This post is part of a series on business model analysis for entrepreneurs. The first post in the series presents a comprehensive list of issues (available as a downloadable PDF) entrepreneurs should consider when designing a business model. Other posts delve deeper into specific issues; this one provides an overview of platforms and network effects.
Network effects are evident when any given customer’s willingness to pay (WTP) for a product depends on the number of other customers with whom they can interact by using the product. In the classic example, the first fax machine sold was worthless until someone purchased a second machine. By providing another potential destination for messages, the subsequent arrival of each new fax machine increased the value of every existing machine and likewise increased the WTP of prospects who had not yet acquired a fax machine. (See the technical appendix below for more on how network effects impact customers’ willingness to pay.)
Network effects arise in platform-mediated networks (PMNs), which include networks of customers—often called users—who wish to interact with each other, along with one or more intermediaries who provide a platform, encompassing infrastructure and rules to facilitate users’ interactions. PMNs form the heart of the computer, telecommunications, media, and Internet sectors. However, PMNs are not limited to information industries. They can also be found in financial services (e.g., stock exchanges, credit cards, ATMs), health care (e.g., HMOs), energy (e.g., the power grid), transportation (e.g., airlines, container shipping, gasoline stations), and retailing (e.g., shopping malls, barcodes). A diverse array of matchmaking businesses mediate network transactions, including auctioneers, executive recruiters, realtors, and travel agencies. Ranked by market value, sixty of the world’s 100 largest companies earn >50% of their revenue from platform-mediated networks, including American Express, Cisco, Citigroup, Time Warner, UPS, and Vodafone. In short, PMNs are essential features of modern economies.
Many startups aspire to either build new platforms (e.g., Facebook) or to profit by offering complements that leverage platforms (e.g., Zynga, which relies on Facebook to host its social games). Platform business models can be extraordinarily complex, and design mistakes are common. Consider Sony’s defeat in the Betamax/VHS standards battle. Or the billion dollars that IBM invested in the failed OS/2 operating system. Or the hundreds of millions of dollars squandered when Yahoo! and Amazon launched full frontal assaults on eBay’s dominant U.S. auction franchise. Or eBay’s own failure to capture auction markets in Japan and China when it imported its U.S. pricing model into those markets, where its rivals initially offered free transactions.
It’s especially difficult to get pricing right in platform-mediated networks, so I’ll focus on that topic here. Specifically, I’ll explain why a platform must be proprietary—controlled by a single company—for its provider to capture value created through network growth. I’ll also give a brief overview of rules for setting prices in a “two-sided” network—one with two distinct user groups.
But getting your pricing right is just the tip of the iceberg when designing business models that leverage network effects. It would take a series of long posts to cover all of the relevant topics, so I’ll just mention a few of them below and point to some articles and notes that should be helpful to an entrepreneur grappling with these topics.
(Links marked below with an asterisk are available as PDFs for $7 apiece from HBS Publishing; others are available as free downloads. Sorry for not being able to provide all of my work for free, but we fund our research at HBS through the sales of cases and notes.)
Proprietary Platforms and Pricing Leverage
When a platform-mediated network’s user base expands, the resulting increase in users’ WTP for platform access does not automatically translate into higher prices for the platform. Providers only gain pricing leverage with network growth when they keep their platforms proprietary. A proprietary platform has a single provider who exclusively controls its technology, for example, eBay, Federal Express, or Google. With a shared platform such as barcodes, DVD, or Wi-Fi, multiple firms collaborate in developing the platform’s technology, and then compete with each other to provide differentiated but compatible versions of the platform. Compatibility ensures that users can switch between products offered by a shared platform’s various providers without incurring significant switching costs. For example, when switching from one Windows-compatible PC to another—say, from Dell to Compaq—a user need not invest in new application software. By contrast, switching to a rival platform—for example, from Windows to Macintosh—would be much more costly.
