a16z Podcast: Not all Network Effects Are Created Equal

Hi, everyone. Welcome to the A6 in Z podcast. I'm Sonal. Today's episode closes out our special series on network effects. You can find the other pieces at ONAsics and Z.com and we're wrapping up by talking about the taxonomy and types and evolution of network effects. Joining us for that conversation is James Currier, who's a co founder and managing partner of NFX Guild, which is an early stage accelerator that runs a program for companies with network effects, which can range with all kinds of businesses. Also joining us for that conversation is a 6 NC partner, Anu Hariharan. And we kick off this conversation, which was recorded months ago, talking about the history and evolution of network effects. Before we break down the taxonomy of network effects and why all network effects aren't equal and why the language for figuring out the differences between them matters.

The first time it was used was 1907 in a annual report by AT&T. And they started noticing that because they had a lot of people connected to the telephone network, other people wanted to connect to the network, and that each of their customers got happier when they got a new customer because they could now call more people. And they said, because of that effect, AT&T said, I think we have a pretty good business here. And that was the first time someone coined the term network effects.

Okay. Because the first time I heard about it was, I think, when it came to the coinage of Metcalfe's Law.

Correct. Yeah. So that was during the Ethernet era. Right. I think there was a gap between when AT and T used it in their annual report. And then Ethernet cards came into practice. If you look at Ethernet Card, it's interesting to note actually network effects began with hardware. It was first the telephone, then the Ethernet. And in today's world, we talk about it in the context of software companies like Facebook. Right. So Metcalfe's law in Ethernet Card, the way it came about is the cost of the printer is really high. And so the Ethernet card actually enabled various people to connect to the printer, which reduced the cost of the device. And that was an incentive for different groups within a company to even start adapting the Ethernet cards. Because the value proposition was you could subsidize the cost of the device and you didn't need a printer for each person. So that's how Metcalfe's Law was born, which sort of became the measure for network effects.

Okay, so just to clarify, by Metcalfe's law, you mean. I mean, it's not really a law like a lot of These laws, it's really just a description, like a descriptive formulation that the power of a network doubles as n number of users get added to it.

I think it's exponential increase in value as each user joins the platform.

Okay, there we go. And it wasn't even him who did this. I think it was George Gilder who actually described this phenomenon on. But it's that the value of a network increases exponentially to square the number of connections in the network.

So that's Metcalfe's Law, you know, from 1907. It took until 1974 until academics started looking at this, because they started to see these effects in particularly in technology businesses. And Metcalfe's law came along a little bit after that. And then, you know, the network effect of the Microsoft operating system came to the fore at the end of the 90s, when the Department of Justice was looking at whether that network effect that they had between their operating system and all of the operating apps on top of it was just too much and crowding everybody else out.

And I think that is the important nugget, which is like, if you build network effects, you do become the dominant player in the industry for a while. And that's really why today, when we look at companies internally from an investing standpoint, we try to figure out a whether they have network effects or do they have a pathway to achieving network effects. Because it's a big differentiator from the rest of, you know, from the competitive landscape.

Network effects are one of the few ways we have left for software businesses to have a moat, to create a moat around their business so that competitors can't eat away at their margins.

Yeah. So in the olden days, there was lots of defensibilities that companies could have. You could have a port in the right spot, you could have a iron mine in the right spot. You could have ways of defending your business. But as we've moved into the software realm, we've eliminated distance, we've eliminated time, we've eliminated a lot of things that used to give people defensibility. And we're really left with only four major types of defensibility.

What are those four types?

Well, the ones that we still have left over are brand. And there's actually three types of brand defensibilities. We have scale, like you'd see with an Amazon. Right. They're just so big, they have such lower prices, it's hard to compete. And then you have what we call embedding, and that's typically used by companies in the B2B SaaS space where you embed your software into another corporation's operations to the extent that they can't take you out. And once you're in there, you're defensible. So again, these are the four defensibilities, and all of them are valuable and all of them are still working in the software and the digital world. But the way we look at it is the network effects is really the only one that's native to what we're doing, and therefore it's the most powerful.

When you say native, there is some power in that word. So when you say it's one of the few native ways.

