You're listening to episode 2, the underlying properties of network theory, from the network effects master class, here on the NFX podcast. To watch the course, visit nfx.com/masterclass. Okay. Let's start at the beginning. What are network So basic idea of network effects is that every new user of your product adds value to the other users of your product. That's a fundamental definition. There's lots of different network effects which fall into that category, but that's the basic idea.
Now most people kind of stop there. And say there are network effects or not. Just not the case. Most people don't understand if you, if you look closely at these things, you notice they behave differently. Okay. They act differently. They have different mechanics to them. How you would build your product to enhance them or to grow them is different depending on which network effects you're actually building. Fundamental properties of a network are essentially you've got links and nodes.
So a node could be a person and a link links them together. And now you could have a telephone be a node. You could have a person be a node. You could have a 3 sided marketplace where you've got 3 different types of nodes. You'd have 20 different types of nodes, and you got different structures.
You might have a network structure where everyone is an equal and they're all connected equally, you might have a hierarchical network structure where people are unequal, but they're still all connected in what is a network. Okay? You could have a 2 sided marketplace network structure where people are different, but they're all connected through your marketplace. So you have lots of different types of structure, but they're all constructed of nodes and links.
Now on those links could travel money, can travel status, can travel information. There's a whole bunch of things that move across these links and different networks have different things moving across those, those links between the nodes. Okay. And then the network has size, how many nodes are in a network, and then it has density. How many links are between the different nodes?
So for instance, a Spotify with connections between people and artists is probably pretty low compared to something like a LinkedIn where the number of connections between people is very high. People have 500 or 2000 or 5000 connections between people. So your density ends up being quite high. Those are 2 core differences between different types of networks. For instance, a Twitter network you might have 3% of people producing 90 percent of the tweets and having most people connect into them.
And then the other folks, they have a very low density connection. Now, once you get to high density, you get what we call the white hot center. So if there's one node where everybody's connecting to them, then you've got the beginnings of a white hot center with the core node and then the nodes that are closest to them. Another important aspect of these links is that they can be directed There can be direction to these.
So for instance, in Twitter, you would have an influencer who's important sending out mostly messages to people who are just reading. That would be directional in one way, or you could have, let's say, a Venmo where people are paying each other, that would be bidirectional. Network structures actually have different shapes to them. The simplest one, of course, is 1 to 1, right, where you could have a network where it's just one person or another person. Where 2 nodes are connected by one Flint.
Then you get into network structures, which look more like one to many. Okay. Think of a TV broadcast station in the 1900s or a radio station or a newspaper. And then you get into the many to many, which was really the innovation of the internet that came along in the 19 nineties where suddenly all the nodes could see each other, and it didn't have to be coming from one central source. What's also fascinating is that we then have laws that are associated with these different network structures.
For instance, in the 19 sixties, you had a guy named Sarnoff, who was sort of broadcast Titan. And so he understood this network structure, and he saw that the value of the network increased with the number of nodes in the network on a 1 to 1 basis. Metcalfe comes along in the 80s, and he was the father of the ethernet where he was connecting all of the computers together with the first ethernet cables and whatnot.
And he noticed that the value was not increasing linearly with the number of nodes, but it was increasing geometrically. So at a sort of geometric growth rate to the value. And then around 1999, along comes a professor from MIT and says, well, Metcalf is correct, but it's actually more intense than that.
Because if you look within networks, you have clusters of Pete, let's say, a parenting group Morgan book club or a soccer team, they are networking on your network and their connection to themselves creates its own very highly dense network effect.
And because they're dedicated to this network, the network's value actually increases at a much more geometric rate to the value of n. And so these things are directionally Currier, even if they're not precise at any one time, but you can actually calculate and map the growth of the value of Facebook to Metcalf's law in particular, and it's actually not not too far off.
So these are useful ways of understanding how much value you're creating by how big you build the network and how much value is at the end of your process and why you should be paying attention to this. Now, This also shows you that over the last few decades, we've been looking for ways of understanding and explaining network because we're just at the point of being able to see them. They've always been there for 1000 years, but we now are able to see them and measure them.
