This essay isn’t about network effects per se. It’s about the end state that they can enable: Defaults. (View Highlight)
Network effects can be broadly divided into three different categories: Direct, indirect and data network effects. (View Highlight)
Direct network effects – as the name suggests – occur when the number of users has a direct impact on the value of a product. A telephone, for example, is worthless if you are the only person who owns one. But with every additional user who joins the telephone network, the number of people you can have a conversation with goes up and therefore the overall value of a telephone increases. The same dynamic is true for chat apps, social networks and p2p payment systems.
Indirect network effects are a little more complicated. They occur in two-sided (or multi-sided) products where the size of one user group affects how valuable the product is to another user group. Operating systems are an often cited example here: The more users an OS has, the more attractive it becomes for developers. More developers and apps, on the other hand, make the platform more valuable for users, thereby generating a positive feedback loop between the two sides.
The third category, data network effects, describes products which become better with more users via the data those users generate. Waze is a popular example of this category as are products that are powered by machine learning. More users lead to more data which lead to better recommendations or predictions and, thus, a better user experience. (View Highlight)
The world’s most successful companies all exhibit some form of structural competitive advantage: A defensibility mechanism that protects their margins and profits from competitors over long periods of time. Business strategy books like to refer to these competitive advantages as “economic moats”. (View Highlight)
One of the most citedtypesofmoats is the concept of network effects. Network effects occur when the value of a product or service is subject to the number of users. A positive network effect means that a product or service becomes more valuable to its users as more people use it. (View Highlight)
Network effects are extremely hyped and have become a bit of a meme in recent years among tech entrepreneurs, investors and policy makers. The five largest companies by market cap in the US all seem to be built on some sort of network effect and some even go so far as to claim that 70% of tech value creation since 1994 is predicated on network effects. (View Highlight)
New highlights added February 14, 2024 at 12:31 PM
Direct network effects – as the name suggests – occur when the number of users has a direct impact on the value of a product. A telephone, for example, is worthless if you are the only person who owns one. But with every additional user who joins the telephone network, the number of people you can have a conversation with goes up and therefore the overall value of a telephone increases. The same dynamic is true for chat apps, social networks and p2p payment systems.
Indirect network effects are a little more complicated. They occur in two-sided (or multi-sided) products where the size of one user group affects how valuable the product is to another user group. Operating systems are an often cited example here: The more users an OS has, the more attractive it becomes for developers. More developers and apps, on the other hand, make the platform more valuable for users, thereby generating a positive feedback loop between the two sides. (View Highlight)
The third category, data network effects, describes products which become better with more users via the data those users generate. Waze is a popular example of this category as are products that are powered by machine learning. More users lead to more data which lead to better recommendations or predictions and, thus, a better user experience.
(View Highlight)
There are two other concepts I would like to briefly introduce you to: Switching costs and multihoming costs. Switching costs describe how difficult or expensive it is for a user to switch from one product to another, whereas multihoming costs explain how easy or likely it is to use multiple competing networks simultaneously.
Both of these costs are not necessarily monetary – they can also be psychological or time/effort-based. The higher the switching and multihoming costs, the more defensible a network typically becomes. (View Highlight)
What gets people so excited (or worried) about network effects is not just their defensibility though. It’s the idea of reinforcing feedback loops and the resulting exponential growth that they enable. As any paper or blog post about the topic will point out, network effects create winner-take-all-dynamics where only one or two firms end up dominating an entire industry and can’t be challenged.
But is that actually true? (View Highlight)
First of all, it feels like many of the prototypical network effect companies are not as defensible as the literature suggests.
Take social networks, for example. Friendster‘s network effects weren’t strong enough to fend off MySpace. And MySpace – which many fearedwould dominate the internet forever – was quickly replaced by Facebook.
(View Highlight)
There is a similar pattern for chat apps: WhatsApp, iMessage, Telegram, Facebook Messenger and Snapchat all have at least 100 million users. Why doesn’t the market tip in favor of one or two of them as the literature suggests? Looking at the speed at which some of these apps have grown, one has to ask if network effects don’t perhaps have the exact opposite effect: Maybe they are not a defensibility mechanism for incumbents but a growth mechanism for new market entrants?