For a shared platform’s providers, lower customer switching costs limit pricing leverage. By contrast, a proprietary platform’s sole provider can raise its price when user base growth increases users’ WTP, up to the point where a user is indifferent between sticking with the platform and switching to a rival platform, if one exists. With such pricing leverage, startups can capture more of the value they create through network effects. Consequently, entrepreneurs will normally prefer proprietary platforms when designing business models. However, as I explain in this article*, there are circumstances when it may be advantageous to share platforms with rivals. In particular, when many equally matched rivals simultaneously enter a new market that is destined to be dominated by a single platform due to very strong network effects, the firms may be better off sharing the market than conducting an expensive winner-take-all battle*. Even though they offer competing versions of a compatible platform, firms may still be able to profit by contributing their technology to the shared platform, perhaps earning licensing fees or gaining a time-to-market edge.
Sharing a platform implies an open architecture. As I’ve written elsewhere on this blog, the concept of an “open platform” can lead to confusion, because platform-mediated networks encompass several ecosystem layers—each of which can be open or closed. In this paper, Geoff Parker, Marshall Van Alstyne and I describe circumstances under which platforms may find it attractive to open or close as they mature. In a more technical paper, we describe a strategy we call platform envelopment, which entails one platform owner moving into another’s market by offering a multi-platform bundle. Through envelopment, firms can surmount the high entry barriers that often protect incumbents who harness strong network effects. Examples include moves by Microsoft (with Explorer, Windows Media Player, Silverlight, etc.), Apple (with the iPhone/iPad, which has enveloped MP3 players, feature phones, PDAs, handheld gaming devices, standalone GPS units, eBook readers, etc.), and Google (with Checkout, Docs, Chrome, Android, etc.).
Platform-mediated networks can be categorized according to the number of distinct user groups they include. In some PMNs, users are fairly homogenous in terms of the platform functionality they require. For example, although a given stock trade has a buyer and seller, these roles are transient; almost all traders play both roles at different times. PMNs with homogenous users are called one-sided to distinguish them from two-sided networks, which have two distinct user groups whose respective members consistently play a single role in transactions (e.g., cardholders and merchants in credit card networks).
Platform providers who seek to mobilize new two-sided networks face a Catch-22: each side will refuse to join until the other side is on board. To avoid this impasse, platform providers often subsidize users on one side, that is, they price below marginal cost to that side or even give its users free access to the platform, as with Adobe’s PDF reader software or Google’s search engine. Due to network effects, attracting users to this “subsidy side” boosts users’ WTP on the network’s “money side,” as with customers for Adobe’s PDF creation software or Google’s advertisers. Generally, it makes sense for platform providers to permanently subsidize the network’s more price sensitive side and charge the side that increases its demand more strongly in response to the other side’s growth.
Parker, Van Alsytne and I discuss pricing rules for two-sided networks in this article.* In a note* for MBAs, Andrei Hagiu and I examine staging strategies that avoid the Catch-22 in two-sided networks by first bringing on board one side in large numbers before seeking to mobilized the other side. Amazon used such a staging strategy. Amazon attracted a big base of customers using a traditional merchant business model, then later morphed into a two-sided platform by hosting third-party merchants who leveraged Amazon’s infrastructure and customer base.
This appendix, adapted from my note* "Racing to Acquire Customers," will only be of interest to readers who like to lift up the hood to understand how a motor works. Below, I dig into the economics of network effects and, using a stylized example, show how they can impact customer willingness to pay, product pricing, and the pace of investment in customer acquisition.
In a market that exhibits network effects, a company must consider the present and future benefits of building its customer base when making current-period pricing and investment decisions. Absent network effects, a company typically maximizes current-period profitability by setting its price and marketing spending at levels that equate current-period marginal revenue and marginal cost. With network effects, however, optimization is less straightforward. Acquiring an additional customer—say, Mary—in the current period yields revenue directly from Mary during that period and also boosts—very slightly—the willingness to pay of all other customers, who value a larger network. Other customers’ willingness to pay remains higher by this tiny increment in future periods, until Mary exits the network. This increase in other customers’ willingness to pay in current and future periods should be factored into decisions about whether to invest in accelerated growth.