And you, just by watching this master class, are actually participating in the continued discovery and understanding these networks and how they function. And so the prize for you to understand this and use them appropriately is very, very high because of this geometric growth pattern of the value of these networks. How do you build that? Well, you've got to get to a critical mass. Okay. You've got to actually get your network to tip.
Now one of the big ways of doing that is by looking at your clustering coefficient because not all nodes are created equal. You're gonna wanna focus on very particular types of nodes and get them to cluster, get them to create their own network effects, and then bridge them to other networks in many cases. That will help you get to a tipping point where the network is growing and growing, but then at some point, the value starts to take off almost by itself.
But getting from here to here is your job as a founder, and it's not easy. Okay. So figuring out your path to that, which we'll talk about later with network bonding theory, figuring that out and building these clustering coefficients to get the network to take off is one of your jobs, and that's what we want to talk about. So to build the network value, you need to get to critical mass.
And what happens is as your network is growing a little bit, a little bit, a little bit all the time, at some Flint, it hits a moment of critical mass where it takes off. It lifts off. You can feel it. It's like, someone's stuck two fingers in your nose and just yanks your head forward. It just starts to take off on its own. And that critical mass moment is often facilitated through developing your clustering.
We just mentioned how Reed's Law really focused in on what the clusters are inside of your overall network. Not all notes are created equal. In fact, they're really created differently. You need to focus on the different types and cluster them together and then bridge the different clusters to each other on your way toward reaching that critical mass. So being able to see and measure and work with your clustering coefficients on the different groups is actually an important skill.
Now to get these clusters going, you're gonna wanna find your minimal viable cluster. Some small group of people that once attached to each other will stay attached to each other that you can then build on. Again, maybe creating another cluster and then bridging those 2 clusters. That minimal viable cluster really helps you to found the overall network. Now To get to critical mass, the product is gonna have to have some value to these users to this cluster.
But at some point, you're gonna notice that the value of the network exceeds the value of the product. Your software, your system is gonna provide some value but then eventually the network takes over in terms of where the value is coming from. And that's when you've hit critical mass. That's when those two fingers get you in the nose and pull your head forward. You'll feel So we just talked about some network structures. Let's talk about some more network properties.
So one big one is whether the nodes on your network have their real IDs, whether they're synonymous or whether they're anonymous. And While there are applications for each of these, what we've tended to notice is that the more valuable and lasting networks have tended to be real ID type of networks. You're gonna have to make this choice and realize it's just a very big choice, what you do. There's one network that's the Currier, which has both, which is Instagram.
Pretty interesting, but pretty much everyone either goes one way or the other and one of them has done both. I would also say maybe Twitter particularly with all the bots is is sort of in between, but as those get cleaned up, we'll have less and less that'll move toward real ID. Our guess is that a lot of web 3 is gonna toward real ID in most applications as well, because that tends to be long lasting. We'll see how it goes.
Another Morgan network property is to understand how homogeneous or heterogeneous it is. For instance, you could have a marketplace where These buyers over here have very different needs than these buyers up here. And you have to treat them very different. And Morgan you will see that, for instance, in Uber, the drivers are very similar. And the riders are very similar. So your homogeneity in that network is very high, whereas Upwork where people are looking for different types of labor.
They have different needs of those different types of labor. You've got a very heterogeneous marketplace. And sometimes it's hard to develop software to satisfy all their needs. Conversely, homogeneity allows you to grow your network faster but then often leaves you more vulnerable to attack by other competitors and other networks simply because people are interchangeable. You think about Uber. You think about Lyft.
You think about businesses where it's it's, the needs are very clear and obvious. But it helps you grow faster at the beginning. So it's a trade off. Another single network property is what we call Asymptoting. And Asymptoting takes place when you might be adding more nodes onto the network, but the value doesn't actually continue to grow it.
So the value of an Uber network or a Lyft network Asymptotes, even though you keep getting more drivers on the network because if you're putting on your jacket or getting your bag or you need to go to the bathroom before they come, you don't really care if they come faster than 4 minutes. It's nice if they come at 3 minutes, but as long as they come in under 10 or 6, it's it's pretty much good. And so the value doesn't actually get much higher.
It adds some totes at some value at about 4 minutes. Okay. That's an example of asymptote.
