A Strategist’s Guide to Platform Thinking
Part 3: Transaction Platforms
This is the third of a five-part series on platform thinking. The series lays out a simple framework designed to help you understand the various types of platforms and how they are used. Each part covers a different aspect of the platform spectrum and can be read independently in any order. In part 3, I will focus on transaction platforms.
Read this article if you are curious about the most powerful type of platforms that are shaping the world and want to understand the nuances of network effects.
What is a Transaction Platform?
Much discussion on the platform today is about transaction platforms, also known as multi-sided platforms in economics. The concept of transaction platform originated from the studies of credit card companies and ad-subsidized free newspapers. In recent years, research focus has shifted towards digital transaction platforms.
Although economists are still debating on the exact definition, they generally agree that the purpose of a transaction platform is to facilitate interactions or exchange of goods, services, or social currency between two or more participant groups.¹ Here are two oft-quoted definitions.
Van Alstyne, Parker, and Choudary (2016): “A [transaction] platform provides the infrastructure and rules for a marketplace that brings together producers and consumers.”²
Evans and Schmalensee (2017): “Multi-sided platforms reduce transactions costs and thereby facilitate value-creating interactions between two or more different types of economic agents.”³
Examples of Transaction Platforms
Transaction platforms have been an important part of the economic fabric of many industries for hundreds of years. Essentially, any market that puts in touch two or more groups of people who need each other can be considered as a transaction platform. Shopping malls connecting consumers and merchants and stock exchange matching buyers and sellers are just two examples of traditional transaction platforms.
What is new is that nowadays, we are surrounded by digital transaction platforms. The key drivers behind the increasing growth and pervasiveness of these platforms have been the rapid development of powerful information and communications technologies. As shown in Figure 1, the cost/performance of computing, digital storage, and bandwidth has been improving exponentially over the last decade. This has made it possible for entrepreneurs to build and scale up online transaction platforms rapidly and cheaply without owing physical infrastructure and assets. The proliferation of smartphone adoption combined with robust internet connectivity has enabled faster and more efficient connection between different sides of a digital transaction platform.
Digital transaction platforms are ubiquitous. Depending on the context and the function they fulfill, they can be broken down into different types. Table 1 provides some examples of two-sided digital transaction platforms.
What are the Defining Characteristics of a Transaction Platform?
In order to distinguish transaction platforms from other types of platforms and non-platforms, we need to understand the main features that make transaction platforms unique.
In general, transaction platforms have serval characteristics (Figure 2):
- Enabling interactions between two or more sides, where each “side” refers to a distinct group of participants;
- Reducing the transaction costs (e.g. search, contracting, and monitoring costs) incurred when participants on one side finding those on the other side; and
- Creating same-side and cross-side network effects by connecting different participants.
It is important not to confuse one-sided services with transaction platforms. For example, the file hosting and sharing service Dropbox has only one distinct side, i.e. end users interact with each other. In contrast, LinkedIn allows a recruiter to contact a potential employee and that interaction is two sided. Based on the characteristic of multisidedness, LinkedIn is a true transaction platform whereas Dropbox is not. Transaction platforms can bring together more than two sides. A good example is the travel platform Expedia which connects travelers with advertisers and service providers such as car rental companies, hotels, airlines, and cruises.
Transaction platforms create value by facilitating distinct groups of participants to get together. Without transaction platforms, participants on different sides would face much higher costs in finding and interacting with each other. By solving a coordination and transaction cost problem between them, a transaction platform becomes a catalyst that makes it easy for participants on different sides to interact, exchange information, and/or do business.
Understanding the Basics of Network Effects
The most valuable economic and business impact of transaction platforms comes from “network effects.” Also known as demand side economies of scale or network externalities, network effects refer to the degree to which every additional participant joining a network makes it more valuable to every other participant.⁷
The basic idea of network effects is simple. Consider the example of Facebook. The value of Facebook was zero if there was no other user to connect with. Facebook became more valuable if a user could reach more people. Figure 3 illustrates how adding another user to Facebook dramatically increases the number of other users that he or she can interact with. The potential value of Facebook to every other user, therefore, increases as more users are connected to Facebook.
