Software may be eating the world, but building and scaling products is still quite a challenge. So how do you get past the awkward “cold start problem” of zero users and build the networks that make your product thrive?
If you’re in the startup world, chances are you know exactly who Andrew Chen is. Moreover, you’ve likely benefited from his advice at one point or another. A mathematician by training and growth expert by trade, Andrew led Uber’s Rider Growth teams in the pre-IPO period, and for the past three years, he’s been a general partner at Andreessen Horowitz, the most widely-known VC firm on the block.
In the last decade, he’s written hundreds of essays on growth, and yet, there has been a burning question in his mind. What was the secret of all the Silicon Valley high-growth startups we’ve seen over the years? How did the tech giants get so… big? Well, the quest for that answer led him to write The Cold Start Problem: Using Network Effects to Scale Your Product.
Even though the underlying principles of network effects have been around for ages (just think about the telephone), we’re only starting to grasp their full potential now. Essentially, network effects describe how the value of a product or service depends on how many users engage with it – the more users engage, the more appealing it becomes. In this book, Andrew provides a practical framework for how companies like Google, Uber, Dropbox, and Tinder use them to break through the competition and reach the holy grail of viral growth.
How do teams create and build them into their products, whether they’re messaging apps or collaboration tools? How do you solve the zero users conundrum? And in a market where everyone else has them, how do you stay on top? This week, we had the pleasure to welcome Andrew Chen back again on the podcast to chat about his book and how the companies you know and love solved the “cold start problem” and scaled into millions and billions of users.
If you’re short on time, here are a few quick takeaways:
- Networks have different underlying structures. While in some companies, like Uber, growth is bound by geography, others, like Airbnb, have a global network effect. Context is important.
- The first thing you need to ask yourself is: what does your network look like? Only then can you think about how many users you need to make the network valuable and come up with a plan to get there.
- If you figure out how to get a single team excited about your product and spread it to others, you’ll figure out how to build 10 or 20 networks. That’s how you get enough momentum.
- Products powered by network effects generally follow some kind of an S curve. When growth slows down, you’re going to need to come up with a new one.
- Big companies can easily get millions of people in a new product launch, but it’s crucial to do the work. Get small groups of people excited about your product and move from there, network by network.
If you enjoy our discussion, check out more episodes of our podcast. You can follow on iTunes, Spotify or grab the RSS feed in your player of choice. What follows is a lightly edited transcript of the episode.
Unpacking network effects
Des Traynor: Andrew, you’re very welcome back to the show. I’m delighted to have you here today. First of all, thanks for joining us.
Andrew Chen: It’s wonderful to be involved. Thanks for having me.
Des: So, your book, I am three-quarters of the way through, which I think is a great point to do this sort of podcast, because it means I’m not going to ask you questions that I know the answer to in all cases. What strikes me as being quite current is that a lot of folks are talking about product-led growth today, and I feel like that’s always floated around. It used to be called self-serve or bottom-up of adoption, or whatever. The marketplace flavor of that is, I think, anchored so heavily on network effects and what you state as, in the title of your book, The Cold Start Problem.
There’s so much here, and there’s so much depth that isn’t captured by an average UX designer. Because there’s so much, I want to say, near-sociological human theory about what works with actual design practices to grow a product. But maybe, to start off and help orient our audience – our audience is a collection of everyone from customer support, sales, marketing, product, engineers, et cetera – we start off with network effects specifically. What do they mean, and where do they start? When did humanity start caring about such things?
“Theodore Veil, who was the chairman of AT&T in 1905, talks about how the phone by itself is a useless thing. It’s not even a toy. It’s only valuable because it lets you call people”
Andrew: Well, I wrote the book primarily because I have been trying to answer the question of what is really the central secret of all the Silicon Valley startups that have been created over the years. Why is it that these companies specifically have gotten so big and become so impactful in the world? And one of the ways that I’ve distilled that down is that a lot of these products are just different ways to connect people. Whether you’re at work, talking about products like Zoom, Dropbox, and products like that in terms of collaboration; products like GitHub, which are for a particular function, a certain workflow. Even Microsoft Office, even the Google Suite Workspace are all different flavors of this.
