Confronting the Limits of Networks

Around 1980 Robert Metcalfe, the inventor of the Ethernet standard and founder of 3Com, developed a simple but powerful model to describe how networks become more important as they grow. He observed that a network’s value lies in the number of links it enables among members and that the quantity of links increases as the square of the total of the network’s members. Result: The value of a network increases in proportion to the square of the number of people using it.

This observation came to be known as Metcalfe’s Law. It was similar to an idea developed by economists about “network effects” — meaning that some resources become more valuable to a person using them according to the number of other people also using them. The telephone system long ago demonstrated the validity of this idea, but at the dawn of the Internet era, network effects became the Holy Grail for many business builders, who wanted to “get big fast” in order to exploit them before the competition did. Many Internet-era business models and technologies, including online marketplaces and communities, person-to-person auction sites, and instant messaging, are based on the assumption that network effects will profitably apply.

A few examples, however, show that Metcalfe’s Law doesn’t always hold. The average American’s telephone service, for instance, would not become much more valuable if everyone in China and India suddenly owned a telephone. And some users can actually make a network less valuable for others, as everyone who receives daily junk e-mail knows. Metcalfe himself recognized the limits of his idea; as he once pointed out, “The law may be optimistic as the number of people on a network gets very large.”1 The good news is that there are strategies that managers can use to maintain the power — or at least halt the decline — of network effects.

Network Trouble

It’s important to understand the phenomena that can put the brakes on Metcalfe’s Law, slowing or even reversing network effects. They include:

Saturation.

In some networks, users add value by bringing resources such as song files or physical goods to be shared or traded. Since the number of different resources that users can provide is finite, the network is eventually saturated once most possible resources are available on it. After that, a new user is unlikely to add value. Napster, for example, had about 5 million users in early 2000. New users didn’t have much to add at that point, because most songs that had been distributed in a few thousand physical copies were already likely to exist on the site. A good indicator of saturation is the success rate of a networkwide search: the higher it is, the closer the network is to saturation.

Cacophony.

When too many users of a network make interaction difficult — in a crowded Internet chat room, for example — cacophony is the result. This problem isn’t just a matter of everyone “talking” at once; online discussion groups, even those with tracking mechanisms such as threads, can become quite hard to follow when too many people are participating.

Contamination.

Users sometimes actively reduce or destroy the value of a network out of malice or desire for gain, or by expressing viewpoints that are sharply different from those of other members. Most users of a network have no trouble agreeing on the kinds of people and activity that destroy value: E-mail “spammers” (purveyors of junk mail), discussion group “flamers” (abusive and profane debaters), fraudulent bidders on auction sites, and adults who interact inappropriately with children are easily identified as contaminators. In other cases, the issue is less clear: Adult content on the Web, for example, is valuable to some but offensive to others.

Clustering.

Clusters form when members consistently use only a small part of a network that is available to them. Virtually all networks exhibit some level of clustering; for some, it is pronounced — people typically exchange instant messages, for instance, with only a small group of family, friends and colleagues. Economist Paul Krugman’s words are apt in this context: “The flaw in Metcalfe’s Law … becomes apparent: Most people have nothing to say to each other!”2

Search Costs.

As a network becomes larger, it can also become harder to navigate — that is, users may find it more time consuming to find a sought-after resource or member, especially if the network is not centrally managed. This is one of the main frustrations of the Web: A searcher often knows that the needed Web page or site is out there somewhere but cannot easily locate it.

Limiting the Damage

These phenomena mean that larger networks can in some cases have less value than smaller ones. But there are ways of maintaining network effects or limiting their decline. We have identified several different strategies that network builders can employ. They fall into two categories: those that acknowledge limits and those that combat them. The distinction is important because some of the limiting phenomena are inherent to networks and cannot be directly counteracted, while others can. When these strategies are followed properly, they are more effective than a blind bigger-is-better approach in which network builders rush to sign up as many users as quickly as possible.

Work-Around Strategies.

Two of the problems we’ve identified, saturation and clustering, are unavoidable. As a resource-trading network grows in popularity, it will eventually contain most available resources, and members of any network will usually want to interact with only a small group of peers. Network builders can deal with both of these issues through the deliberate selection of prospective members. When a resource-trading network is still relatively new, for example, companies should try to identify people with the most (or most valuable) resources and offer them incentives to join. Deliberate selection can also turn clustering to the advantage of the network builder by getting all cluster members signed up quickly. MCI increased the size and value of its network by selecting new members based on its customers’ already existing clusters. Its Friends & Family calling program gave members incentives to sign up everyone they talked to frequently.

It’s also possible to encourage the formation of new clusters. EBay continually subdivides its product categories, helping site visitors to home in tightly on their interests. The Internet’s Usenet maintained its value over time partly by making it easy to set up ever more specific or focused news groups. Instant messaging “buddy lists” are also convenient ways to build and access clusters.

A related strategy is bridging: introducing members to new clusters that might be of interest to them. Again, eBay provides a good example. It lets buyers easily see the other products that a seller is offering. In many cases, the seller offers goods in other clusters that might be of interest to the buyer. The buyer-seller link thus forms a bridge between the clusters for the buyer. Recommendations at the DVD rental site Netflix, which are based on the rental patterns of other members, often jump from one film category to another and so offer bridges for members to cross.

Work-Through Strategies.

Unlike clustering and saturation, cacophony and contamination can be fought actively. Although it may sound strange, turning away prospective members and kicking out current ones can be highly effective ways to keep Metcalfe’s Law in force. Moderated news groups and chat rooms, Web site membership policies, e-mail spam filters, and software that blocks objectionable Web sites all help keep members engaged. Gatekeeping slows the growth of networks but also delays the onset of cacophony and keeps contaminators out. Monitoring is intrusive, but it provides a way to identify value destroyers within a network so they can be removed. The networks of programmers that collaborate over the Internet on open source projects such as the Linux operating system feature both monitoring and gatekeeping.

Another strategy is to offer effective search assistance within the network. The search engine Google has become popular because it does an excellent job of helping Web users quickly find what they’re looking for. Since the Web is huge, growing and almost completely decentralized, surfers often have difficultly finding what they need or even knowing where to look. By addressing that problem, Google helps maintain and increase the entire Web’s value.

When a network has a single owner, it is technically straightforward to incorporate search capabilities. EBay’s sprawling listing of goods for sale is easy to search, as is the phone network. But some networks might not want to be searchable. The instant-messaging networks built by AOL, Microsoft, Yahoo!, and others are extremely useful to their members but are also highly clustered. While some users might gladly accept messages and chat invitations from anyone else using the same software, most probably would prefer to remain invisible to all but family, friends and colleagues. A comprehensive searchable member list would likely be a bad idea unless there were clear opt-in and privacy policies.

Finally, managers in many businesses can find ways to build in network effects. The value of an e-tail site, for example, wouldn’t necessarily increase with the number of shoppers; once the company has enough transactions to obtain bulk discounts from suppliers, new customers could actually be bad news for current ones since they could slow down the site and consume scarce merchandise. Amazon and many other e-tailers, however, have found ways to make their customers valuable to one another. Amazon does that by generating recommendations based on what other shoppers have bought, by including customer reviews of products, and by allowing customers to post “favorites” lists and even to sell used merchandise (including merchandise that Amazon stocks). The ultimate goal of all of these techniques is to make the Amazon site as valuable as possible; the means to this end are network effects.

The bottom line is this: Business builders should not automatically expect that a network’s value will continue to increase geometrically as new members join. That doesn’t mean that managers should ignore Metcalfe’s Law, which does provide real insight into the reason that networks become ever more valuable — up to a point. The law does not explain, however, why that process slows down or even reverses course as networks become very large. Managers who understand the reasons for this can begin to take a more nuanced and effective approach to network building in which size isn’t everything.