Harvard Business Review

In 2016, Didi became the world’s largest ride-sharing company, reaching 25 million trips a day in China and surpassing the combined daily trips of all other ride-sharing companies across the globe. It had arrived at this milestone by merging in 2015 with its domestic rival, Kuaidi, and pushing Uber out of the Chinese market after a fierce, expensive battle. With its competition gutted, Didi gradually began to improve its margins by reducing subsidies to drivers and passengers.

But just as the company began to reach profitability, in early 2018, Meituan, a giant player in online-to-offline services such as food delivery, movie ticketing, and travel booking, launched its own ride-hailing business in Shanghai. Meituan didn’t charge drivers to use its platform for the first three months and afterward took only 8% of their revenues, while Didi took 20%. Drivers and passengers flocked to the new service. In April, Didi struck back by entering the food delivery market in Wuxi, a city close to Shanghai. What followed was a costly price war, with many meals being sold for next to nothing because of heavy subsidies from both companies. So much for Didi’s profitability.

Didi was taking other hits too. In March 2018, Alibaba’s mapping unit—Gaode Map, the largest navigation service in China—had started a carpooling business in Chengdu and Wuhan. It didn’t charge drivers at all, and in July it began offering passengers the option of ordering from several ride-hailing services. Meanwhile, Ctrip, China’s largest online travel service, had announced in April that it had been granted a license to provide car-hailing services across the country.

Why hadn’t Didi’s immense scale shut down its competition for ride services in China? Why wasn’t this a winner-take-all market, as many analysts had predicted? Moreover, why do some platform businesses—such as Alibaba, Facebook, and Airbnb—flourish, while Uber, Didi, and Meituan, among others, hemorrhage cash? What enables digital platforms to fight off competition and grow profits?

A video game console with a small technical advantage can easily steal share.

To answer those questions, you need to understand the networks a platform is embedded in. The factors affecting the growth and sustainability of platform firms (and digital operating models generally) differ from those of traditional firms. Let’s start with the fact that on many digital networks the cost of serving an additional user is negligible, which makes a business inherently easier to scale up. And because much of a network-based firm’s operational complexity is outsourced to the service providers on the platform or handled by software, bottlenecks to value creation and growth usually aren’t tied to human or organizational factors—another important departure from traditional models. Ultimately, in a digital network business, the employees don’t deliver the product or service—they just design and oversee an automated, algorithm-driven operation. Lasting competitive advantage hinges more on the interplay between the platform and the network it orchestrates and less on internal, firm-level factors. In other words, in the digitally connected economy the long-term success of a product or service depends heavily on the health, defensibility, and dominance of the ecosystem in which it operates.

And as Didi is learning, it’s often easier for a digital platform to achieve scale than to sustain it. After all, the advantages that allow the platform to expand quickly work for its competitors and anyone else who wants to get into the market. The reason that some platforms thrive while others struggle really lies in their ability to manage five fundamental properties of networks: network effects, clustering, risk of disintermediation, vulnerability to multi-homing, and bridging to multiple networks.

Strength of Network Effects

The importance of network effects is well known. Economists have long understood that digital platforms like Facebook enjoy same-side (“direct”) network effects: The more Facebook friends you have in your network, the more likely you are to attract additional friends through your friends’ connections. Facebook also leverages cross-side (“indirect”) network effects, in which two different groups of participants—users and app developers—attract each other. Uber can similarly mine cross-side effects, because more drivers attract more riders, and vice versa.

Less well acknowledged is the fact that the strength of network effects can vary dramatically and can shape both value creation and capture. When network effects are strong, the value provided by a platform continues to rise sharply with the number of participants. For example, as the number of users on Facebook increases, so does the amount and variety of interesting and relevant content. Video game consoles, however, exhibit only weak network effects, as we discovered in a research study. This is because video games are a hit-driven business, and a platform needs relatively few hits to be successful. The total number of game titles available isn’t as important in console sales as having a few of the right games. Indeed, even an entrant with only a small technical advantage (and a good business development team) can steal significant market share from incumbents. That explains why in 2001 Microsoft’s new Xbox posed such a threat to Sony’s then-dominant PlayStation 2, and why each console has gone up and down in market share, alternately taking the lead, over the years.