When I started to explore this concept of network effects, if you go backward in history, what you realize is, “Wow, this idea has been around for a really long time.” When you go back to telephones, and I have a quote in the book by Theodore Veil, who was the chairman of AT&T in 1905, he talks about how the phone by itself is a useless thing. It’s not even a toy. It’s only valuable because it lets you call people. It’s only valuable based on who it lets you connect with. And so, I think when you start to think about it that way, what you realize is that you can build a product that has all the right features, that has all the features your customers ask for, but if you don’t build the network alongside it that lets people then connect with each other in the right way, if your coworkers aren’t using it, your colleagues aren’t using it, it doesn’t matter that you’ve built all the right features, the product’s not going to be successful.
I find it so fascinating to be able to go back and look at telephones, at railroads, which actually have network effects, and cable TV has network effects. A lot of these are sort of pre-tech companies and product ideas actually have network effects, but we’ve certainly taken them to the next level in the digital age.
“You actually do have a cold start problem when you start a nightclub. Because you’re like, ‘Well, how do I get all the cool, attractive people to the club at the same time?’”
Des: It’s fair to say even nightclubs would have network effects, right? People go there because there are people there, and people wouldn’t go there if there weren’t. The quality of the product is determined by the volume of usage.
Andrew: That’s right, and the funny thing about that exact thing is that you actually do have a cold start problem when you start a nightclub. Because you’re like, “Well, how do I get all the cool, attractive people to the club at the same time?” And then, on the other side of it, and this is the other part of the theory, once you have too many people, then it’s also bad. It becomes too saturated. In the same way that when your grandma’s using Facebook and everyone’s using Facebook, you can’t post your party photos, and you have kind of that context collapse. In the same way, when a nightclub gets too popular, it faces some of the same issues.
Des: It reminds me of that famous quote, “No one goes there anymore, it’s too crowded.” It’s like that type of thing. There are both cross-side and same-side network effects. The idea that, say, in Uber, more drivers attract more passengers, but even the passengers can attract more passengers by me getting an Uber and introducing it to my friend on the way home or whatever. This idea of same-side and cross-side. And similarly, the attractive one-half of the marketplace can bring all their friends the next time and that makes the place even more attractive. I think a lot of these things are more complicated than people actually realize, and you think, “Okay, God, I have to see this whole thing,” right?
Andrew: That’s right, yeah. And that’s actually a great point. You can go through and build this crazy taxonomy of different forms of networks because the reality is that the networks you use in a work situation when you’re growing from company to company to company, team by team by team, are very different from something like Tinder that grows from one college campus to another college campus. But what I try to do, within the book The Cold Start Problem, is to really abstract out some of these ideas into broader ones.
“Almost every single one of these network products has some kind of a small minority of users that become the power users that do all the work and create all the value”
And so, one of the things that you just mentioned is one of the core concepts, which is that these networks actually all have different sides to them, and the sides do different things. Sometimes it’s really obvious, in Uber’s case, riders and drivers are very different than each other. But funny enough, even in a case like Slack, where it sort of feels like, “Well, everyone just kind of types and everyone just kind of talks,” actually, there’s a small percentage of people inside the Slack ecosystem that make all the channels, that invite all the people, that talk all the time.
Des: Productive employees, usually.
Andrew: That’s right, exactly. And so, one of the arguments that I make is that almost every single one of these network products has some kind of a small minority of users that become the power users that do all the work and create all the value, but they’re also more expensive to acquire. And as you might imagine, a lot of the Slack channels are being created by first-line managers, for example, that are building these channels for their teams. And they’re just more expensive to acquire than a run-of-the-mill individual contributor. And so, that’s true in almost any product that you look across. So, I totally agree with you on that.
Scaling a minimum viable community
Des: Is it true, or what’s your take on the following, that whenever significant network effects exist for a product, that category will tend towards being a winner-takes-all environment? If it is the case that whoever gets most usage on most same-side and cross-side network effect attraction, et cetera, becomes the dominant winner, and it becomes more difficult for anyone else to catch up.
