All about network effects (A16z mini-moocs)

Credit to Esther Aarts

Anu Hariharan, a current partner of Y combinator, has a great presentation that demystifies the concept of network effects while was a partner at Andreessen Horowitz. Almost all startups love to brag about their ability to achieve network effects in some ways or another but very rarely does one actually achieve that. Today I am on a mission to take a deeper dive in understanding the network effects through Anu’s presentation.

Anu’s presentation has the following components and let’s go through each one by one:

  1. What are network effects
  2. Properties, terms, and laws of networks
  3. Case studies of companies with network effects
  4. Strategies for building network effects.
  5. What aren’t network effects?

What are network effects

Network effects in business is defined as a phenomenon that occurs when a product or a service becomes more valuable to its users as more people use it. It is also known as the demand-side economies of scale.

People care about network effects because it is a means to a higher end — building better products and businesses. Specifically, network work effects can:

  1. Create barriers to exit for existing users and barriers to entry for new companies (help create moats)
  2. Protect software companies from competitors’ eating away at their margins
  3. Create or tip winner-take-all markets. This point is specifically important. If you follow Peter Thiel and his theories, he comments profusely about how independent the two variables — revenue and profit — truly are. Revenue represents the value the company creates for its users but the ability to retain some of that value for the company is a whole other matter. The creation of a winner-take-all markets is a necessary ingredient for retaining large portions of the value the company creates for tis customers.

Properties, terms and laws of networks

We have defined network effects in the context of running a business. Networks are a set of nodes connected by links (Think a group of interconnected people or system of things). The way the nodes are connected, or the structure of a network, determines key properties of a network.

  1. Whether nodes are homogeneous or heterogeneous. (Homogenous nodes means that the network comprised of similar types of nodes. Skype is an example. Heterogeneous nodes comprised of different types of nodes. Open table is an example)
  2. Their type of clustering and degree of connection. (Degree measures number of connections to a single node. Clustering coefficient measures degree to which nodes cluster together — how likely are two nodes that are connected part of a highly connected group of nodes. Type of cluster can range from hub and spoke (star) to connected (clique).
  3. Directionality of those connections (unidirectional or bidirectional). Facebook is a bidirectional network while twitter is a unidirectional network.
  4. Whether they have complements. (Two products are complementary when they are separate but are more useful to users when the networks are together. This would result in increasing in usage of one product by a set of users reinforces and increases the value of a complementary but separate product. An example is the Microsoft Operating system network and the Microsoft Office Suite network)

Additional properties of a network but specific to communication networks:

  1. Sarnoff’s law: the value of network is proportional to the number of viewers. This is a broadcast network. For example, Yahoo’s value derives form this.
  2. Metcalfe’s law: the value of a network is proportional to square of number of connected users. Think Facebook
  3. Reed’s law: The value of a group-forming network is proportional to number and the ease with which groups form within it (subgroups grow faster than sheer number of P2P participants). Think Slack or WhatsApp groups.

Common Terms:

Marketplace: this is a network where money/transactions flow between two or more sides with distinct (heterogeneous) groups of users on each side; a successful marketplace is where supply and demand are attracted to the same place. (E.g. online auction marketplace, dating sites)

Platform: this is a network of users and developers; The multi-sided feedback loop between those users, developers, and the platform itself creates a flywheel effect increasing value for each group. (Think Salesforce Appexchange, Operating systems, Wechat.)

The discussion needs to start with these fundamental and detailed information of a network. These details can suggest the right questions to ask of a network. (E.g. is this network defensible?) and what the corresponding strategy should be. All of these questions boil down to what’s the initial growth lever or tactic to help get to scale?


For example, if the start up is a network, then the following questions should be asked:

  1. What should be the entry point be (to build a network effect)?
  2. What are the growth levers/tactics/hacks to get to critical mass?
  3. What’s the critical mass inflection point ( at which a network effect occurs)?
  4. How do you drive engagement?
  5. How do we take advantage of irregular topologies to find clusters and sub-clusters?


  1. How do we build liquidity in the marketplace/solve the chicken-and-egg (which side comes first) problem?
  2. Which is the money side vs subsidy side of the marketplace?


Will the market we are eyeing eventually be served by a single platform and will it be shared (ethernet) or will it be a fight for proprietary control?

Additionally, one has to understand whether the ‘network’ has a single player mode or a multiplayer mode. When the product has immediate utility for a single user, the user can start using the product (E.g flickr in the early days is a tool to store private photos while Foursquare in the early days allow you to book mark restaurants you have been to). When the product doesn’t have immediate utility, the user can’t be signed on by themselves. Instead the growth tactic should be different. (E.g. Skype in the early days need connections to other users to make calls. Slack is a messaging platform for teams).

Case studies of how networks grew


  1. Started off as a social network (P2p)
  2. Became a platform with developers
  3. Has elements of marketplace (users/advertisers, instant articles)

The key take-aways from Facebook’s growth:

  1. Facebook’s entry strategy was taking a clustered approach before rolling it out to other clusters
  2. They focused on engagement as well as growth
  3. Facebook kicked off offering immediate utility in single player mode (online school directory), but people started connecting with each other right away too


Airbnb is a two sided market place. Network effect comes from both sides of the network. A unique aspect of Airbnb is that its supply and demand overlaps at times.

It took Airbnb to build sufficient liquidity and to start seeing signs of network effects.

The key takeaways of Airbnb’s growth:

  1. Focused early features on building the demand side and in a marketplace, supply will always go to where the demand is (and will stay if you help growth the business)
  2. Traditional marketing methods — branding, design, targeting, direct advertising can help
  3. Trust and safety is paramount in all market places


Network effect from both sides of the network. More writers = more time readers spend on medium. More readers = more writers begin to write.