Another often cited network effect example is Google Search. Data network effects have made Google the best search engine on the planet. With each additional search query, Google’s algorithms become a little bit smarter and its search results better – which in turn attracts even more users and usage. This positive feedback loop makes Google unstoppable … and yet, Google pays Apple $15bn per year to remain the default search engine on iOS. Isn’t this a bit of a narrative violation?
Now, I’m not arguing that Facebook and Google aren’t extremely powerful companies. They clearly are. But I’m not convinced that their power is really based on network effects. Network effects might have helped them to get to where they are today, but their defensibility lies somewhere else. (View Highlight)
The second thing that I found interesting hit me when I looked at examples of industries that had indeed achieved winner-take-all (or winner-take-most) dynamics:
• Telephone
• Railroads
• VHS
• Smartphones
• QWERTY keyboards
If you study the list closely, you’ll notice that these businesses all have something in common. Something that the before-mentioned industries don’t have. They are based on atoms, not just bits. (View Highlight)
In contrast to bits, atoms have marginal costs. You can copy and paste a piece of software at virtually zero cost, but producing an additional piece of hardware has all sorts of marginal costs associated with it (material, production, shipping, …). This means that atom-based network effect businesses have switching and multihoming costs.
If you want to switch from Android to iOS (or use both OSs simultaneously) you have to spend at least a few hundred dollars on a new iPhone. Software, on the other hand, is typically free to use, so switching to a new chat app, social network or search engine doesn’t come with a real price tag. Because of their marginal costs, atom-based businesses have a greater lock-in effect and thus defensibility.
Another under-appreciated aspect that makes multihoming less likely is that atom-based products have physical space constraints. You can’t have eight competing rail road networks because there simply isn’t enough real estate to build them. Similarly, most consumers wouldn’t be willing to carry around more than one smartphone with them. Your pockets also have limited real estate. (View Highlight)
You may have heard about the concept of “value chains” before. Value chains are horizontal visualizations of all the value-adding business activities involved in creating a product or service from start to finish. They are great to analyze traditional businesses and industries but they are not a very useful framework for tech companies.
If you want to understand the power dynamics between different platforms, aggregators and other players in a tech ecosystem, it’s better to look at them as a vertical stack with different layers.
On each layer of the stack, companies are trying to create value and to capture value. The lowest layers of the stack are typically the most powerful. If you are able to take control of a layer, you can dictate the terms of most of the value creation and value capture that is happening in the layers above you. (View Highlight)
As a result, you see companies trying to
create layers on top of their business (everyone wants to be a platform)
move down the stack to get closer to the base layer (to increase defensibility)
What is the base layer?
The base layer is the final interface between the stack and the end user – which is typically an operating system tied to a piece of hardware. This is why atom-based network effects are so powerful: They help companies gain control of the most powerful layer of the stack. (View Highlight)
Google started as a simple website that allowed users to search other websites. Thanks to the superiority of its PageRank algorithm, more and more users started using Google Search, resulting in better search results and thus in even more users switching over to Google Search.
Thanks to this powerful data network effect, Google was able to move down the stack. Google wasn’t “just a website” anymore, it became an aggregator that commoditized all other websites and made them layers on top of Google’s.
The data network effect, however, is not the real moat here. It’s just a means to an end. The end goal is to become a default on the layer below. (View Highlight)
Let’s go back to atom-based network effects real quick. Earlier, we identified that they have two significant characteristics:
• Marginal costs (resulting in high switching & multihoming costs)
• Space constraints (resulting in high multihoming costs)
As we discussed, software doesn’t have marginal costs. The fact that you can copy and paste software at virtually zero cost has led us to a world of abundance where many things are free and infinitely available.
But what about space constraints?
Your initial reaction might be that software doesn’t have space constraints. We are talking about bits here, after all, not atoms. But if you look closely, you’ll notice that that’s not really true. Every layer in the stack has some sort of limited (pixel) real estate – and in a world of abundance, that scarce resource is extremely valuable. (View Highlight)
If Google wants to become a default on the layer below the search engine, it must find and occupy limited real estate on the browser layer. There are billions of websites out there … but only one default: Your browser homepage.
Becoming the default browser homepage became Google’s actual moat. There can only be one homepage (no multihoming) and users are typically too lazy to change it (friction = switching costs).
(View Highlight)
As you well know, Google didn’t stop there. It moved down the stack and successfully conquered the browser layer with Google Chrome – which obviously shipped with Google Search as the default search engine. More importantly, it turned the address bar into a search box thereby melting the browser and the search engine into sort of one layer.
With Android – one of the most underrated acquisitions of all time – Google then moved even further down the stack. It now owns a significant chunk of the world’s most important base layer: the smartphone operating system. Unsurprisingly, Android ships with Chrome and Google Search pre-installed.
If you think about it, it’s kind of amazing that the largest operating system humanity has ever seen only exists to protect an advertising business two layers further up the stack.
Unluckily for Google, smartphone operating systems are not winner-take-all but winner-take-most markets. Not only does Apple have considerable market share in the smartphone market, its user base is significantly wealthier and thus more interesting to advertisers. (View Highlight)
Apple is an interesting case study because its stack strategy has been so reciprocal to Google’s. It started with the base layer and then worked its way up the stack.
The first iPhone was released in 2007, but only in 2008 did the company launch the App Store. Opening up iOS to third-party developers created a new layer above the operating system: apps.
The apps layer is interesting because it created indirect network effects between developers and users, which added defensibility to the OS layer. That defensibility in turn meant that Apple was able to capture a lot of the value that was created on top of its platform. 30% to be exact.
30% value capture is great, but you know what’s even better?
100% value capture.
Over the last couple of years, Apple has increasingly launched and monetized its own apps (and services) on top of iOS. It has moved up the stack and is now competing with the third-party developers on its own platform.
(View Highlight)
In contrast to third-party apps, however, Apple’s own apps come pre-installed. They are defaults.
Apple realized that it owns some of the most valuable pixel real estate in tech: The home screen. And the best way to monetize that real estate is by occupying as much as possible of it yourself.
The beautiful thing about defaults is that they beat almost any competing product – even if that competitor has strong network effects or is technically superior.
This is why Apple Maps has higher market share than Google Maps, why Apple Music is able to catch up with Spotify, and why Google pays Apple $15bn a year to remain the default search engin (View Highlight)
It’s important to point out that not all of Apple’s default apps exist to increase value capture. That might be true for Apple News, Apple Arcade and Apple Fitness, but clearly not for apps like Maps or Health. So why do those exist?
Some of them add stickiness and switching costs to the iOS platform. iMessage, for example, creates network effects that make it harder for users to switch to a competing OS layer like Android.
Other default apps – like Maps – are solely there to mitigate risk from upper layer apps that might become too powerful. You’ll remember from the Google case study that Search became synonymous with the browser. And that the browser became the real operating system on desktop computers. That’s a scenario Apple wants to avoid at all cost. So instead of taking the risk that a service like Google Maps or Spotify becomes too dominant and threatens the base layer, Apple will launch its own (default!) maps and music apps.
This also explains why Apple doesn’t allow any competing app stores, browser apps (that aren’t built on WebKit) or game streaming services. Users are also not able to add 3rd-party apps to their iOS lock screen, change their default camera app or give quick access to a payment app other than Apple Pay. All of these limited real estate defaults are fully in Apple’s grip.
Apple Pay and Wallet are particularly important for Apple because they are tied to the user’s identity – which is another important layer in the stack. (View Highlight)
So far, the layers we have looked at were all pretty distinct. The Google and Apple stacks we analyzed are simply different platforms and applications that build on top of each other. But not all layers are distinct pieces of software. Some layers are blurry and don’t have a clearly marked place in the hierarchy of the stack. One of those layers is the identity layer.
Earlier in this essay, we described the base layer of the stack as “the final interface between the stack and the end user”. Operating systems like iOS or Android are great examples because there are no other layers between the smartphone and the end user. (View Highlight)
Identity is a crucial component of almost every single layer in the stack – especially if that layer is a network (which most layers are). Every network is just a collection of nodes and if you want to build connections between those nodes you need an identity layer. Even in a pseudonymous network, each user has a consistent identifier.
The fascinating thing about identity is that it’s kind of a stack of its own. When you sign up to a new app or service, you almost always use an existing identity. Until recently, that identity was typically your email address or phone number. Both email and phone numbers are great base layers for online identity because they are (sort of open) standards.
Unsurprisingly, in their attempts to become defaults, other companies have tried to create their own identity base layers. One of them is Facebook. (View Highlight)
Facebook is a typical network effect business: The value to its users grows as a direct result of attracting more users. On top of the core social network there are several other network effect loops including Messenger (direct network effects), Newsfeed (data network effects) and Marketplace (indirect network effects).
Facebook’s real stickiness, however, comes from another network effect. One that allowed the company to move a layer down the stack: Facebook Login
Facebook Login is a single sign-on service (SSO). It allows people to sign up to other social platforms and services using their Facebook account. SSO is more convenient for users and thus promises greater sign-up conversion rates for 3rd-party platforms. The more people use Facebook Login, the more platforms will offer it, thereby increasing the total number of accounts that a single user has connected to their Facebook identity. (View Highlight)
As a result, it becomes really difficult not to use Facebook. In contrast to Google and Apple, Facebook doesn’t own an operating system and, thus, doesn’t enjoy the defensibility of a pre-installed default. But because Facebook is the de-facto online identity layer for so many people, it is almost guaranteed to secure some of that limited pixel real estate on the user’s home screen.
Unsurprisingly, Facebook is not the only company trying to become the default identity layer – dozens of companies are. Which brings us to the question of “winner-take-all-dynamics” and defensibility in the identity stack.
(View Highlight)
We started this essay with the observation that neither social networks nor messaging apps have tipped in favor of just one or two winning companies despite the strong network effect dynamics in these verticals. Interestingly, both of these product categories are also very closely tied to identity.
Instead of winner-take-all-dynamics, we are seeing not just multiple identity aggregators (SSO services) but – more importantly – many, many different identity networks building on top of them. When you sign up to a new social app using Facebook Login, your (pretended) identity in this new social app is most likely going to be distinct from your Facebook identity – which is probably the reason you are signing up to a new social service in the first place. (View Highlight)
As a result, we see intense multihoming in the online identity stack. There might be clear winners in certain verticals (e.g. LinkedIn for “professional networking”), but “social” as a whole is not a winner-take-all market.
That changes once we start to move down the stack and get closer to the base layer.
(View Highlight)
Your various different online identities on the top of the stack get bundled in a handful of identity aggregators. Most online services don’t offer more than three or four different SSO options because – you guessed it – there is limited pixel real estate in the sign-up flow.
Chances are you don’t actively use more than two different email addresses (one for work, one for personal life). Your phone number is directly tied to a physical device which makes multihoming unlikely because of the aforementioned multihoming costs and physical space constraints. (View Highlight)
The most interesting identity defaults to occupy are therefore all the interfaces between your real world identity and your digital self. For example, your driver’s license (see Apple Wallet), your payment details (see Apple Pay), or your medical records (see Apple Health).
All of these touchpoints represent sources of truth – which brings us to the last but most important kind of network effect default: Intersubjective realities. (View Highlight)
If you have ever worked in a revenue organization, there’s a good chance that you have heard the following sentence before: “If it’s not in Salesforce, it doesn’t exist”
It’s like “pics or it didn’t happen” but for sales people. A Salesforce entry is a timestamped proof point that a customer interaction or sales deal has actually taken place. It makes Salesforce an important source of company data. A system of record.
Salesforce is of course not the only available data source within a company. There are dozens of internal BI dashboards, spreadsheets and third-party software tools that track customer and revenue data. All of them are accurate in their own unique way, but they all tell you something slightly different.
A company can have multiple sources of data, but it can’t have multiple sources of truth. There is no multihoming on the truth layer. You need a single source of truth. A default. (View Highlight)
Salesforce is not more accurate than any of the other company data sources. It is not objectively better. But it becomes intersubjectively better if enough people agree that it should be the single source of truth.
“If it’s not in Salesforce, it doesn’t exist” is a reinforcing flywheel. A social network effect. The more people believe in it, the truer it becomes.
This is Salesforce’s core defensibility. You can replace a CRM system, but can you replace “the truth”? (View Highlight)
Intersubjective realities like being the default source of truth are not just intra-company network effects.
For many decades, a popular intersubjective reality was that “nobody ever got fired for buying IBM”. Salesforce feels like the 2021 version of that belief – and they are not the only company that fits that description. (View Highlight)
It would be amazing if we lived in a world where companies with terrible UX did badly in the stock market, but that is clearly not the case. The fact that you can build a successful business without a truly great product is a real testament to how strong of a moat belief-based defaults really are.