These dynamics can be illustrated using an example of a fictional company—let’s call it “Blossom”—selling spreadsheet software in 1989 (before most office productivity software was sold in bundled suites). Then as now, spreadsheet software was subject to a fairly strong network effect: users valued the ability to exchange files and required software that employed a compatible standard to do so. I use this example because scholars have estimated the impact of network effects on customer willingness to pay for spreadsheet software in 1989 using an econometric technique called hedonic regression (Brynjolfsson & Camerer, 1996, “Network Externalities in Microcomputer Software,” Managment Science 42: 1627-1647) Those estimates are employed in the stylized example below to show how network effects can influence optimal marketing spending levels.
Assume that Blossom’s unit price was $360 and its variable expenses (excluding customer acquisition costs) for supplying an additional unit were $72, or 20% of revenue. To keep the analysis simple, we also assume that a new customer would use the product for four years, at which point the product would be rendered technologically obsolete and the customer would be in no way loyal to Blossom. This assumption is unrealistic, but it lets us avoid factoring the present value of future product sales into our calculations. In reality, customers might prefer to buy Blossom’s next-generation product if it provided backward file compatibility or reduced the time spent mastering a new interface.
First, consider a scenario in which spreadsheet software was not subject to a network effect. What is the largest amount Blossom should be willing to spend to acquire an additional customer in 1989? This calculation is straightforward: the company could afford to spend up to $288 ($360 minus pre-marketing variable expenses of $72) to acquire a customer. Beyond $288, the marginal cost of acquiring and supplying a new customer would exceed the marginal revenue from that customer.
How would network effects influence the maximum amount that Blossom should be willing to spend to acquire a customer? The research mentioned above provides the following formula for predicting the logarithm of a spreadsheet’s product price in 1989 (“P”), based on its publisher’s installed base share (“S”), after controlling for differences in product quality. Consistent with the concept of network effects, the formula indicates that customers would pay more for a spreadsheet used by a bigger installed base:
Log P = 5.7376 + .0075S
If we assume that Blossom had a 20.000000% share of the total installed base of spreadsheet software, then this formula would predict customer willingness to pay equal to $360.5389517 (the reason for showing so many decimal places should be apparent in a moment). Assume further that:
Obviously, a company cannot raise its price in increments as small as $0.0000054, but larger market share gains may yield material pricing benefits. Specifically, based on the formula above, a 1% share increase yields a 0.75% increase in customer willingness to pay for spreadsheet software in 1989. For this reason, network effects provide an incentive for companies to invest more aggressively in accelerated growth strategies—that is, to race for scale.
Network effects are evident when any given customer’s willingness to pay (WTP) for a product depends on the number of other customers with whom they can interact by using the product. In the classic example, the first fax machine sold was worthless until someone purchased a second machine. By providing another potential destination for messages, the subsequent arrival of each new fax machine increased the value of every existing machine and likewise increased the WTP of prospects who had not yet acquired a fax machine. (See the technical appendix below for more on how network effects impact customers’ willingness to pay.)
Network effects arise in platform-mediated networks (PMNs), which include networks of customers—often called users—who wish to interact with each other, along with one or more intermediaries who provide a platform, encompassing infrastructure and rules to facilitate users’ interactions. PMNs form the heart of the computer, telecommunications, media, and Internet sectors. However, PMNs are not limited to information industries. They can also be found in financial services (e.g., stock exchanges, credit cards, ATMs), health care (e.g., HMOs), energy (e.g., the power grid), transportation (e.g., airlines, container shipping, gasoline stations), and retailing (e.g., shopping malls, barcodes). A diverse array of matchmaking businesses mediate network transactions, including auctioneers, executive recruiters, realtors, and travel agencies. Ranked by market value, sixty of the world’s 100 largest companies earn >50% of their revenue from platform-mediated networks, including American Express, Cisco, Citigroup, Time Warner, UPS, and Vodafone. In short, PMNs are essential features of modern economies.
Many startups aspire to either build new platforms (e.g., Facebook) or to profit by offering complements that leverage platforms (e.g., Zynga, which relies on Facebook to host its social games). Platform business models can be extraordinarily complex, and design mistakes are common. Consider Sony’s defeat in the Betamax/VHS standards battle. Or the billion dollars that IBM invested in the failed OS/2 operating system. Or the hundreds of millions of dollars squandered when Yahoo! and Amazon launched full frontal assaults on eBay’s dominant U.S. auction franchise. Or eBay’s own failure to capture auction markets in Japan and China when it imported its U.S. pricing model into those markets, where its rivals initially offered free transactions.