Another example of asymptoting is often in the medical area with a data network effect where you might get 25,000 cases of kidney disease, you get all this data, your algorithm is able to spot that cancer really, really well Beller than most doctors, let's say, But the fact is a competitor can come along and put in 8000 kidney cancer cases and get their algorithm to be almost as good as yours because your algorithm is 98.6 percent accurate, and theirs gets it to be 97.2 percent accurate.
With 8000 cases instead of 25. So you can't ever get ahead and stay ahead. Now there are other networks like a Google, where the net new searches per day is 20 or 30 percent, somebody's always coming up with something new to search for. And so that's one of the reasons Google continues to be worth a $1,000,000,000,000 is that that long tail doesn't really asymptote, or is it what does with an Uber or with medical data network effects.
So another important network property is what we call asymmetry. So think of a marketplace, let's say, which has a marketplace network effect, and your, one side is going to be harder than the other. So let's say you've got your basic marketplace, and you've got your supply on this side, and you've got your demand on this side. In most marketplaces, one side is harder than the other, in some James, much harder than the other. Example, outdoorsy doing an RV marketplace.
They discovered that the supply side was much harder to get because there was, let's say, 2500 mom and pops renting out about 50,000 RVs in the United States, but there was 30 million people trying to rent those RVs. So if they could build software to allow them to work with and understand where these were, they would be able to offer that supply out for the voracious demand that was out there and build a pretty good marketplace, which they've done.
The opposite would be something like a lending club where the number of people willing to loan money to people who want to borrow it at, let's say, 18%, is really high. You can find them in Baltimore and in DC and in New York, in Europe. You can find them all over the place to loan.
So the supply side was easy to get, but what was hard was to find consumers who are credit worthy, who are currently paying 26 or 28% of their credit card companies who now wanna borrow and replace that loan with a loan for 18%. And once you find those people and they were hard to find, then the supply side of the money would show up right away.
So what you'll notice here is you've got what we call a cross side network effect, and you can use that term cross side network effect to help understand and explain amongst your team what you're looking at. The two sides in this case are the supply and the demand. So the value is the more supply on the network the more value to the demand side. So that's cross side network. Correct. Cross side value being created, and that can be quite intense. And that's pretty intuitive about how that works.
Sometimes you actually have what we call same side network effects where the value the demand side actually increases. So you might, like, think about a group on or group buying on Pinduodua in China, where the more people you have buying the more the suppliers, the grouping together, the better the price you're gonna get from the suppliers. And this happens pretty often.
Excel is another example where the more of us use Excel on top of the Microsoft platform and can exchange files between Excel, the better off we all are. There's something else you need to be careful of when building your network business, which is negative network effects. Now some people like to go on and on about these, and generally negative network effects only take place 6 or 7 or 10 years into your business.
However, when designing You need to avoid them at the beginning so that you can get to critical mass. You still need to avoid negative network effects. And there's basically 2 kinds. One is what we call congestion. And that means that there's so many nodes on the network that are trying to use the network, that the network starts to break down.
A good example of that would be Ethereum, where the more people using a theorem, the more the gas prices go up, it just becomes, you know, unaffordable to use that network. And there's also what we call pollution. Which is more about what is happening between the nodes. So congestion could be there's too many nodes. The pollution is when the nodes on the link are behaving badly.
So an example of behaving badly could be on Facebook when your grandmother gets on she starts posting pictures of you as a kid and embarrassing you, you don't want to be on Facebook anymore. That's network pollution. Another example of network pollution would be on, let's say, monster.com, where work from home Pete, come on and start scamming all the people who are on monster Beckham. You have to clean them off of that. That just makes the whole that degrades the whole experience for everyone.
To have these sort of, you know, the underbelly of the work world existing on Morgan. And you see this on Twitter at the trolls. This is network pollution. Too many people, too many crazies, and they end up dominating. Right? And so you have to figure out software solutions and network solutions to tamp down the negative network effects, the pollution and the congestion that's going to start to happen, even sometimes early in the growth of your network business.
So on building your network business, you're gonna encounter a bunch of obstacles. The first one, if you're building a marketplace, it's called the chicken or egg problem. If you're building a direct network effect business, It's known as a cold start problem. But the basic idea is that you can't get anyone to join your network unless there's already somebody there.