Same-side vs. Cross-side Network Effects
What I just described is an example of same-side network effects, i.e. adding a new participant changes the value to all other participants on the same side. Same-side network effects are not limited to transaction platforms. For example, the more users join Skype, the more valuable Skype becomes to existing users on the same side. Skype exhibits same-side network effects, but it is not a transaction platform.
Cross-side network effects are often viewed as the holy grail for value creation in transaction platforms. In this case, increasing the number of participants on one side impacts the value to all other participants on the other side. Let’s continue with the Facebook example. The value of Facebook to an advertiser increases significantly as the number of users grows.
Positive vs. Negative Network Effects
Network effects are commonly associated with positive network effects. David Sack’s famous napkin sketch of Uber’s virtuous cycle has frequently been used to illustrate how network effects can contribute to a positive feedback loop.⁹ What is missing in the sketch is that network effects can also work in the opposite direction, resulting in a negative feedback loop. Feedback loops can enhance or dampen changes that occur in a system.¹⁰
Let’s look at each separately using Uber as an example.
In the case of positive network effects, more riders using Uber will attract more drivers because there are more business opportunities and hence less downtime for drivers. As more drivers joining the Uber network, riders also benefit due to better geographic coverage and faster pickup (Figure 4). A virtuous cycle emerges provided the roads are underutilized and all other factors remain fixed.
Conversely, negative network effects can arise when every additional Uber driver makes it less valuable to other drivers. If you live in any major city, you have most certainly experienced traffic jams. When roads are congested, more drivers will make it even more difficult for other drivers to navigate through, thus lowering the utility of existing drivers. As pickup time becomes slower, some riders may use alternative routes (e.g. subway) to get to the destinations. Fewer riders decrease the appeal to drivers because business opportunities are shrinking and there will be more downtime for the drivers (Figure 5).
The Uber example illustrates that it is possible for a transaction platform to exhibit both positive and negative network effects after reaching a certain threshold.
Direction and sidedness are two fundamental properties of network effects. By combining these two properties, Professor Amrit Tiwana of the University of Georgia developed a framework for analyzing network effects in transaction platforms (Figure 6).⁷ This framework is a useful tool for designing the architecture of the platform and the way in which it is governed.
One caveat to keep in mind: In order to harness the power of network effects and generate value, a transaction platform needs to introduce the right incentives and appropriate governance to get all sides on board first. Network effects may sound simple conceptually. In practice, there are many challenges that a transaction platform needs to overcome.¹¹
“Network effects result from getting the right customers, and not just more customers. Platforms create value when customers find good matches and enter into exchanges.” (Evan and Schmalensee)¹¹
How to Estimate the Value of a Network?
There is a strong association between scale and value in transaction platforms with network effects. As a network grows in size, the two key questions become: What is the value of the network? And in what proportion does the value increase relative to network size? The answers to these questions would drive different investment and M&A decisions.
Telecommunication pioneers and economists have developed several approaches to model how a network increases in value since Jeffery Rolhfs published his seminal paper on network effects in 1974.¹² Figure 7 outlines three valuation methods of a communication network. These methods are typically called “laws” in the literature to sound more scientific. But unlike physical laws, these methods are general rule-of-thumbs that attempt to assess the value of different types of communication networks.
Consider a fully connected network, where every participant can communicate or interact with every other participant. The total value of the network V is proportional to the number of participants n (e.g. users, subscribers, customers) multiplied by the connectivity value for each participant. The unknown proportionality coefficient A could be thought of as the value of a single connection for one participant and it is dependent on the type of the network.
Depending on the nature and the architecture of a communication network, the connectivity value for each participant can be estimated differently using different methods. It turns out that the assumptions used in each method have a lot to do with how the value is assigned.
Let’s examine each method in more detail.
Sarnoff’s Method
The first and most simple way to estimate the value of a communication network was named after David Sarnoff, a radio and television pioneer and the Chairman of Radio Corporation of America. His method states that the value of a broadcast network is directly proportional to the number of viewers.¹³
Sarnoff’s method is intended to estimate the value of a traditional broadcast network where content is transmitted by the hub and consumed by many subscribers (i.e. “one-to-many” structure). Modern digital networks often employ a “many-to-many” structure where users are connected to other users.
In a one-to-many network structure, if pricing remains fixed, the value of connectivity to each subscriber is the same regardless of the size of the network. A new subscriber would not automatically benefit from other existing subscribers already on the network. That’s the key assumption of Sarnoff’s method. Although his method is suitable for calculating the value of a broadcast network, if we use it on a social network, it will result in underestimation because network effects are not taken into account.
Metcalfe’s Method
Robert Metcalfe, co-inventor of Ethernet and founder of 3Com, has been widely credited for what is now called Metcalfe’s Law. Simply put, it says that the value of a communication network is proportional to the square of the number of its users.¹⁴ This means that if a network doubles in size, it does not double in value but quadruples.
What is rarely known is that Metcalfe introduced the hypothesis around 1980 on a transparency slide for a different purpose (Figure 8).
In his own words, Metcalfe explained:
“The original point of my law (a 35mm slide circa 1980, way before George Gilder named it, in 1995) was to establish the existence of a cost-value cross-over point — critical mass — before which networks don’t pay. The trick is to get past that point, to establish critical mass, which is why it has been so important for generations of Ethernet to be backward compatible.”¹⁶
It is important to recognize that Metcalfe originally conceived the idea as a way to sell more Ethernet cards to his 3Com customers.¹⁷ He used the slide to compare the benefits for the customers with the costs for the customers. He never claimed inexhaustible returns to scale in which the bigger the network, the better it is. His point was to establish the importance of establishing a critical mass so his customers can reap the benefits of network effects. However, the popular press continued to call Metcalfe’s model a “law” and use it to explain the success of the Ethernet and many Internet-based companies.¹⁴
Today, Metcalfe’s method has been widely used to estimate the value of any type of communication network. The foundation of his method is based on the assumptions that each member of a fully connected network can make (n-1) connections with other participants and those connections are equally valuable. Under these conditions, the value of a communication network is proportional to n(n-1) which is roughly n².
Briscoe, Odlyzko, and Tilly’s Method
Some scholars long questioned the assumptions of Metcalfe’s method, particularly the assignment of equal value to all connections.¹⁵’¹⁶ In large networks, many connections are not used at all. Let’s use the mobile network as an example. As of November 2019, 5.166 billion people around the world are subscribed to mobile services.¹⁸ Metcalfe’s method assumes that each mobile subscriber gains equal value from each of those 5.166 billion. But are we really going to call everyone who is connected to the mobile network at the same frequency? Even if we can reach 1,000 people a day, it will take us well beyond our lifetime — 141,534 years — just to contact all mobile subscribers only once.
Bob Briscoe, Andrew Odlyzko, and Benjamin Tilly proposed that the value of a communication network is proportional to nln(n).¹⁹ Their method was based on the principle of diminishing marginal returns through a rule-of-thumb known as Zipf’s Law. Arose out of an analysis of language, linguist George Kingsley Zipf observed that if a large sample of words is ordered by popularity, the most frequent word will occur approximately twice as often as the second most used word, which occurs twice as often as the fourth most frequent word. Stated more generally, the frequency of any word is inversely proportional to its rank in the frequency table.²⁰
This relationship between rank and frequency has been demonstrated to occur naturally in an extraordinarily diverse range of phenomena. For instance, the number of hits on websites and the number of calls received on a single day have all been shown to follow Zipf’s observations.²¹
Let’s apply the rank/frequency relationship in a mobile communication network. Following Zipf, the nth ranked member contributes to the total value about 1/n of the value of the first member. For instance, if you talk to your favorite person 100 minutes a month, then your second favorite person will only get half of the time, i.e. 50 minutes. From your perspective, the total value to you will be the sum of the decreasing 1/n value of all the other members of the network. This value will be proportional to 1 + 1/2 + 1/3 + … + 1/(n-1), which approaches ln(n).