I’ve never worked in a marketplace-type business – you have. So I’m not trying to put words in your mouth, but as I was playing this out, I was like, if I’m building a note-taking app and you’re building a note-taking app, and it’s just for personal use, I think it’s just my product versus yours. If it’s collaboration, whenever we win a user, we possibly win their whole sphere of influence as well. As a result, if I can build network effects into mine and you can’t into yours, I have a better chance of saturating the market quicker. What’s your take?
The team that decides to start making their documents in Google Docs will probably get everyone on the team to standardize on it. If you can win that network, all the adjacent teams become more likely to go.
Andrew: Well, on the competitive side, first, it is often the case that the most interesting companies are taking areas that were not previously considered network effects-driven and adding network effects. If you think about what Figma is, for example. This is a product category that Adobe had for a very, very long time. Or if you think about Google Docs, the same thing. This is a category that Microsoft had for many, many years. And in many ways, there was some kind of light collaboration, because you do your mockups and maybe email the PSDs around, there’s that type of thing. Microsoft had track changes, but Google went in and added commenting, real-time collaboration, permissions, added all these different things. And what happens is, to your point, oftentimes, the team that decides to start making their documents in Google Docs will probably get everyone on the team to standardize on it. If you can win that network, all the adjacent teams maybe become more likely to go.
Now, the interesting question, though, does it become a winner-take-all? And I think one of the things I would argue is that the answer is… kind of. The problem is that you have to think about the underlying network structure for it to make sense. So two companies that are near and dear to my heart, Airbnb, Andreessen Horowitz did this Series B investment in, led by Jeff Jordan, one of my colleagues, and then also Uber, where I ran many of the growth teams there. And you compare the two and they’re just fascinating. The problem with Uber is that if they are very successful in San Francisco, that does not help them win in London. It doesn’t help them win in New York because each of their networks, each of their sub-networks is pretty separate.
Andrew: Geofenced, exactly. Versus something like Airbnb where most people are renting rooms halfway across the world. And so, because of that, you have this global network effect, you have a lot more intertwining of all of their various sub-networks. The B2B example of this would be like Zoom versus Slack. Zoom you’re often using across organizations, Slack you’re usually using within an organization. And so, they’re going to have different dynamics. I think what often ends up happening is you’re often winner-take-all but in the context of your local network. It just sort of depends on the structure. So, if everyone decides, in a company, to use Google Docs, that’s the kind of thing that might actually spread through the entire company versus something like storage, which is a little bit more fungible, where it’s more like, “Well, I share these one-off files with people all the time.” Some people might do that over email. Some people might do that over Dropbox. Some people might do that over Box. And there’s little standardization as a result.
Des: Yeah, exactly. Because it doesn’t actually improve or change most people’s experiences just clicking a link to download, in a sense.
Andrew: Yeah. The level of collaboration is just not as deep for that.
“The very first version of Tinder was actually very good. It had the big profile photos, it had swiping, it had messaging (…) However, in the first few months, they couldn’t get anyone to use it”
I don’t care if anyone else is looking at my expense report. It’s not useful, right? But there will be areas where it’s like, “Oh, you were about to send this email? In that case, I won’t.” I think we’ll see people build tools to simplify how to add some of the superficial aspects. But I actually think what you’re saying is that it’s deep and meaningful collaboration that wins the day, not the sort of surface-level stuff, although that might be useful for starting it.
Andrew: Yeah. Realistically, you’re going to need to design your network features very thoughtfully. That is one of the main arguments I’m making. One of the case studies I have in the book is building Tinder version one. For the early team led by Sean Rad, the very first version of Tinder was actually very good. It had the big profile photos, it had swiping, it had messaging, it had all the key ingredients of what you would describe as Tinder. However, in the first few months, they couldn’t get anyone to use it. They were just texting their friends, and their friends would kind of be like, “Oh, are you insulting me? You’re telling me I’m lonely so I should have a dating app?” It was during a time when it was less accepted.