Of course, intersubjective realities don’t just lead to suboptimal product defaults like Salesforce or Workday. These are just extreme edge cases to prove my point. Most of the default software products of today (think AWS, Stripe, Notion) became defaults because their products are just really good. But even great products need shared beliefs to become defaults. (View Highlight)
Beliefs are the world’s most powerful network effects. They don’t just explain the success of software products. The price of Bitcoin, democracies and dictatorships, capitalism, and the power of the church are all based on intersubjective realities. No moat is more durable than a shared belief.
Jesus turned water into wine – but Salesforce turns shit into gold.
See, the real base layer of the stack isn’t an interface. And it’s not our identity either. It’s our collective minds. (View Highlight)
I very deliberately downplayed the importance of certain network effects and took a few shortcuts to make the narrative work. But as I pointed out at the beginning, this post isn’t really about network effects. It’s about thinking in layers.
This essay doesn’t end with a definitive answer to the “Are network effects overrated?”-question, because there is none. Some are overrated. Some are underrated. Some are rated just right. It really depends on where in the stack you are. (View Highlight)
Indirect network effects are a little more complicated. They occur in two-sided (or multi-sided) products where the size of one user group affects how valuable the product is to another user group. Operating systems are an often cited example here: The more users an OS has, the more attractive it becomes for developers. More developers and apps, on the other hand, make the platform more valuable for users, thereby generating a positive feedback loop between the two sides.
The third category, data network effects, describes products which become better with more users via the data those users generate. Waze is a popular example of this category as are products that are powered by machine learning. More users lead to more data which lead to better recommendations or predictions and, thus, a better user experience. (View Highlight)
(View Highlight)
(View Highlight)
If you study the list closely, you’ll notice that these businesses all have something in common. Something that the before-mentioned industries don’t have. They are based on atoms, not just bits. (View Highlight)
create layers on top of their business (everyone wants to be a platform)
move down the stack to get closer to the base layer (to increase defensibility)
What is the base layer?
The base layer is the final interface between the stack and the end user – which is typically an operating system tied to a piece of hardware. This is why atom-based network effects are so powerful: They help companies gain control of the most powerful layer of the stack. (View Highlight)
The data network effect, however, is not the real moat here. It’s just a means to an end. The end goal is to become a default on the layer below. (View Highlight)
(View Highlight)
If you think about it, it’s kind of amazing that the largest operating system humanity has ever seen only exists to protect an advertising business two layers further up the stack.
Unluckily for Google, smartphone operating systems are not winner-take-all but winner-take-most markets. Not only does Apple have considerable market share in the smartphone market, its user base is significantly wealthier and thus more interesting to advertisers. (View Highlight)
The apps layer is interesting because it created indirect network effects between developers and users, which added defensibility to the OS layer. That defensibility in turn meant that Apple was able to capture a lot of the value that was created on top of its platform. 30% to be exact.
30% value capture is great, but you know what’s even better?
100% value capture.
Over the last couple of years, Apple has increasingly launched and monetized its own apps (and services) on top of iOS. It has moved up the stack and is now competing with the third-party developers on its own platform.
(View Highlight)
The beautiful thing about defaults is that they beat almost any competing product – even if that competitor has strong network effects or is technically superior.
This is why Apple Maps has higher market share than Google Maps, why Apple Music is able to catch up with Spotify, and why Google pays Apple $15bn a year to remain the default search engin (View Highlight)
Facebook Login is a single sign-on service (SSO). It allows people to sign up to other social platforms and services using their Facebook account. SSO is more convenient for users and thus promises greater sign-up conversion rates for 3rd-party platforms. The more people use Facebook Login, the more platforms will offer it, thereby increasing the total number of accounts that a single user has connected to their Facebook identity. (View Highlight)
(View Highlight)
(View Highlight)
“If it’s not in Salesforce, it doesn’t exist” is a reinforcing flywheel. A social network effect. The more people believe in it, the truer it becomes.
This is Salesforce’s core defensibility. You can replace a CRM system, but can you replace “the truth”? (View Highlight)
Jesus turned water into wine – but Salesforce turns shit into gold.
See, the real base layer of the stack isn’t an interface. And it’s not our identity either. It’s our collective minds. (View Highlight)