Different Estimates of Network Value
“All models are wrong, but some are useful.” (George Box)
Depending on which method is used, the calculations of network value can lead to drastically different results. Figure 9 shows the value of a communication network estimated by three different methods as the network grows from 1 to 20 users. The bottom curve shows linear growth estimated by the Sarnoff’s method and the top curve represents quadratic growth estimated by the Metcalfe’s method. The middle curve, estimated by the Briscoe, Odlyzko, and Tilly’s method, grows faster than linear but slower than quadratic.
An accurate valuation of a communication network is complicated. None of the methods can be applied universally. The shape of the network value curve is not only influenced by the nature and the structure of the network, but also by the assumptions we impose. It is worth noticing that some of the assumptions are more accurate than others. Unrealistic assumptions can lead to an overly optimistic estimation of network value. This can, in turn, drive different investment decisions.
An Alternative Model
Let’s not forget that early studies of network effects focused on the connection of physical devices such as telephony and fax machines. Today’s communication networks connect people. Because we are human beings, we either don’t want to talk to everyone on the network, or don’t have the time or energy to talk to everyone. Joe Weiman described this intrinsic limits of consumption of value as follows:
“Regardless of how much value might be “in” the network, bottlenecks in discovering, accessing and acquiring, or consuming that value may prevent all of the value from being realized by each node or user. Time is one limit, but so is capacity, e.g. storage capacity of a user’s portable music player or personal video recorder.” (Joe Weiman)²²
The intrinsic limits of consumption suggest that each user has an upper bound in value derived from the network regardless of how large the network gets. The total value derived from the network is therefore bounded. Briscoe, Odlyzko, and Tilly used a divergent series to model the value of a communication network. Even though their approach is more accurate than others, it can still lead to overestimation because the distribution of value to a user remains unbounded. You can talk to 100 people on a monthly basis. It is possible to stretch the number to 1,000 people. But after that, you probably no longer want to talk to anyone “lower down” on the list. It is extremely unlikely that you will ever call the 10,000th ranked person for 0.6 seconds.²²
In the real world, we only talk to the same, small group of people again and again. Empirical studies by Paul Adams show that the average Facebook user communicates directly with just 4 people per week and 6 people per month despite these people are checking Facebook every day. This includes private messages, chats, wall posts, and likes and comments on status updates. According to Adams, in general, people have consistent communication with between 7 and 15 people, but that most conversations are with our five strongest ties who we are emotionally close to and communicate frequently with.²³
Figure 10 depicts how frequently we communicate with others in a social network. Human relationships are shown as layers of nested circles from closest to the furthest. Numerous studies have confirmed that most people have about 5 intimate friends (the closet circle), 15 close friends, 50 general friends, 150 meaningful contacts, and 500 acquaintances.²³
Robin Dunbar, an evolutionary psychologist and anthropologist, demonstrated this pattern in the 1990s and hypothesized that there are cognitive constraints of maintaining relationships in primates as well as in humans. Limitations on brain capacity and the time required in a relationship to maintain an adequate level of emotional intensity mean that we can only maintain about 150 meaningful relationships at any time. The cognitive limit to the number of individuals with whom we can maintain meaningful relationships is known as Dunbar’s number.²⁴
Recently, Robin Dunbar analyzed thousands of Facebook users. He found that despite social media’s potential to connect us with many people on the network, we only have the capacity to maintain 150 friendships at a time.²⁵ In an interview with ABC, he commented that:
“People can (and sometimes do) have 500 or even 1,000 friends on Facebook, but all they are doing is including people who we would normally call acquaintances.” (Robin Dunbar)²⁶
One implication of Dunbar’s research is that users do not value the number of potential connections as much as they value close connections in social networks both online and offline. If we agree that there is an intrinsic limit of consumption of value, then how do we factor this in when estimating the value of a social network? If we use Briscoe, Odlyzko, and Tilly’s method, the long tails of the divergent series will inevitably contribute to an overestimation of value.