When readers invite other readers (via highlights, mentions, replies, and annotations), the overall value of the entire network increases as more ideas are shared in the network itself.

Key take-aways from medium:

  1. Reminder that single player mode can help get to multi-player mode. The appeal of the tool attracts users initially to help bulid enough critical mass, and then getting those users to participate over time creates the network come for the tool, stay for the network.
  2. They didn’t just build the tool and wait for the users to come. A lot of up-front work went int curating and editing early content and community.


Whatsapp users have around 20 connections compared to 980 in Facebook. Although on whatsapp the connections are few, they are highly clustered among close family and friends or whatsapp groups and therefore has a ton of engagement.

Key take-aways:

  1. Usage — not just growth — is what helps indicate network effects
  2. WhatsApp Launched globally at outset but still pursued a clustered approach by making sure product was working in one sub-cluster first, Product continued to grow in clusters, not just p2p
  3. No ads, no gimmicks, no games — focused on simplicity first which tends to viral before adding extra features
  4. Phone login is a very low barrier to entry for users

Strategies for building network effects

From the way the case studies presented above, we can see that the network effects are achieved through:

  1. Product should provide inherent value, whether in single or multiple player mode
  2. Growth tactics to drive adoption (Viral growth is helpful)
  3. Engagement triggers
  4. Sustain network effects

Key dimensions to consider in building network effects:

  1. What is your entry strategy? Bowling pin strategy. This is the strategy used to overcome the chicken-egg problem. The bowling pin strategy is how Geoffrey Moore has outlined in his Crossing the chasm book: start with niche segment where the chicken and egg can be both easily overcome and then eventually move to other niches and the broader market.
  2. What are the growth levers to drive adoption? Growth strategy.
  3. What is your critical mass inflection point? (Critical mass goals)
  4. What are the engagement triggers? (Engagement strategy). Having specific triggers to sustain engagement in network.
  5. How an you leverage an irregular network? The key task is to surface key clusters where companies can reach critical mass with those sub clusters and expand beyond.
  6. What do you attract the harder side of the marketplace? This is the subsidizing strategy. In almost every two-sided market, one side is harder to acquire than the other. Common way to attract the harder side is to subsidize that harder side. Reducing the costs for the hard side of the market can help build critical mass. However, unit economics should be carefully considered in order to build a sustainable business. This is why understanding which is the money side of the market place and the side of the market place where the most value is coming from matters so much because then you know which side to carefully subsidize.
  7. Show long term commitment to platform (Oculus and Microsoft both gave heavy subsidies to developers to build and enhance the platform)
  8. Provide stand alone value of the base
  9. Vertically integrate when supply uncertain. By vertically integrating the complement product as well as the base product, a company can attempt to ensure adequate supply of both goods. In platforms, one doesn’t necessarily have to be dependent only on outside developers — companies can ensure critical complements are built by themselves as well.

Some common Misconceptions about network effects

Network effects != Virality

Network effects increases value as more users join a network, whereas viral growth increases just the speed of adoption. The two happen at the same time but aren’t the same.

Product that spreads from one user to another through direct customer to customer contact. Viral growth implies low CAC (customer acquisition cost). Vriality is usually measured in viral coefficient K factor = [avg. number of invitations sent by each current user] * [conversion rate of invitation to the new user]

On the other hand, network effects are where products become more valuable when more users use it. Network effects help build a moat for the business, leading to high engagement/repeat rates and higher margins.

There are different flabors of viral growth:

  1. Network effects (Product Virality): A product that has inherent virality (spreads from one user to another as an organic consequence of use — will have network effect)
  2. Word-of-mouth (Traditional Virality): Customers organically recommend the product or distribute it via other platforms
  3. Referrals with no incentives (Traditional Virality): This spreads without financial or other sharing incentive due to being exclusive, invite only or other.
  4. Casual contact (Traditional Virality): This is where a product spreads virally via customer to customer contact (not via users intentionally inviting others)

Traditional virality doesn’t always lead network effects but the product virality leads to network effects.

A platform having scale doesn’t mean it has network effects.

Having scale or equivalently, having economies of scale, just means that product becomes cheaper to produce as business increases in size and output. The increasing scale leads to lower cost per unit of output.

Network effects means that product becomes more valuable as more user s use it. Network effects help build a moat — leading to high/repeat rates of engagement, higher margins.

Network effects is also called the demand side economies of scale while supply side economies of scale is a function of production size; so scale leads to lower cost per unit of output (unit economic efficiency).

Both effects are valuable and help build competitive moats but network effects tend to be stronger — users have higher barriers to exit.

A special note of data driven network effect is that a network effect that results from data. The driving factor behind is that when product powered by machine learning, becomes smarter as it gets more data from your users. The more the user uses the product, the more data they contribute; the smarter the product becomes.

Some thoughts I have on these material

Anu really dropped the bomb on the last point of data network effects. The idea is simple but the strategy is crucial to building a slew of new-age startups. Using AL/ML as a core vehicle to building moats. Having data doesn’t equate to having the ability to achieve data network effects. The key is that the product/company needs to have the right infrastructure/process that can take the user generated data and readily/efficiently convert them into insights in improving the product. Now the question comes to how to build a scalable infrastructure/pipeline that can make this conversion happen in real time or efficiently? I haven’t come across any topic that expounds on this topic. Once I found more materials related to this, I will share it here.

The original post and slides are found here: