A dummy’s guide to Network Effects
This article was written with Emily Glassberg Sands.
In the tech industry today, network effects are more widespread and powerful than ever. Google search gets better with higher search volume, Facebook is more valuable the more friends you have on it, Apple’s ecosystem is strengthened when everyone you know uses Apple.
In this post, we lay the foundations — what network effects are, what they are not, and how they work. But first, three reasons to care:
- Network effects can be magical for profits, especially for businesses with low variable costs.
- Most internet companies have very low variable costs.
- If you’re at a tech company, chances are you can actively architect network effects through data.
A brief history of network effects
Twenty-five years ago, Nobel Prize-winning economist Gary Becker wrote a classic paper trying to solve a mystery: why don’t restaurants with lines out the door just raise prices? Traditional economic reasoning would say they could increase profits by raising prices until the line disappears (and the “market clears”). But this just doesn’t happen.
Becker explained restaurant lines with the novel economic idea of network externalities, also known as network effects: for some products, the demand of one consumer is increasing in the demand of others. Basically, popularity is self-reinforcing. So a restaurant benefits from deliberately underpricing and keeping that line out the door.
Before the internet, the power of network effects were limited. Why? Because most businesses that exhibited network effects did not have downward sloping average cost curves. Their variable costs did not diminish as the business grew larger.
Take the restaurant example: once a restaurant is built, the costs of adding an extra table are low, but only until the restaurant is at capacity. From there, adding another table requires opening a new location, which is expensive. What’s more, the second restaurant location might not benefit from the excess demand for the first. This principle is true for most brick-and-mortar businesses that require physical floor space to make a sale: because the costs of adding more floor space are large, the power of network externalities in brick-and-mortar businesses are generally capped.
The historical exceptions were natural monopolies which always had exceptionally low variable costs. The telephone is a classic example. First, there were network effects at play: you valued a landline phone the more others had them. Second, telephones were a natural monopoly: adding another telephone customer cost the provider next to nothing. Industries like these with network effects and low variable costs were among the most profitable businesses of the early- to mid- 20th century.
The internet altered the basic economics of business by dramatically reducing variable costs across a range of industries. In today’s tech startups, variable costs like hosting and streaming tend to be small potatoes relative to the more fixed costs of, say, office space and technical talent. And low variable costs mean the profit potential of viral growth through network externalities is huge.
Many of the biggest internet businesses are built on some kind of network effect. Google, Facebook, and Apple are only the tip of the iceberg. And just as the classic natural monopolies were the most valuable businesses to own before the rise of computing, low variable cost businesses with strong network effects are among the most lucrative today.
So what, exactly, are network effects?
In economics (and business), a network effect is the effect that one user of a good or service has on the value of that product to other people. Put simply, the value of the product is dependent on the number of others using it.
Let’s work through a very simple — we promise! — model to understand where network effects take hold. Company profit is quantity times price minus cost:
Profits = Quantity x (Price — Costs(Quantity)).
Quantity sold is the number of people who receive net positive utility from consuming:
Quantity = Sum(1*(Utility — Price > 0)).
Now, suppose the good has quality alpha determined by the firm. Let’s consider different models for utility:
- Utility is constant: Utility = alpha
- Utility is affected by the number of consumers: Utility = alpha(Quantity)
The latter admits the possibility of network effects.
What we generally refer to as network effects can actually more precisely be called positive network effects. That is dU/dQ > 0: the good or service is more valuable the more others use it. Positive network effects create demand-side economies of scale.
With negative network effects, in contrast, the good or service is less valuable the more others use it. Examples include the 101 Highway, or Comcast. Negative network effects are more commonly referred to simply as congestion.
And some products even have both positive and negative network effects at play. The on-demand economy is one salient example. With Uber, my wait time is longer — or, with surge pricing, my price is higher — the more others in my area are using the service; that is, I face congestion on the demand side directly. But I also benefit from ample demand in my area because that induces a lot of drivers in my area, a positive network effect. Similar forces are at play on Airbnb and in most on-demand economies.
So what are they not?
Day-to-day we hear people refer to general herd behavior and to supply-side economies of scale as network effects. This usage is confused at best and misleading at worst.
First, on herd behavior. Herd behavior, or crowd following, is the tendency of people to adopt what those around them have adopted. But herd behavior does not imply network effects. If a product is objectively good, you get herd behavior; and as users learn about the goodness of a product, you get herd behavior. In both cases, the herd behavior relies on the quality of the good. (Network effects do not rely directly on quality.)
Let’s consider a couple of examples. Suppose the product is inherently “good”, and consumers have similar tastes, i.e. they like to consume “good” things. Then we see consumers adopting what those around them have adopted because their utility functions are similar — but not because they care about what their peers are doing. If you ever walk downtown Palo Alto on a hot day, you’ll see a lot of people eating Fraiche; that’s probably not because of network effects, but because we can all agree that nothing beats froyo on a hot day.
In a different case, suppose the product has uncertain “goodness”. Consumers can learn from those around them whether or not the product is actually good via social learning — in which an acquaintance tells them about the quality of the product — , or via observational learning — in which they infer “goodness” from seeing others consume. Imagine you’re trying to choose a good movie. If you consider IMDB ratings, you’re leaning on social learning; if you consider box office sales, you’re leaning on observational learning. While learning effects are important drivers of behavior, they are not network effects: the IMDB rating and the #1 box office sales do not in and of themselves increase your utility of attending the movie; they merely signal to you that the movie is good.
Second, on supply-side economies of scale. These typically shift the cost curve down, so that marginal costs are lower at larger quantities sold (dMarginalCosts/dQuantity < 0). Many businesses benefit from some sort of supply-side economy of scale. Walmart’s leverage over its suppliers is an extreme example: the more Walmart sells, the better prices it is able to provide for its customers. But supply-side economies of scale are distinct from network effects, which directly impact the benefits of the product to consumers (dUtility/dQuantity > 0).
Final Thoughts: Creating Network Effects Through Data
Think about your product as a fire. If the fire spreads quickly on its own, you probably have demand-side network effects. If you’re forced to manually shovel the coal, you probably have supply-side network effects that aren’t (yet) automated. And if you’ve built a conveyer belt that feeds the fire, you have automated supply-side network effects. The fastest growing fires generally have all three.
We won’t dive into the mechanics for now, but many of the most powerful tech players are leveraging a combination of decision science and data product to actively architect network effects. Are you?