So if you want to build, let's say, a marketplace, you have to get enough supply sitting there so that when you bring in your first demand, number 1, they have somebody to look at and see if they wanna buy their stuff. Okay. But how do you get them to sit there long enough? How do you attract them there at all? And then once you get them there, how do you attract enough of the demand so that these people stick around? There's a timing issue with bringing people in at the right time.
This is known as the chicken or egg problem and you might call the cold start problem the same situation, but when you've got a network and there's nobody there yet. So when the first person comes in, there's no one there. How do you teach them to stick around so that they are there for their friends when their friends arrive. Okay? We've actually written about 19 different tactics that can be used to break the chicken or egg problem or the cold start problem.
And when you've solved the chicken or egg or the cold start problem, what you have, you have what we call liquidity. This is a word that's very useful to that you can actually measure, in your networks. If a if someone Flint, something for sale on your marketplace, What percentage of the time do they sell it? What percentage of the time do they get the transaction they were looking for? If someone posts on your social network, What percentage of the time do they get likes? Do they get attention?
Do they get responses, comments? Okay. You can actually look at liquidity both in terms of transaction on marketplace and in terms of interaction on a network. So the second thing I want to mention that you're gonna face as you're growing your business is multi tenanting. So if you've got a spline, you've built your marketplace for the demand, they might be using your marketplace. But like with Lyft or Uber, I could sell my services on somebody else's marketplace.
Number 2, or maybe marketplace number 3. This happens if you're selling on eBay, you also wanna sell on Amazon marketplace or on Shopify or on Etsy. And this happens all the time. Same thing is happening with Uber, and this is called multi tenanting, and it makes economic sense. Can you build a system?
Where either or both the supply and the demand do not want a multi tenant so you can capture their share of wallet, capture their average order value, capture the overall annual take that you're gonna be taking from them. That's one thing you need to think about when designing your marketplace is how to keep them for Pete tenanting. Alright. Now I might also multitenant on Instagram or snap or on Facebook.
I I'm I'm socializing on different platforms Each of those platforms wants to capture much of my time so they can put ads against my time and make more revenue. Everyone's facing this multi tenanting challenge. And then the third thing to mention, of course, is what if someone just decides to go direct? And disintermediates you and the other marketplaces.
So in an Uber example, if I'm an Uber driver, I can give you my card in the back seat the next time you wanna ride to the airport, you just call me direct. And I don't have to pay the 30% to Uber. Right? Or if I'm a big enough buyer, on Amazon marketplace, you eventually get my phone number, and you just start shipping me stuff directly. I don't have to go through Amazon marketplace and have them take their their rake. Okay. So that's disintermediation. And this happens all the time.
There's lots of ways of solving this. I know in higher.com's example, they were trying to figure out if someone one of the engineers was being hired, and the employer was not notifying hired.com. Because they were gonna have to pay $15,000. So why would they tell them? So what they would do is they would send a bottle of wine or champagne to the person who got hired and say congratulations. Who hired you? And then they would tell them who hired them and say, hey.
I know you just hired that person. Give me my $15. That's how they solved the disintermediation for a time. Now a couple things to first know. Network effects are not viral effects, and most people mistake this. Okay? Viral effects are where near existing users, get you new users for free. There are playbooks for that, and that's great. That's not what we're talking about. Network effects are about retention. Network effects are about defensibility.
Okay. And the defensibility is what ultimately adds value to your product. There have been tons of companies with lots of viral effects like Myspace or Chat Roulette or daily Booth that had lots of viral effects, but no lasting retention. And therefore, their companies didn't produce a lot of value in the long term. Wanna talk about the network effects that produce value. Okay. So we're not talking about viral effects.
So two concepts that aren't network effects, but are related to network effects, They're just useful for you to understand and be able to talk about with your co founding team. 1 is just linear growth versus geometric growth. We all kinda know what this means. If you are growing every month like this, that's growth. And sometimes you can be happy with that. But what you want is geometric growth where you start to see a growth rate that goes up and eventually gets very big, very quickly.