“You have to decide how many users you need before the network is actually valuable”
And it wasn’t until they actually had a whole plan for how they were going to get their first 500 users, which was: They were going to go throw a party at USC, the local campus, and they were going to make everybody who went to the party download the app, and they were going to have bouncers, and you had to show the thing. But they rented a really cool house, and they sponsored this birthday party, and then, all of a sudden, they had their first 500 users. I think my point in telling that story is that you have to have a very thoughtful understanding of what you think your network’s going to look like first. Is it a hyperlocal network? Or is it a network more like marketplace companies like eBay? Is it more like that? Reddit is another topical network as opposed to hyperlocal.
You have to decide that, and you have to decide how many users you need before the network is actually valuable. Zoom might be useful with two or three people, Slack might be useful with five or ten people, the people in your team. Something like Tinder, Airbnb, or Uber needs hundreds of network participants all in a hyperlocal region for it to work. And that’s why I would find it hard to copy and paste other people’s features into your products. You have to think about, “Does my app need to be geo-aware, or is it fine if it’s a company?” It sounds like what Cord has done is really interesting if your particular approach is teams within companies for collaboration.
Des: Just looking at what each other are doing versus actively collaborating together, right?
Andrew: Exactly. Right.
Following the S curve
Des: To get into specifics for a second, in the book, one of the really nice frameworks you have is how you break down this theory, if you like, the Cold Start Theory, which you break into five primary stages – the problem, the tipping point, escape velocity, the ceiling, and then, lastly, the sustainable moat. Without trying to give away your entire book, could you break those down a little bit further, so our readers will know what’s going on in each of them?
Andrew: The first part of the book, which is focused on what the book is titled, The Cold Start Problem, is all about what happens when you have zero users and you don’t know what to do. What I’m doing in that part is teaching the reader some of the core central concepts of what I’m discussing later. I talk about some of what we’ve already discussed, which is that networks have these sides, these benefits. Network effects often are classically defined as products that become more useful as more people use them. In the context of a world that is increasingly competitive, there are millions of apps and everyone’s trying all these different things, and by having a network, you have a way to break through, and you’re able to build one of the most valuable products in the whole industry. They’re all often built on network effects.
And so, I talk about how the core strategy is really figuring out how to build these atomic networks. How do you get a single team to utilize your product and get excited? If you can figure out how to build one of those and the second one and the third one, you’d probably figure out how to build 10 or 20, and off you go. And once you achieve that, you’re able to get enough momentum to hit the tipping point, which I talk about as a repeatable process of generating these atomic networks.
“You’re trying to get the market to choose you. If you can win the market, then you become the monopoly”
If you’re able to do that, and I have a bunch of examples and playbooks for how companies like Reddit and Zoom achieve that, what ends up happening there is you’re able to really spin up your user base. Some of the ideas I talk about are: What does it mean to subsidize the market? Why is it that so many of these marketplace companies, for instance, whether you’re talking about a B2B marketplace or consumer marketplace, end up subsidizing the first few years of their existence? Economics are often completely upside down, and it takes years to get to a point where you can even talk about being breakeven. Why is that? And why is that often a smart idea, right? And a lot of that has to do with the fact that you’re trying to get to the tipping point. You’re trying to get the market to choose you. If you can win the market, then you become the monopoly.
Des: And then you can apply the price, basically.
Andrew: Exactly. And then you can figure out how to do everything in the next phase, the escape velocity, which is all about doing things in the product and on your team so that you’re able to double down, triple down, and quadruple down on something that’s working. And if you’re doing that, you’re probably building a growth team, you’re probably finding your second, third, and fourth go-to-market channel. Whether that’s more like paid marketing or SEO or one of these things, you’re often thinking about how to build your product in a way where it will spread inside of a company automatically, so you’re building these viral sharing features and doing all of that acceleration.
And then, I talk about the ceiling, which is what happens when you’re a later-stage company that has network effects and you find that your growth just slows down. You just can’t help it. It’s just very, very hard. And the reason is because if you were tripling or quadrupling in your early years, the problem becomes, “Well, in year seven or eight, there’s just not enough customers in the market.” And so, you’re starting to hit saturation. You’re starting to hit a point where, like Dropbox, a lot of your product usage is being driven by people who are pirating movies in Southeast Asia. You’re never going to make any money on those users, so what do you do to focus your audience down?