It’s especially difficult to get pricing right in platform-mediated networks, so I’ll focus on that topic here. Specifically, I’ll explain why a platform must be proprietary—controlled by a single company—for its provider to capture value created through network growth. I’ll also give a brief overview of rules for setting prices in a “two-sided” network—one with two distinct user groups.
But getting your pricing right is just the tip of the iceberg when designing business models that leverage network effects. It would take a series of long posts to cover all of the relevant topics, so I’ll just mention a few of them below and point to some articles and notes that should be helpful to an entrepreneur grappling with these topics.
(Links marked below with an asterisk are available as PDFs for $7 apiece from HBS Publishing; others are available as free downloads. Sorry for not being able to provide all of my work for free, but we fund our research at HBS through the sales of cases and notes.)
Proprietary Platforms and Pricing Leverage
When a platform-mediated network’s user base expands, the resulting increase in users’ WTP for platform access does not automatically translate into higher prices for the platform. Providers only gain pricing leverage with network growth when they keep their platforms proprietary. A proprietary platform has a single provider who exclusively controls its technology, for example, eBay, Federal Express, or Google. With a shared platform such as barcodes, DVD, or Wi-Fi, multiple firms collaborate in developing the platform’s technology, and then compete with each other to provide differentiated but compatible versions of the platform. Compatibility ensures that users can switch between products offered by a shared platform’s various providers without incurring significant switching costs. For example, when switching from one Windows-compatible PC to another—say, from Dell to Compaq—a user need not invest in new application software. By contrast, switching to a rival platform—for example, from Windows to Macintosh—would be much more costly.
For a shared platform’s providers, lower customer switching costs limit pricing leverage. By contrast, a proprietary platform’s sole provider can raise its price when user base growth increases users’ WTP, up to the point where a user is indifferent between sticking with the platform and switching to a rival platform, if one exists. With such pricing leverage, startups can capture more of the value they create through network effects. Consequently, entrepreneurs will normally prefer proprietary platforms when designing business models. However, as I explain in this article*, there are circumstances when it may be advantageous to share platforms with rivals. In particular, when many equally matched rivals simultaneously enter a new market that is destined to be dominated by a single platform due to very strong network effects, the firms may be better off sharing the market than conducting an expensive winner-take-all battle*. Even though they offer competing versions of a compatible platform, firms may still be able to profit by contributing their technology to the shared platform, perhaps earning licensing fees or gaining a time-to-market edge.
Sharing a platform implies an open architecture. As I’ve written elsewhere on this blog, the concept of an “open platform” can lead to confusion, because platform-mediated networks encompass several ecosystem layers—each of which can be open or closed. In this paper, Geoff Parker, Marshall Van Alstyne and I describe circumstances under which platforms may find it attractive to open or close as they mature. In a more technical paper, we describe a strategy we call platform envelopment, which entails one platform owner moving into another’s market by offering a multi-platform bundle. Through envelopment, firms can surmount the high entry barriers that often protect incumbents who harness strong network effects. Examples include moves by Microsoft (with Explorer, Windows Media Player, Silverlight, etc.), Apple (with the iPhone/iPad, which has enveloped MP3 players, feature phones, PDAs, handheld gaming devices, standalone GPS units, eBook readers, etc.), and Google (with Checkout, Docs, Chrome, Android, etc.).
Pricing in Two-Sided Networks
Platform providers who seek to mobilize new two-sided networks face a Catch-22: each side will refuse to join until the other side is on board. To avoid this impasse, platform providers often subsidize users on one side, that is, they price below marginal cost to that side or even give its users free access to the platform, as with Adobe’s PDF reader software or Google’s search engine. Due to network effects, attracting users to this “subsidy side” boosts users’ WTP on the network’s “money side,” as with customers for Adobe’s PDF creation software or Google’s advertisers. Generally, it makes sense for platform providers to permanently subsidize the network’s more price sensitive side and charge the side that increases its demand more strongly in response to the other side’s growth.
