategy for launching a product. And for Gmail, LinkedIn, Facebook, and many other networked products, it has worked. Why?

Some have espoused the invite-only tactic as a method to generate hype, since potentially a buzzy new product might cause people to head to social media to ask their friends for invites. Others say the value of invites comes from a method of limiting audience growth so that teams can fix bugs and scale a product’s infrastructure, before going to market more fully. Yes, these are all true, but they miss the most important reason to have invites in a product with network effects. Invite mechanics work like a copy-and-paste feature—if you start with a curated network, and give them invites, that network will copy itself over and over automatically.

On the first week of LinkedIn’s launch, employees and investors of the company could invite as many people as they wanted, but you couldn’t sign up from just the website. We intentionally seeded the network with the mid-tier of successful professionals that wanted to take time to connect.

Famously, Facebook initially required a harvard.edu email address to sign up, both defining an atomic network where everyone trusted each other, and also providing an explicit way to think about school-by-school launches.

These are all clever, and while invites-only strategies are often described as taking advantage of the fear of missing out—FOMO—that’s not the core driver. When a new product carefully curates a network, followed by implementing invites so that it can copy and paste similar networks, then it can grow to take over the market.

The Welcome Experience Invite-only mechanics provide a better “welcome experience” for new users as well. To explain why, imagine arriving at a large dinner party. A good friend welcomes you at the door, and as you step in, you see acquaintances, close friends, and a number of new people who’ve been curated to be absolutely fascinating. If that’s the ideal experience for a dinner guest, it’s also an apt metaphor for the best possible entry into a new product experience. Invite-only products can facilitate this, because every new user that signs up is already connected to at least one person—their inviter. For products like Slack or Zoom where you only need a few people to make it useful, having the guarantee of at least one connection is a giant step toward solving the Cold Start Problem.

Mathematically, it tends to work even better than that—the most connected people tend to be invited earlier, and in turn they tend to invite other highly connected people. The connected bring in the more connected. This results in a dinner party of socia...

It’s fascinating when you analyze the initial networks. The earliest users tend to have large contact lists, often way into the thousands, and tend to invite people with similarly large lists. Over months and years, the late adopters of the product might have just a few hundred connections. The end result of this mathematical property is that early adopters can be instantly connected with dozens of their friends and colleagues right away. Again, it’s a dinner party of social butterflies.

People with an invite to an exclusive product will post praise, critiques, and other commentary. People without an invite will ask for it, prompting discussion and sometimes controversy, driven by scarcity and exclusivity dynamics. This in turn attracts more attention and engagement. It works!

However, invite-only launches have been a key feature of many products precisely because for networked products, there are huge advantages. It allows the early network to gel as a community, develop a high density of connections, and grow organically via virality.

There are big advantages to an in-app wait list experience. For example, Robinhood, the commission-free online brokerage, had a hot and widely anticipated launch, and launched by signing users up to a wait list. On the back end, the team slowly let people in, pacing the growth so servers weren’t overwhelmed. The Robinhood mechanic asked wait list users to tweet or post to social media in order to jump ahead, ultimately bringing a million users in before a widespread release. Another variation of this is to ask users on the wait list to fill out detailed information about themselves, including their potential use cases, giving the teams a way to let a curated, small trickle of users in to form the initial network.

For networked products, the curation of the network—who’s on it, why they’re there, and how they interact with each other—is as important as its product design. Starting with a deliberate point of view on who’s best for your network will define its magnetism, culture, and ultimate trajectory.

Instagram launched on October 6, 2010, to the App Store, and at the end of the first week, it had been downloaded over 100,000 times. Within another two months, it had hit a million, and it just grew from there.37 To this day, it is still one of the fastest-growing apps ever created.

Eighteen months after its launch, Facebook acquired the company for $1 billion in stock and cash.

over time, the “tool” part of the product—the photo filters—have waned in importance, as users often post photos with the “#nofilter” tag. A recent analysis has shown that the vast majority of photos—82 percent39—had no filter used at all. Eight years after its initial launch, network effects had fully taken over for the utility of photo editing—it’s more network, and less tool.

Looking backward, this is widely seen as one of the best tech acquisitions in history, as Instagram would likely be worth several hundred billion as a stand-alone entity. It has more than a billion active users and generates $20 billion in revenue as part of Facebook. Not bad.

But for a large class of products centered on content creation, organization, and reference, it can be a winning strategy. When it works, the tool can help take on an entire network, and once atomic networks start to form, the entire market will come over.

But when this strategy works, it can be highly effective. It aids in the Tipping Point of a market because it is much easier to spread a tool than a network—after all, the latter suffers from the Cold Start Problem. Spread a tool far and wide, and then grow it properly, and it might start to build networks upon networks around the tool. Continue executing, and the entire market might follow.

Coupons presented a solution to the problem, and marketing legend Claude Hopkins used them to their full advantage, as he describes in his 1927 memoir, My Life in Advertising. He writes about solving this for his client, Van Camp’s Milk, which made a powdered milk: So I devised a plan for making Van Camp’s Milk familiar. In a page ad, I inserted a coupon, good at any store for a ten-cent can. We paid the grocer his retail price. For three weeks we announced that this ad would appear. At the same time we told the story of Van Camp’s Evaporated Milk. We sent copies of these ads to all grocers, and told them that every customer of theirs would receive one of these coupons. It was evident that they must have Van Camp’s Milk. Every coupon meant a ten-cent sale which, if they missed it, would go to a competitor. . . . The result was almost universal distribution, and at once.41

The common wisdom was, “If you have a chicken and egg problem—buy the chicken.”

