A Framework for Quantifying Network Effects on Platforms

Alex Sambvani
7 min readMar 12, 2018

Not all network effects are created equal. The value of a network to a given user depends on several discernable factors. This post attempts to identify the most important of those factors and proposes a qualitative framework for measuring the strength or value of platform network effects. Hopefully this framework is useful to those interested in understanding the strategic positioning of a network-based business — investors, advisors, employees, and passive observers alike.

First, let’s define what “network effects” are — this term is used liberally throughout the tech industry and is often loosely defined. Here, we’ll define a network effect as demand side economies of scale, meaning that the value of the network to a user increases as the size of the network increases (i.e. the number of users on the network grows). To make this definition more tangible, imagine you just downloaded a brand new messaging app, like WhatsApp, and you are the FIRST user on it. This app wouldn’t be very valuable to you, because there would be no one for you to send a message to! Now imagine your best friend downloads the app. The app is instantly more valuable to you because there is another person on the platform you can interact with. Now imagine 100 more of your friends download the app. The size of the network has just grown ~50x and the app is exponentially more valuable to you — that’s because of network effects!

So everyone knows that network effects are great and that many of today’s leading tech companies have them, but how do we quantify the network effects that a company has? Every company can’t have the same type of network effects, right? The answer is: Definitely not.

Metcalfe’s Law

I must confess here that I’m not the first person to think about quantifying the strength network effects. Perhaps the leading methodology for valuing a network is Metcalfe’s Law, which states that the strength of a network with n users is proportional to (n²). So if a network is worth $X to a user for each user on the platform, 10 users on the platform would make the network worth ~$100X to a user (Farrell & Saloner, 1985; Katz & Shapiro, 1985; Economides, 1996). The graph below demonstrates how, according to Metcalfe’s law, the value of a network grows exponentially as the number of users in the network grows.

Cool, so all you have to do is square the number of users in the network and you’re done, right? WRONG!

Metcalfe’s law actually estimates the potential value of a network; Metcalfe’s law assumes that each user can potentially interact with every other user in the network. The issue with applying Metcalfe’s law to many networked businesses is that a lot of them have constraints that prevent users from interacting with the entirety of the network. In fact, some businesses are fundamentally incapable of reaching their maximum network potential given their product offering / network structure (more on this later). It is for this reason that we might consider other frameworks for quantifying the strength of network effects or the value of a network.

Constraints on Network Effects

There are many factors that can constrain or dampen network effects. My thinking is that there are two broad categories in which these most of these constraints fall into: Openness and Localness.

What I mean by Openness is the ease with which and the likelihood of which a user can discover and interact with another user on the platform. The most open type of network does not require users to have a pre-existing relationship in order to interact — these types of networks often are anonymous (think Craigslist or Ebay). The most closed type of network would be one where users have to have a pre-existing relationship in order to participate, and they likely will only interact with users that they know personally. An example of a closed network would be something like an invite-only messaging app. The level of discoverability matters, too. For example, if users can search for other users to interact with, especially users they don’t know, then the app would be more open than if you couldn’t discover users organically. One drawback of the original Snapchat app was that it was low on discoverability, because you had to know a user’s screen name or have them as a contact in your phone in order to follow them (there was no way to search and discover by name, etc…) — this is something that has been addressed in the newly-released version of the app. Now, you can more easily find users to follow through search. In the “Stories” portion of the app, you can now explore popular stories, regardless if you follow the user that posted the content (similar to Instagram’s “Explore” tab). Higher levels of Openness will lead to stronger network effects.

I define “Localness” as the extent to which the location of a given user matters. The less location matters, the higher the potential for interaction and thus the stronger the network effect. For example Ebay doesn’t require you to live in the same city as the person you purchase an item from — you can in theory buy from any seller in the world. This doesn’t work for a network like Uber’s or Lyft’s. Rideshare companies have local network effects, because users on both sides of the platform must be in the same location in order to transact. Services with local network effects must work hard to grow their networks on a location-by-location basis. Being the dominant player in, say, New York, doesn’t help you dominate another city such as Los Angeles (other than capital you might be able to raise by already being successful somewhere). To illustrate this point even more simply, let’s think back to the definition of a network effect. The value of the network is supposed to increase as the size of the network grows. Let’s pretend you live in Los Angeles and you are both an eBay and an Uber customer. Now let’s pretend your friend that lives in San Francisco just joined both services today (assume they join Uber as a driver). When your friend joins eBay, the platform is more valuable to you because there is one more person on the platform that you can potentially buy from or sell to. However, the value of Uber to you is unchanged, because you can’t catch a ride from your friend in San Francisco when you use Uber in LA! Metcalfe’s law doesn’t fully apply to an app like Uber, because the potential number of interactions is not based off of all active users in the app, it’s based off of all active users in your area.

Figure 1 below demonstrates how some of the leading networked-based businesses out there would be classified under this framework. You’ll notice that there aren’t any companies in the bottom left; my hypothesis is that networked business that are constrained by both location of users and by pre-existing relationships of those users do not last. Let’s try to hypothetically bring one of these businesses into the bottom left. Imagine if you could only add a friend on Facebook if they lived in your city. This would make Facebook constrained by both location and by each users’ personal friend network. If this were the case, do you think Facebook would still be the social media juggernaught that it is today? Possibly, but probably not.

Figure 1

When trying to evaluate Openness, consider:

  • Discoverability — High levels of discoverability will correlate to stronger network effects
  • Familiarity required— Do users need to have a pre-existing relationship to interact? Network effects will be stronger if no pre-existing relationship is required
  • Language — What types of languages are spoken on the platform? Do users need to speak the same language to interact? The less language matters, the stronger the network effects will be. If language does matter, then network effects should be estimated by considering each group of same-language speaking users independently.
  • Idiosyncratic constraints — The nature of the transactions being made on the platform might introduce some other constraints that limit the number of interactions. Think about what platform-specific attributes might affect the potential for users to interact with each other.

When trying to evaluate Localness, consider how much location matters to the interaction. Can the interaction take place with users on each side of the platform being on opposite sides of the globe? The less location matters, the stronger the network effects will be.

REMINDER! It’s important to remember that network effects aren’t all that matters when determining the viability of a technology business. Product quality also matters as do many other factors. There are many successful technology companies that make great products and compete solely on product quality (i.e. do not have network effects).



Alex Sambvani

Co-founder and CEO @ Slang.ai. On a mission to improve phone-based customer service.