The reason why I think we refer to network effects as being native right now is really the evolution of the Internet from web to mobile. And I usually think of it as the pre 2000 versus the post 2000 era, because Facebook came post 2000 and social really became predominant. Facebook is not valuable if you're the only user on the platform. And it was not just social connections. Right. You find news through Facebook, you find content through Facebook, you find events through Facebook. And so it became easier to derive that value. If you compare that to telephone or Ethernet, you still needed a piece of hardware. There was more friction to get more people to adopt the hardware. That changed then from the Facebook web era, it changed to mobile. That's how you saw all these messaging, even photo sharing apps. It became even easier because mobile phone enabled you to take pictures. You could communicate via images. It was easier for platforms to build a network, I would argue, on mobile more than web.

I don't disagree. I would maybe take a different shot at it, which is that it's native to technology in the sense that the technology wants to connect. So every Ethernet switch connects to every other ethernet switch. Printers are connected to printers. Mobile devices are connected to mobile devices. So ever since the 70s when we started using chips or 60s, it was endemic to these technologies that they wanted to be connected. And once connected, they become networks. And once networks, you have the chance for network effects. And other cars, other products like cars or toasters or houses or whatever aren't natively connected physically or through information sharing. So the.

Like an inherent property.

Yeah. So everything that is silicon wants to connect to everything else that's silicon through these little wires that we have. So it's not. It's not a. It's not unusual that we would find the first network effect come when we first had our first wires, which were the copper wires that connected the telephone system.

I think actually, Doug Engelbart was the first to make that observation about silicon wanting to connect. Anyway, I think that's an interesting observation.

Everything that we touch is connected, and you can't really have a network effect unless things are connected. Now, there are types of network effects that are outside of the technology, which we can talk about.

Let's actually talk about types of network effects because I think this is a good opportunity to really unpack these terms.

Yeah. So academic literature I know tries to classify network effects at the highest level into four types. Right. One is direct network effects, which Facebook is a classic example of, where the organic consequence of the product itself is that you need to have connections with other users and therefore you derive value from the network.

All of this is just language to try to help give us insight. Right. Insight as we build these businesses, insight as we analyze other businesses and where we want to take these companies and these experiences that we create for other people. So this is, you know, the language is evolving and we're changing it all the time. We sometimes use different words to try to piece it apart, but we actually have identified nine different types of network effects. As Anu pointed out, the first one, the first cluster of network effects is what we call direct network effects. You then have what we call, instead of calling it local, we call it personal because there's a personal relationship that the human beings have to each other that drive their deep adoption of a particular platform. And Facebook would be a good example of that. But it's not just personal. I don't get a lot of utility out of Facebook. It's really sort of a media. It's like instead of watching tv, I use Facebook, but I use Facebook messenger to do useful Things. I use WhatsApp to do useful things. I've got to pick up my kids at school. I need that done, and I'll use those two services to get that done. Now, what's happening with Facebook messenger is they're gonna adding payments and they're adding all sorts of other the utilities. And I think that they, from what we can tell from their behavior, have realized that the personal utility direct network effect is actually more powerful than a regular personal direct network effect.

Well, that was actually the big point of the primer we put out around WeChat. It's not the social connection, only the personal connection that you're describing. But the more utility you drive, the more people will use it. Why does that distinction matter? Like, why would you go that far in the taxonomy?

It matters because you're going to get a different level of engagement in terms of activity. So how many times you use it a day. And you're going to get a longer life arc of that product. Right. So we all still, we all had telephones between 1900 and 2014, and now we're on the mobile phones and we have a personal number as opposed to being attached to a phone. So it didn't change for 115 years. Facebook may be around in 10 years, but it might not be. But it's pretty clear that messaging will be around because we're always going to be able to scan text with our eyes. That's a good interface for us. And so messaging is probably a longer term thing for a network effect business than just a personal connection.

And there's a fourth one, and there's.

A fourth one which is what we've called market networks. And a market network is a personal utility. But there's money involved. So there's an even greater reason for you to want to stick with and use it because that's where you get your money. You have a network, a personal direct utility network where there's money being transacted. And so now you have even more reasons for you never to leave that because, hey, that's where I get my income.

Do all marketplaces have network effects by definition?

So I don't think that all marketplaces have network effects by definition. They have the potential, but different marketplaces are at different stages of development in their evolution. So a marketplace that is new and just begun and say it's six weeks in or 10 weeks in, they don't have sufficient liquidity, someone could start that down the line or even a few weeks from now and they may have stiff competition. A classic example, and we've talked about this before as well, is Airbnb. The first three years of Airbnb was a real slog because it's a global marketplace and they were trying to build supply. At the same time they were also trying to build demand. And they needed to sign up the homes, they needed to make sure that people trusted the marketplace and felt secured to make a booking. But three years into until then, the growth was really sluggish and you can almost see it in their graphs that they share the room nights book just started growing enormously after the third year because they had sufficient liquidity, both supply and demand. And it started working.

Then we've got three that are more two sided network effects. And one of them is, let's say where Microsoft was in the 90s, which was they had what we call a platform two sided network effect, where you have a piece of software that everyone wants to plug into and then you have developers or other companies that plug into your business so that everyone kind of has to use it because otherwise they don't get access to the other side of the platform. We thought that the word platform described it better. It made us understand what was going on better so that we can measure are the developers actually writing software to that platform and have enough developers adopted that platform so that we feel like they're going to have a chance to win.

What I find fascinating about this conversation, and also as the person who loves words for a living, is how. Why the words matter.

What we're doing is we're looking for playbooks to grow the companies with the network effects. And so we need to identify which type of network effect are we going for. And then we run the playbook on the growth around that. So now we're up to the two sided network effects and there's three of them. One of them we call the asymptoting two sided marketplace network effect. It means it doesn't keep going, doesn't get better. Past a certain point, we're capable of seeing someone come in and compete with them pretty effectively on a lower price. The eighth and ninth network effects, which are again now less strong than the direct network effects or the direct network personal utility effects of the center of the core is what we call bandwagon effects. Literally just think of it as popularity. So the most recent good example of that is Slack. They do have a real network effect. They do have a real personal direct utilitarian network effect, but they also have a bandwagon network effect, which is if you're using hipchat, it's kind of uncool. The bandwagon effect and the language effect are actually the network effects that exist between people without any software, without any technology actually connecting. Which is, if I come to you, Anu, and say, you know what, I actually think that hipchat's better, you're going to look at me cross eyed and say, dude, you're not cool.

You know, why aren't you on Slack? We would say that. Cause Slack's a portfolio company.

God damn it.

No, I'm just kidding.

Right?

So that's a social reason why it's hard for me to use another network and I have to stick to the network that everybody else has already adopted. And when I use it, it then benefits everybody else for real network effects. But there's a bandwagon effect which comes first that actually is impactful.

Just one quick question though. Because when you talk about different types of defensibility that companies have in this from a previous age and in this current age, that's just brand.

Correct. But it's important to tease out the difference between brand and a bandwagon. When something's just getting going and there's everyone wants to get on it for social reasons, that's before it has a brand.

A lot of people make this mistaken assumption. My business is growing just fine and adding all kinds of great value. I don't need any sales and marketing. So it kind of is an interesting. I like that you're distinguishing this because it does make an argument for why certain levers need to be pulled at certain points.

That's exactly the point. The only reason we changed this language around is so that we can figure out what playbook to run, what feature to build. How much should we spend on the next feature? How much time should we spend on trying to do it? That's it. That's the only reason we tease these things out.

And I think in the bandwagon, also tying it to the viral growth point. It's not sometimes all of a sudden. So the product has to be good for the viral growth. But the key influencers also matter. Right? Like who is using the product. Going back to your point, who is using the product, who's actually promoting the product? If you look at the original, when they launched Instagram, the people who tweeted that here is the new app that is Instagram actually influenced a lot of people to sign up.

Yeah. This point about, you know, who starts it to create the bandwagon effect matters for almost all the big networks you can think of. There was a guy named Scott who started Rise R Y z e in 2000, which was LinkedIn, two years before LinkedIn.

I don't remember that.

But he wasn't Reid Hoffman. And Reid knew the top 4,000 people in Silicon Valley and emailed. And then once we were on, all the guys in New York wanted to be on. So where these networks start matters. Look, there was three college social networks that came before Facebook, but it was the first one started at Harvard.

Besides, where. What about when, when does it actually reach a point? You know, you have a network effect.

I think these businesses start with network effects by their design.

Right. So you're saying it's baked in. From the get go, it's baked in.

And then the question is, when do the network effects kick in so that you can tell it's going to be a $10 billion company, when do we.

Know it is at that point that it is winner take all.

But it's very hard to determine the tipping point. And I think if you ask the companies themselves, they would tell you that they don't know when it tipped. And there is no single measure of network effect for a company and different value to the users. At the end of the day, you just want to make sure as there are more users on the platform, is it becoming more valuable? How do you measure what is valuable? And that's different depending on which company.

And also the competitive set. Because if there's three people going after the same thing and they're all growing quickly, you're like, whoa.

Yes. Especially these days with mobile on demand services where you have a company in San Francisco and other company in New York and both are growing really well in their respective cities. You know, that's not really a network effect yet. Well, it's a natural effect.

It is a network effect. It's just not one you want to invest in.

So let's actually talk about what happens when they don't work. I mean, let's take the case of like MySpace and even a case like Pat, because that to me felt like. I mean, granted it was capped at 150, but like, why wouldn't that. Why didn't that work?

Let me talk about the MySpace example. I think one of the key things about network effects is you need to have product market fit. Like, you can't just build network effects and people are not just going to use your product without that. The challenge with MySpace was it allowed you to sign up anybody. It allowed you to have connections with any, even strangers or, you know, personalities with the page or influencers. And so you don't have a daily communication with them. And I think that was. There are many other reasons that I cited, but that is one of the key drivers for why I think MySpace did not succeed. I don't think they had the strong retention that Facebook did as well. Facebook from day one, measured their repeat usage and that kept growing.

Our take on MySpace was that they didn't actually have a personal network effect because they didn't have real names. And in fact, what happened with Facebook is that they were the lucky ones who got to real names in a social context first. So LinkedIn got their first, Facebook got their second. But no one at that time was willing to accept real names in a social network. Everybody was a pseudonym. MySpace, high five. Tickle social networking. We had 30 million people before Facebook ever launched. But they had real names. And so then it becomes a personal direct network effect, which is stronger than just a regular network effect. Now I would argue that MySpace became entertainment. It became like a TV show you're going to watch for two or three years. It didn't become anything with utility. So when you look at Facebook, Mark Zuckerberg on year three, changed the homepage to say Facebook is a social utility. He knew that he needed to go toward utility in order to get a real network effect. Real names and real utility was going to provide you with that foundation. And without that, MySpace withered away like any TV show does.

And in fact, Facebook in the first two years, I believe, when they were trying to get the students to sign up, called it as a student directory, because these universities didn't even have a good student directory where they could actually look up and see who else was taking this course.

Well, that was the origin of Facebook because there was a physical Facebook and.

That was the utility for students to sign up. Right. It was a utility even from day one.

Right, that makes sense. And then how about the case of Path? I know that it had a built in cathedral, but then at the end of the day, my WhatsApp group only has like 20 people in it and it's incredibly engaged compared to my Facebook group.

Yeah, I think for WhatsApp they did crack the utility piece which helped them build product market fit. Yes, they hacked the phone, but that's not what the most exciting piece about them. The most exciting thing about them is they targeted international communities where SMS is expensive and WhatsApp was free. I remember I was one of the early WhatsApp users because I had to communicate with my family in India and I was paying steep amounts being in the US per minute while WhatsApp was free and I could engage with them on a daily basis.

It's actually a dollar for a year. I remember that's when I paid a dollar when I signed up.

Yeah, I don't remember paying that. But the international community signed up because it was an alternative option to sms and it was essentially free. And they used the phone book and I think they kept the product really, really simple. And over time they had partnerships with carriers which pushed WhatsApp onto the phones as well. That helped their growth.

Going back to the Path. I think Path just suffered from competition. I think that once Facebook Mobile came out and Once you had WhatsApp, you had so many other ways of getting the personal utility out of your phone. Had there been no competition like there was for Facebook when They were growing for two years in colleges. Path would have been a big company.

Okay, so, you know, earlier I was saying, don't confuse virality with network effects.

A network effect is essentially for defensibility and retention, right? You continue to use a product that has a network effect because. Because of the network effect. It's not. A viral effect is different in the sense that it gets you new users. So the viral effect is about getting new users and the network effect is about keeping those users. And they're very, very different things. Now, a network effect type of a business often has a lot of interesting hooks, a lot of good raw material to develop viral hooks and viral paths. Right? And viral loops on top of. But they're different. And when you're building a network effect business, you build certain features and language and functionality, and when you're building your viral loops, you build different stuff on top of it.

And I think the key distinction, especially from an investment point of view we try to do is because is it viral growth or does it also have a network effect? And as James pointed out, platforms or marketplaces need to have network viral growth to have network effects. The other layer I would add is the result of a viral growth is that your CAC is pretty much zero. Right. Your customer acquisition cost is zero, which is great. We like businesses that are really growing organic and word of mouth, but then we also really care about do they have the potential to have strong networks? Defensibility and defensibility.

Great. Okay, well, you guys, thank you for all those insights. A lot of terminology, but I think it's important. Thank you for joining the ASICS and Z podcast, James. And thank you, Anu.

Thank you.

Thank you.