Here, I propose a new model based on a truncated Zipf’s series. The key assumption of my model is the existence of intrinsic limits of consumption of value. I use nₘₐₓ to denote the effective number of connections beyond which any further increase in the size of the network will not contribute to any noticeable benefits to the users. The total value of a network is proportional to nlnnₘₐₓ.
We can apply this model to estimate the value of a social network. In the case of Facebook, nₘₐₓ can be approximated by Dubar’s number which is 150. Figure 11 compares the estimation of the Facebook network value using Truncated Zipf’s series with Briscoe, Odlyzko, and Tilly’s method. As you can see, even for a small number of connections, the value calculations can be very different depending on which method you choose.
Where is Real-world Evidence?
Surprisingly, despite the importance of network effects in transaction platforms, no empirical evidence was available to validate any of the valuation methods described above until late 2013. As of today, I can only find two original papers attempting to prove Metcalfe’s Law. One paper was authored by Metcalfe himself where he used Facebook’s actual data over the past 10 years to show a good fit for his model.¹⁴ Specifically, Facebook’s revenue is proportional to the square of the number of its monthly active users. A more recent paper published by Chinese researchers confirmed Metcalfe’s results with more data set from Facebook and Tencent.
Contrary to Metcalfe’s conjecture that costs increase linearly, the Chinese authors also found that the costs of Facebook and Tencent are proportional to the squares of their network sizes.²⁷ This finding may help explain why Uber is still losing $3 billion and has a total debt of $6.5 billion.²⁸ Metcalfe’s assumption of linear increase in costs may be plausible if the costs of adding each additional Ethernet card would be precisely the same. Whether this assumption holds true in transaction platforms today is still questionable. What we know from Uber is that there are many other sources of costs. For example, Uber has been spending huge amounts of money to acquire drivers, incentivize users, expand geographically, and overcome regulatory hurdles.²⁹
Results from the two papers should be interpreted with caution. Both Metcalfe and the Chinese authors used non-inflation adjusted revenue as a proxy for value over a long period of time. Growth in revenue could be attributed to M&A and expansion of services rather than network effects per se. Moreover, Facebook and its close cousin, Tencent, which is the holding company of the largest social network WeChat in China, have tremendous pricing power over advertisers. Indeed, research conducted by AdStage revealed that the average CPM on Facebook ads increased 171% during the first six months of 2017, from $4.12 to $11.17, while impressions remained flat. Are the authors simply modeling the effects of the price increase?
The two papers have raised more questions than answers. What proxy should we use to measure the true value of the network? What assumptions should we make to better reflect human behaviors in real life? How fast does the value increase relative to network size? Can the impact of negative network effects also be modeled?
More research is needed so we can use facts to develop effective business strategies and make sound investment decisions.
References:
1. Wismer, S., & Rasek, A. (2018). Market Definition in Multi-sided Markets. In OECD, Rethinking Antitrust Tools for Multi-Sided Platforms 2018 (pp. 55–67).
2. Van Alstyne, M.W., Parker, G.G., Choudary, S.P. (2016). Pipelines, Platforms, and the New Rules of Strategy. Harvard Business Review, April.
3. Evans, D.S., & Schmalensee, R. (2017). Multi-sided Platforms. In Palgrave Macmillan (Eds.) The New Palgrave Dictionary of Economics. Palgrave Macmillan.
4. Deloitte Center for the Edge. (2016). 2016 Shift Index. Deloitte University Press.
5. Hagel, J. (2015). The Power of Platforms. Deloitte University Press.
6. Hagiu, A., & Wright, J. (2015). Multi-Sided Platforms. Harvard Business School Working Paper 15–037.
7. Tiwana, A. (2014). Platform Ecosystems: Aligning Architecture, Governance, and Strategy. Morgan Kauffmann.
8. Yoo, C.S. (2016). When Are Two Networks Better than One? Toward a Theory of Optimal Fragmentation. Global Commission of Internet Governance Paper Series №37.