You need to nail the killer product, and prove that you can gain an atomic network, before reaching for the financial lever. But once a team can reliably launch a product’s initial atomic networks, financial levers can rapidly accelerate the speed in which the market hits the Tipping Point.

No one wants to live in a ghost town. No one wants to join an empty community. In the early days, it was our job each day to make sure there was good content on the front page. We’d post it ourselves, using dozens of dummy accounts. Otherwise the community might dry up.

All of these dummy accounts looked and acted like real users, but it was Steve and Alexis controlling them. And while in the early days it might require a lot of manual searching and posting of content, the two became more clever over time. They began to build software to help them scale this activity—which Steve describes: I wrote some code that would scrape news websites and post them with made-up usernames. That way, it looked like there was an active community. Problem was, it still needed my attention—about a month after launch in July of that year, I went camping with my family and didn’t submit any links. When I checked Reddit, the homepage was blank! Whoops.

Services like DoorDash and Postmates Flinstoned by showing a large selection of restaurants, regardless of whether those restaurants had actually signed up. When customers ordered, the apps would send couriers to pick up the food, unbeknownst to the small, local businesses! They would just act as customers, grab the food, and deliver to the users who ordered it. Later on, once the demand had been proven the food apps would form direct relationships with restaurants.

Five years after Dropbox was founded at the Massachusetts Institute of Technology, the company had done the hardest part—solving the Cold Start Problem. The company did it with a classic “come for the tool” strategy, with file syncing across a person’s computers as the initial tool, followed by a network of shared folders with colleagues, friends, and family. They innovated with a referral program where users could give and get storage by inviting friends. User growth was explosive.

Dropbox had initially been built on Amazon’s cloud platform, and the product was growing so fast that the hosting bills had become very expensive. With hockey stick growth would come a second hockey stick, but this time, in file hosting costs. Building out in-house infrastructure48 would save nearly $75 million in just the first two years, and hundreds of millions over time. It was obvious it had to be done. But it would require a tremendous amount of up-front capital to lease and build out data centers, which in turn would push the company to unprofitability for the first time in years.

Just as there are high- and low-value users, there are also high- and low-value networks. In 2012, nearly 100 million people had signed up to use Dropbox. This vast network was made up of smaller, atomic networks—comprising hundreds of thousands of businesses, both big and small. The Dropbox sales team was able to “fish in their own pond,” prioritizing outreach to companies that had many, many users already on the product—using their email domains as a clue. Just as years back, Facebook had used edu email domains to partition smaller, engaged networks and spread from Harvard to other universities, Dropbox could do the same with the dot-com corporate equivalent.

An even more important signal was how many shared folders were being used at a company—the more collaboration via Dropbox, the stickier the product, and the easier upgrades would be to sell.

The product was launched with a self-narrated, four-minute video49 showing a “magic folder” that automatically synced files across computers, obviating the need for fragile USB drives. The first version didn’t have shared folders, though that would quickly come later. The video was released in April 2007, and it attracted a torrent of demand from users of social media sites Reddit, Hacker News, Digg, and others.

Drew recounted later: It drove hundreds of thousands of people to the website. Our beta waiting list went from 5,000 people to 75,000 people literally overnight. It totally blew us away.

While you may only need a small handful of employees to achieve product/market fit—famously, Instagram had thirteen employees and 30 million users when it was bought by Facebook—you need a significant coordinated effort to scale a product to its full potential.

Let’s start with a surprising idea that goes against the grain of industry jargon: the network effect is not one effect. Instead, the network effect is a broader umbrella term that can be broken down into a trio of underlying forces: the Acquisition network effect, the Engagement network effect, and the Economic network effect. Each one of these can contribute to a business in a different way, and is stronger the more dense a network is.

The “Acquisition Effect” is the ability for a product to tap into its network to acquire new customers. Any product can buy Facebook or Google advertising, for instance, to attract new users, but only networked products can tap into viral growth—the ability for users in its network to tell others in their own personal networks. This keeps customer acquisition costs low over time, fighting against the natural rise that comes with market saturation and competition. The types of projects that amplify the Acquisition Effect are oriented around viral growth: referral features that reward users when they invite others, tapping into contacts to create suggestions for who to add to an app, and improving conversion along the key moments in the invitation experience. All of these help increase metrics like new user sign-ups, the so-called viral factor of a product, and bring down the cost of acquiring a customer (CAC).

The “Engagement Effect” describes how a denser network creates higher stickiness and usage from its users—it is a more specific form of the classic description of network effects that I covered at the beginning of the book, “the more users that join the network, the more useful it gets.” However, the classic definition can be refined to include the underlying system that drives the value—use cases and “loops” that define how users derive value when engaging with a product—as well as the specific metrics that increase with a denser network. For example, Twitter is a lot more interesting to use, now with media outlets, celebrities, and politicians on it, than in the early days when you might just have a nerdy friend or two on the platform. Because there’s more types of content creators in the network, what might have felt like an app to stay in touch ...