Even more critically, the strength of network effects can change over time. Windows is a classic example. During the heyday of personal computers in the 1990s, most PC applications were “client based,” meaning they actually lived on the computers. Back then, the software’s network effects were strong: The value of Windows increased dramatically as the number of developers writing apps for it climbed, topping 6 million at the peak of its popularity. By the late 1990s Windows seemed entrenched as the leading platform. However, as internet-based apps, which worked across different operating systems, took off, the network effects of Windows diminished and barriers to entry fell, allowing Android, Chrome, and iOS operating systems to gain strength on PCs and tablets. Mac shipments had also begun to rise in the mid-2000s, increasing more than five-fold by the end of the decade. This turn of events illustrates that when an incumbent’s network effects weaken, so does its market position.

It is possible for firms to design features that strengthen network effects, however. Amazon, for example, has built multiple types of effects into its business model over the years. In the beginning, Amazon’s review systems generated same-side effects: As the number of product reviews on the site increased, users became more likely to visit Amazon to read the reviews as well as write them. Later, Amazon’s marketplace, which allows third parties to sell products to Amazon users, generated cross-side network effects, in which buyers and third-party sellers attracted each other. Meanwhile, Amazon’s recommendation system, which suggests products on the basis of past purchase behavior, amplified the impact of the company’s scale by continually learning about consumers’ preferences. The more consumers used the site, the more accurate the recommendations Amazon could provide them. While not usually recognized as a network effect per se, learning effects operate a lot like same-side effects and can increase barriers to entry.

Network Clustering

In a research project with Xinxin Li of the University of Connecticut and Ehsan Valavi, a doctoral student at Harvard Business School, we found that the structure of a network influences a platform business’s ability to sustain its scale. The more a network is fragmented into local clusters—and the more isolated those clusters are from one another—the more vulnerable a business is to challenges. Consider Uber. Drivers in Boston care mostly about the number of riders in Boston, and riders in Boston care mostly about drivers in Boston. Except for frequent travelers, no one in Boston cares much about the number of drivers and riders in, say, San Francisco. This makes it easy for another ride-sharing service to reach critical mass in a local market and take off through a differentiated offer such as a lower price. Indeed, in addition to its rival Lyft at the national level, Uber confronts a number of local threats. For example, in New York City, Juno and Via, as well as local taxi companies, are giving it competition. Didi likewise faces a number of strong contenders in multiple cities.

Now let’s compare Uber’s market with Airbnb’s. Travelers don’t care much about the number of Airbnb hosts in their home cities; instead, they care about how many there are in the cities they plan to visit. Hence, the network more or less is one large cluster. Any real challenger to Airbnb would have to enter the market on a global scale—building brand awareness around the world to attract critical masses of travelers and hosts. So breaking into Airbnb’s market becomes much more costly.

It’s possible to strengthen a network by building global clusters on top of local clusters. While Craigslist, a classified ad site, primarily connects users and providers of goods and services in local markets, its housing and job listings attract users from other markets. Facebook’s social games (like FarmVille) established new connections among players who were strangers, creating a denser, more global, more integrated network, which is easier to defend from competition. Both Facebook and WeChat, a popular social-networking app in China, have been enhancing their networks by getting popular brands and celebrities—those with national and often international appeal—to create public accounts and post and interact with users.

Risk of Disintermediation

Disintermediation, wherein network members bypass a hub and connect directly, can be a big problem for any platform that captures value directly from matching or by facilitating transactions. Imagine that you hire a house cleaner from a platform like Homejoy and are satisfied with the service. Would you really go back to Homejoy to hire the same person again? If a user has found the right match, there’s little incentive to return to the platform. Additionally, after obtaining enough clients from a platform to fill his or her schedule, the house cleaner won’t need that platform anymore. This was exactly the problem that doomed Homejoy, which shut down in 2015, five years after it was founded.

Which Network Structure Is More Defensible?

Some digital networks are fragmented into local clusters of users. In Uber’s network, riders and drivers interact with network members outside their home cities only occasionally. But other digital networks are global; on Airbnb, visitors regularly connect with hosts around the world.

Platforms on global networks are much less vulnerable to challenges, because it’s difficult for new rivals to enter a market on a global scale.

Platforms have used various mechanisms to deter disintermediation, such as creating terms of service that prohibit users from conducting transactions off the platform, and blocking users from exchanging contact information. Airbnb, for example, withholds hosts’ exact locations and phone numbers until payments are made. Such strategies aren’t always effective, though. Anything that makes a platform more cumbersome to use can make it vulnerable to a competitor offering a streamlined experience.