9. Gurley, B. (2014, July 11). How to Miss By a Mile: An Alternative Look at Uber’s Potential Market Size [Blog post]. Retrieved from: https://abovethecrowd.com/2014/07/11/how-to-miss-by-a-mile-an-alternative-look-at-ubers-potential-market-size/
10. Farnam Street. (2011, October). Mental Model: Feedback Loops [Blog post]. Retrieved from: https://fs.blog/2011/10/mental-model-feedback-loops/
11. Evans, D.S., & Schmalensee, R. (2017). Debunking the ‘Network Effects’ Bogeyman. Regulation, 40(4), Winter 2017–2018.
12. Rohlfs, J. (1975). A Theory of Interdependent Demand for a Telecommunications Service. Bell Journal of Economics, 5(1), 16–37.
13. Farris, P., Pfeifer, P.E., Johnson, R.R. (2009). The Value of Networks. Darden Case No. UVA-M-0645.
14. Metcalfe, B. (2013). Metcalfe’s Law after 40 Years of Ethernet. Computer, 46(12): 26–31.
15. Briscoe, B., Odlyzko, A., & Tilly, B. (2006). Metcalfe’s Law: A misleading driver of the Internet bubble. University of Minnesota Working Paper Version February 11, 2016.
16. Briscoe, B., Odlyzko, A., & Tilly, B. (2006). Metcalfe’s Law is Wrong. IEEE Spectrum.
17. Metcalfe, R.M. (2007, April 20). It’s All In Your Head. Forbes. Retrieved from: https://www.forbes.com/forbes/2007/0507/052.html#526a4c0f47d3
18. GSMA Intelligence. [2019, November 30]. Definitive Data and Analysis for the Mobile Industry. Retrieved from: https://www.gsmaintelligence.com/
19. Odlyzko, A., & Tilly, B. (2005). A refutation of Metcalfe’s Law and a better estimate for the value of networks and network interconnections. University of Minnesota Working Paper Version March 2, 2005.
20. Zipf, G.K. (1949). Human Behavior and the Principle of Least Effort: An Introduction to Human Ecology. Reprinted by Martino Fine Books in 2012.
21. Newman M.E.J. (2005). Power Laws, Pareto Distributions and Zipf’s Law. Contemporary Physics, 46, 323–351.
22. Weinman. J. (2007). Is Metcalfe’s Law Way Too Optimistic? Business Communications Review, August, 18–27.
23. Adams. P. (2012). Grouped: How Small Groups of Friends are the Key to Influence on the Social Web. New Riders.
24. Ro, C. (2019, October 9). Dunbar’s number: Why we can only maintain 150 relationships. BBC Future. Retrieved from: https://www.bbc.com/future/article/20191001-dunbars-number-why-we-can-only-maintain-150-relationships
25. Dunbar, R.I.M. (2015). Do online social media cut through the constraints that limit the size of offline social networks? Royal Society of Open Science, 3, 150292.
26. ABC Science. (2016, January 20). Facebook: 150 is the limit of real friends on social media. ABC News. 20 January 2016. Retrieved from: https://www.abc.net.au/news/science/2016-01-20/150-is-the-limit-of-real-facebook-friends/7101588
27. Zhang, X.Z., Liu, J.J., & Xu, Z.W. (2015). Tencent and Facebook Data Validate Metcalfe’s Law. Journal of Computer Science and Technology, 30(2), 246–251.
28. McBride, S. (2019, August 14), Uber Gives Investors the Worst of Both Worlds. New Constructs. Retrieved from: https://www.newconstructs.com/uber-gives-investors-the-worst-of-both-worlds/
29. CB Insights Research Report. How Uber Makes — And Loses — Money. Retrieved on 2019, November 30 from: https://www.cbinsights.com/research/report/how-uber-makes-money/#6
30. Prater, J.D. (2017, September 18). Facebook CPMs Increase 171% In 2017. AdStage. Retrieved from: https://blog.adstage.io/2017/09/18/facebook-cpms-increase-2017