What I would encourage you to do is to notice whether you are growing linearly or geometrically in every aspect of your business. Because that helps you engineer how you're approaching building these network effects. Okay. And eventually, I really wanna push you to grow for the geometric growth. If you don't have geometric growth in a startup, it's it's not necessarily a startup. It's okay. You can be encouraged, but until you hit that, You're not really doing what you need to be doing. Okay?
Really focused on that geometric growth. Now the concept of geometric growth has been highly popularized and discussed around the concept of viral growth. Okay. So let's talk about that for a second because, again, viral effects and viral growth is not network effects. Right? Viral growth is where you get your existing users to get you new users for free. That's the idea. We have lots of playbooks for how to do that.
We're not gonna talk about them in this master class, but you can measure this, what's called the K factor, by the number of invitations that your existing users send out times the conversion rate of the person being invited over the time. Okay. For instance, with Evite, they had a viral delay. Their time about 6 months between when you were invited to a party and when you would use Evite to invite other people to a party. Because of that viral delay, it just never got big enough.
Too many competitors came in, and it never became a big company because of the viral delay. Whereas sometimes you'll have something like a birthday alarm, which was started by Michael Birch years ago, where you would get an email that says, James wants to remember your birthday, and you'd click on the Flint, you'd land on the page, say put in your birthday, now get your own birthday alarm. Set up, and people would use it immediately.
And so there was no viral delay, and so the growth rate of that could be measured on a daily basis, like Beller 24 hours, how what is the K factor? How much have we grown compared to yesterday? Now if you get a k factor of 1.01, you will for for many days in a row, you will eventually start growing very geometrically. Okay. And if you can get a viral factor of 1.25 back in the day, maybe 10 or 15 years ago, it was possible to get viral factors high as 3.0, meaning you would triple every day.
And in many cases, we were registering 250,000 people a day. Okay. I know of one application that was registering 2,000,000 people a day on top of the Facebook platform because there was no viral delay. The invites were going out like crazy. The conversion rate was really high because people were already on Facebook, one click install the app easy.
Those conditions don't exist today anymore, but understanding viral effects understanding virality, how to measure your K factor can be useful to lowering your CAC today. I doubt you'll be able to get over 1.0. But it's possible if you're really talented at this and you iterate. Now I'll also mention once you cross 1.0, you will not stay there.
You will either float up to 1.13,1.4, but then you will come down as the conversion rate degrades, or the excitement of your users to invite more people degrades over time. You're gonna have to constantly reinvent that viral engineering. Okay, is not something you do and then you're done. It's it's like you're a shrew. You're constantly eating and burning energy. Okay. You're like a small mammal trying to stay alive all the time. It's gotta be your mentality. If you wanna do viral engineering.
Now this has been confused with network effects for 2 reasons. 1, It has geometric growth just as network effects adds geometric value to your network. So they're both geometric properties with a good feedback loop. So they kinda smell the same. The other reason that they've been confused is because both viral effects and network effects have been connected to many companies that we all know about like Facebook and Twitter, etcetera.
So for 15 years, It looked like they were happening together, and people weren't looking hard enough at it to tease them apart to really understand the mechanics of the different playbooks to get each of them to happen. And so a lot of people are out there confused and not really understanding it. So that's why I'm breaking it apart for you today. You can have a company with network effects, with 0 viral effects.
For instance, I could pay $10,000 per node on a marketplace for some, you know, hospital equipment manufacturing market network. And I could pay $10,000. I could get a thousand people buying, and I could get a thousand people Beller, and I would have a network effect that would keep them all there buying and selling with each other for a decade. But I didn't have any viral effect. I've had $10,000 per node. Okay. The same thing can be true for when you have viral effects.
Like we did on chat Roulette, where millions of people joined this thing, lots of people took their photos, lots of people did videos, was an excitement. All the networks were talking about it. Wall Street Journal was talking about it, but nobody stayed. There was no reason to continue to get value from the network. It was a novelty. Very viral, but only a novelty, and people would leave high k factor, low value.
So you could have a a lot of viral effects without any network you can have network effects without any viral effects. They're very different things. Stay tuned to the NFX podcast. As we'll post 1 episode per week until we complete the course. You can also watch this entire master class online at nfx.com/mass class, where you can log in, track your progress, and watch full videos, retranscripts, and find other related material. Thanks for listening to the NFXpod guess.