“Oftentimes, when you ask a startup, ‘Hey, how are you going to compete against everyone else?’ They just say, ‘We have network effects.’ Well, it turns out they have network effects, too”
Similarly, if you are a big social product, or your eBay or marketplace or something like that and you have to deal with trolls and fraudsters, and spammers – this whole section describes how these products that we honestly think of as invincible are actually not and are struggling just to stay on a growth rate that makes sense. And I think we all intuitively understand that. We both talk about a company like Facebook like it’s invincible, and we also talk about how we’re all using Instagram more often, right?
Des: Or TikTok or whatever. Yeah, totally.
Andrew: Right. And then, the final section, the moat. Oftentimes, when you ask a startup, “Hey, how are you going to compete against everyone else?” They just say, “We have network effects.” Well, it turns out they have network effects too, most likely because they’re also marketplace companies or collaboration tools or whatever. In that case, if you’re the big guy and they’re the smaller player, what do you do differently? And then similarly, if you’re a smaller player and you’re going after a big guy, maybe you can figure out how to cherry-pick some of the various smaller networks they have. Airbnb’s not going after Craigslist’s entire business – they just went after their shared rooms product. If you are competing against Microsoft Office, you’re probably not building everything in the Suite and all the features. What you’re trying to do is you’re trying to pick off some smaller piece, like what Notion is doing, and just build a great product off on that and then grow from there. And so, I talk about some of the techniques and strategies involved in a world where everyone has network effects.
Going back to the start
Des: On hitting the ceiling piece, as you go mainstream, you are at risk of, let’s just say, worst usage. As you keep chasing it, you’re going to get pirates, trolls, or whatever. You’ve gotten all the natural usage, and now you’re trying to force, it in a sense. Is the best strategy, in your opinion, to guide the market to say, “Hey, don’t expect 100% growth for us when we’re already at two billion users”? Or is it more like finding other sources of stickiness, like Dropbox, I feel like I’m unfair saying this, but, “Hey, we’ve dominated file sharing, we need to find something else that’s not file sharing. So maybe that’s why we build Paper.” It’s like, find the next sticky thing because the old one’s going to grow old too quickly.
“When you’re a larger company, you often get to a state where everyone who solved the cold start problem for your first product has retired, has left the company, and is off doing other things”
Andrew: Number one, absolutely, you have to figure out the new network product that drives everything forward. That’s the actual solution to the whole thing. These products generally follow some kind of an S curve, and then you’re going to need to come up with a new one. Now, the whole problem is that, when you’re a larger company, you often get to a state where everyone who solved the cold start problem for your first product has retired, has left the company, and is off doing other things. And the people that you have left running things have only known peacetime. They’ve never known what it takes to build something from zero.
Des: You have another problem that Google had a lot of. Even if you have a really bad idea for a product, you get a hundred million users on launch day. Because you’re Google. You see bits of this with Facebook trying to expand their product line, even if Slack wanted to do something, you’re going to bump into this challenge. We get a flavor of this at Intercom. We launched something new for two, three, four weeks, and it’s popular. We get no active real signal because asking 25,000 people to try something when they already try all the rest of your stuff, of course, they’re going to try it. And so, this is the sympathy I have for Google. They do launch a lot of software – they certainly went through a period of seven different messaging apps, three social networks, and things like that. And I think it can be quite a while before they realize, “Huh, this isn’t sticking at all. What looks like a lot of usage is just slush, in a sense.” Right?
“When you’re looking for things that look like big numbers, like getting a hundred million people, you’re ultimately focusing on the wrong metric”
Andrew: That’s right. One of the things I really admire about the Snapchat team as I’ve gotten to know them over the years is that they’ll just go to a high school and build apps for one high school for a whole year. And they’ll dispatch a little tiger team to go do that, and they won’t go and tell them, “We’re Snapchat! Do this, do that!” They’ll just do the hard work it takes to start from zero and get really small groups excited about what you’re doing.