Parker, Van Alsytne and I discuss pricing rules for two-sided networks in this article.* In a note* for MBAs, Andrei Hagiu and I examine staging strategies that avoid the Catch-22 in two-sided networks by first bringing on board one side in large numbers before seeking to mobilized the other side. Amazon used such a staging strategy. Amazon attracted a big base of customers using a traditional merchant business model, then later morphed into a two-sided platform by hosting third-party merchants who leveraged Amazon’s infrastructure and customer base.
Technical Appendix: Network Effects and WTP
In a market that exhibits network effects, a company must consider the present and future benefits of building its customer base when making current-period pricing and investment decisions. Absent network effects, a company typically maximizes current-period profitability by setting its price and marketing spending at levels that equate current-period marginal revenue and marginal cost. With network effects, however, optimization is less straightforward. Acquiring an additional customer—say, Mary—in the current period yields revenue directly from Mary during that period and also boosts—very slightly—the willingness to pay of all other customers, who value a larger network. Other customers’ willingness to pay remains higher by this tiny increment in future periods, until Mary exits the network. This increase in other customers’ willingness to pay in current and future periods should be factored into decisions about whether to invest in accelerated growth.
These dynamics can be illustrated using an example of a fictional company—let’s call it “Blossom”—selling spreadsheet software in 1989 (before most office productivity software was sold in bundled suites). Then as now, spreadsheet software was subject to a fairly strong network effect: users valued the ability to exchange files and required software that employed a compatible standard to do so. I use this example because scholars have estimated the impact of network effects on customer willingness to pay for spreadsheet software in 1989 using an econometric technique called hedonic regression (Brynjolfsson & Camerer, 1996, “Network Externalities in Microcomputer Software,” Managment Science 42: 1627-1647) Those estimates are employed in the stylized example below to show how network effects can influence optimal marketing spending levels.
Assume that Blossom’s unit price was $360 and its variable expenses (excluding customer acquisition costs) for supplying an additional unit were $72, or 20% of revenue. To keep the analysis simple, we also assume that a new customer would use the product for four years, at which point the product would be rendered technologically obsolete and the customer would be in no way loyal to Blossom. This assumption is unrealistic, but it lets us avoid factoring the present value of future product sales into our calculations. In reality, customers might prefer to buy Blossom’s next-generation product if it provided backward file compatibility or reduced the time spent mastering a new interface.
First, consider a scenario in which spreadsheet software was not subject to a network effect. What is the largest amount Blossom should be willing to spend to acquire an additional customer in 1989? This calculation is straightforward: the company could afford to spend up to $288 ($360 minus pre-marketing variable expenses of $72) to acquire a customer. Beyond $288, the marginal cost of acquiring and supplying a new customer would exceed the marginal revenue from that customer.
How would network effects influence the maximum amount that Blossom should be willing to spend to acquire a customer? The research mentioned above provides the following formula for predicting the logarithm of a spreadsheet’s product price in 1989 (“P”), based on its publisher’s installed base share (“S”), after controlling for differences in product quality. Consistent with the concept of network effects, the formula indicates that customers would pay more for a spreadsheet used by a bigger installed base:
Log P = 5.7376 + .0075S
If we assume that Blossom had a 20.000000% share of the total installed base of spreadsheet software, then this formula would predict customer willingness to pay equal to $360.5389517 (the reason for showing so many decimal places should be apparent in a moment). Assume further that:
- The total installed base for spreadsheet software was exactly 50,000,000 in 1989—about half of the actual size of the worldwide installed base of personal computers in that year.
- The spreadsheet market was not growing. This is clearly not realistic, but it simplifies our analysis in ways that do not meaningfully change the conclusions.
- Annual unit sales for the industry equaled one-quarter of the industry’s total installed base, corresponding to the four-year replacement cycle described above.
Obviously, a company cannot raise its price in increments as small as $0.0000054, but larger market share gains may yield material pricing benefits. Specifically, based on the formula above, a 1% share increase yields a 0.75% increase in customer willingness to pay for spreadsheet software in 1989. For this reason, network effects provide an incentive for companies to invest more aggressively in accelerated growth strategies—that is, to race for scale.