Some platforms try to avoid disintermediation by enhancing the value of conducting business on them. They may facilitate transactions by providing insurance, payment escrow, or communication tools; resolve disputes; or monitor activities. But those services become less valuable once trust develops among platform users—and the strategies can backfire as the need for the platform decreases. One of us, Feng, and Grace Gu, a doctoral student at Harvard Business School, saw this effect in a study of an online freelance marketplace. As the platform improved its reputation-rating system, trust between clients and freelancers grew stronger, and disintermediation became more frequent, offsetting the revenue gains from better matching.

Some platforms address disintermediation risks by introducing different strategies for capturing value—with varying results. Thumbtack, a marketplace connecting consumers with local service providers such as electricians and guitar teachers, charges for lead generation: Customers post requests on the site, and service providers send them quotes and pay Thumbtack fees if those customers respond. That model captures value before the two sides even agree to work together and has helped save the company from withering like Homejoy. Thumbtack today is handling over $1 billion worth of transactions annually. The downside of its revenue model is that it doesn’t prevent the two sides from building a long-term relationship outside the platform after a match.

Alibaba took a different approach with its Taobao e-commerce platform. When Taobao entered the market, in 2003, eBay’s EachNet had more than 85% of the Chinese consumer-to-consumer market. However, Taobao didn’t charge listing or transaction fees and even set up an instant-messaging service, Wangwang, that allowed buyers to ask questions directly of sellers and haggle with them in real time. In contrast, EachNet charged sellers transaction fees and, because it was concerned about disintermediation, didn’t allow direct interactions between buyers and sellers until a sale had been confirmed. Not surprisingly, Taobao quickly took over leadership of the market, and at the end of 2006, eBay shut down its Chinese site. Taobao today continues to offer its C2C marketplace services free of charge and captures value through advertising revenues and sales of storefront software that helps merchants manage their online businesses.

After estimating that it could lose as much as 90% of its business to disintermediation, the Chinese outsourcing marketplace ZBJ, which launched in 2006 with a model of charging a 20% commission, began looking for new revenue sources. In 2014 it discovered that many new business owners used its site to get help with logo design. Typically, the next job those clients would need done was business and trademark registration, which the platform started to offer. Today ZBJ is the largest provider of trademark registration in China—a service that generates more than $70 million in annual revenue for the firm. The company has also significantly reduced its transaction fees and focused its resources on growing its user base instead of fighting disintermediation. As the experience of ZBJ, which is now valued at more than $1.5 billion, shows, when disintermediation is a threat, providing complementary services can work a lot better than charging transaction fees.

Vulnerability to Multi-Homing

Multi-homing happens when users or service providers (network “nodes”) form ties with multiple platforms (or “hubs”) at the same time. This generally occurs when the cost of adopting an additional platform is low. In the ride-hailing industry, many drivers and riders use both, say, Lyft and Uber — riders to compare prices and wait times, and drivers to reduce their idle time. Similarly, merchants often work with multiple group-buying sites, and restaurants with multiple food-delivery platforms. And even app developers, whose costs are not trivial, still find it makes sense to develop products for both iOS and Android systems.

When multi-homing is pervasive on each side of a platform, as it is in ride hailing, it becomes very difficult for a platform to generate a profit from its core business. Uber and Lyft are constantly undercutting each other as they compete for riders and drivers.

Incumbent platform owners can reduce multi-homing by locking in one side of the market (or even both sides). To encourage exclusivity, both Uber and Lyft gave bonuses in many markets to people who completed a certain number of trips in a row without rejecting or canceling any or going offline during peak hours. And while rides are in progress, both platforms provide drivers new requests for pickups very close to current passengers’ drop-off locations, reducing the drivers’ idle time and hence the temptation to use other platforms. Yet because of the inherently low cost of adopting multiple platforms, multi-homing is still rampant in ride sharing.

Attempts to prevent multi-homing can also have unintended side effects. In one research project, Feng and Hui Li of Carnegie Mellon University examined what happened in 2011 when Groupon retooled its deal counter—which tracks the amount of people who have signed up for a specific offer on its site—to show ambiguous ranges, rather than precise numbers. It then became more difficult for LivingSocial to identify and poach the popular merchants on Groupon. As a result, LivingSocial started to source more exclusive deals. While Groupon was able to reduce merchant-side multi-homing, the research found, consumers became more likely to visit both sites, because there were fewer overlapping deals on them, and it cost little to multi-home. That finding points to a key challenge platform firms face: Reducing multi-homing on one side of the market may increase multi-homing on the opposite side.