Because the main issue, to your point, is that when you’re looking for things that look like big numbers, like getting a hundred million people, you’re ultimately focusing on the wrong metric. You have to be so focused on retention and engagement. And you have to be really, really focused on small groups of people getting entire groups on at the same time and then building network by network. You can’t do the top-down – I call it a big bang launch. I think everyone gets used to, at some point, just writing a blog post and talking to all the news and media outlets and PR outlets, and then you expect to get a huge amount of usage, but the problem is you get a lot of disjointed usage. You get a lot of random spray paint of random users that don’t know each other as opposed to really, really focusing and taking time. I think one of the asymmetric advantages of startups is the patience of the founding team. They’re able to work with really small numbers of people for substantial periods of time versus larger companies, as you know, which get in a place where you just want to go big, really fast.
Des: I think it’s the willingness to focus on true product usage as in, the people are using this thing for exactly what I designed it to be used for as opposed to just, “Hey, our numbers are looking really high.”
The dawn of web3
Des: It would not be 2021 if I didn’t ask you a question about crypto, blockchain, or something in that space. And I’ll phrase this one loosely so you can tackle it any way you wish. When I think about the various webs: If web1 was multiplayer and web2 was the social era of the net, and if web3 is built on all of those, even when I think about NFTs or coins, do you see the idea of the cold start problem into web3? Will all solutions from here onwards need to solve this problem? Do you think it’s relevant at all?
Andrew: I think it is super, super interesting. I had various starts and stops of trying to write more web3 content into the book. It is changing so fast, unless you’re literally talking about Bitcoin and Ethereum. And so, that’s what I’ve done. I have a mention of Bitcoin and Ethereum, but I don’t go super deep into it. Look, I think the reality is that the network effect is an inherent part of web3 in such a deep way. Why do I think Bitcoin is valuable? And why does Alice think Bitcoin is valuable? It’s because, well, everyone else we know thinks Bitcoin is valuable, and that’s why it’s valuable. It’s very self-referential. Even if you were to literally take the Bitcoin codebase and fork it, and the Blockchain and fork it, that new Dipcoin would not be more valuable than the original one because you literally just need the market participants in there. And this is true for Bitcoin, and this is true for Bored Apes, this is true for CryptoPunks, and so on.
“Counter-Strike has been around for over 20 years, League of Legends has been around for over 10 years. And it’s the inherent network effects that allow them to stay popular for so long”
What that means is, if you are in a world where you are building NFTs and you’re going to do a big NFT drop, then you’re going to have to solve this cold start problem. You’re going to have to figure out where the cool kids are hanging out on Discord and on the right subreddits to get the right Twitter influencers into the drop. And that becomes a thing that you’re going to have to do. Now, we are in an age right now where every single cool NFT project is getting a lot of attention. And I think we’ll get to a point where it’s going to have to be more sophisticated because all the techniques that are working now, I think, in a year, will stop working. People are going to ignore when folks are super noisy about something on social media, for instance.
The other variation of this I think is interesting is that, over the last couple of quarters, we’ve also started to see a lot of consumer-facing web3 projects. I spend a lot of time in crypto gaming and companies like Axie Infinity and many of the smaller companies, and the big thing that you see there is that you’re going to overlay a series of networks on top of each other. There’s both the economic aspect of it, but also, a lot of these games are going to be multiplayer. A lot of these games are going to feel like they’re 3D immersive social networks. And that also has a very network effects-driven set of dynamics. 20 years ago, you went to a retailer, like a Best Buy, bought a cartridge or a DVD, and stuck it into your console. But these days, Counter-Strike has been around for over 20 years, League of Legends has been around for over 10 years. And it’s the inherent network effects that allow them to stay popular for so long. I think it’s very exciting.
Des: It absolutely is. Okay. Andrew’s new book, The Cold Start Problem: How to Start and Scale Network Effects, is out now, and I suggest you all pick it up. Andrew, thank you so much for your time today.
Andrew: Thank you for having me.
Des: Thanks a lot.