Category Archives: Network Effects


Content craves a network, not elitist isolation, which is why clever harnessing of content can drive the creation of a social network.  This is the example.

In academia, each field of study has a prestige journal–or two or three of them–and they guard their articles jealously. This is how they make money. Price worked around this. There are actually three stages an academic article goes through. The first is the precious baby produced by the authors. This is the pre-print version. Virtually all journals permit authors to do whatever they want with this. Then that baby is sent out for review and learned peers weigh in with comments and these are incorporated into the paper. Revisions can be minor or major. This is the author version and about half of all journals permit authors to publish it. Lastly, there is the final published version of the article in journal format. Journals guard this closely and do not permit easy dissemination. encourages researchers to post their pre-print and author versions of papers. Most of the important information is in these and it becomes immediately accessible.

Data Karma

My wife and I were raised Hindu and Jain respectively, both religions that have karma as a central element.  One part of the notion of karma is that you get what you give.

That may be an useful way to think about the cold start problem that one encounters when trying to create data network effects.  One company that has done it well is Markit in various financial markets.  The simple insight is that in order to benefit from the valuable data output, you need to input your own data: i.e, you give, then you get.  Its story has it all: opaque markets, prices all over the place, and a benefit from aggregation and transparency.

This is from the Economist:

Mr Uggla’s big insight, as a former bond trader, was that opaque market structures made it tricky for banks accurately to price some of the complex financial instruments they were dealing in. Only by pooling their proprietary data could they get reliable marks. The trick was first applied to credit default swaps (CDSs), a sort of insurance policy against borrowers going bust. Unlike shares, CDSs are not traded on exchanges with transparent prices. They are bilateral contracts that give rise to a jumble of erratic price quotes. Only those banks that fed their prices into the Markit system got access to the aggregate data. They also got a majority stake in the company: Markit’s owners include the likes of Goldman Sachs, JPMorgan Chase and UBS. Staff own 30%, a handful of other investors the rest.

What came next distinguishes Mr Uggla’s approach from that of Michael Bloomberg, another former bond trader with a successful data franchise. Markit has moved into a broader role as a partner for banks that want to pool or outsource costly non-core activities.

At a facility in Dallas, for example, it receives and processes 7m faxes a year on behalf of customers. Few are of any interest to the recipients—they are mainly updates on the progress of loan repayments—but regulations require that they must be kept on file regardless. Another Markit product helps make sure that complex deals devised by hotshots on trading floors are recorded and processed accurately. Yet another ensures all sides to a transaction are using the same formula when calculating collateral payments. Last year it acquired a company central to securities lending, a practice that enables short-sellers to borrow the shares they want to bet against.


You are nobody on Wall Street without a hulking Bloomberg terminal on your desk.

in an age of cost-cutting and easily accessible information, that reality is in tension with its $20,000/year annual cost per terminal

The reason for the terminal’s enduring power lies in one of the most powerful and enduring local network effects of our time in the messaging feature of the terminal:

A private network since its foundation in 1982, before email was widely used or social networks emerged, Bloomberg has used its messaging technology to turn its financial data service into the hub of a vast social network.

“Bloomberg is like a very expensive Facebook,” an executive at one rival data company observes.

In a world where many market infrastructure operators provide a cheap, often free messaging tool, Bloomberg reigns supreme. Each day, its 315,000 subscribers exchange 200m messages and have 15m to 20m chats. Rivals have tried for years to break its dominance in messaging, with little success.

Financial groups with security and compliance concerns about Facebook or Twitter like Instant Bloomberg for its security, including biometric identification, and the fact messages are archived and auditable. Users like functions allowing them to share complex data sets, integrate with Yahoo or AOL chat services, or simply see whether someone has received a message. Others have to have it simply because their customers use it.

But network effects while robust are not indestructible, and the current scandal is giving those customers that pay the bills the opportunity to reevaluate alternatives and chip away at the reliance on the terminal.

Transactional Local Network Density

So local network effects suggests that the cold start problem to start building network effects may not actually be a great problem.

This is particularly true for enterprise oriented transaction platforms.

Like the homecoming queen sending out invites  to join an emerging IM network 15 years ago, local network density that matters to the customer can quickly be established.  The local network density of having the homecoming queen on the network is what matters for the teenage boy even if the IM network does not yet have reach beyond that local network effect.  A big enterprise can get its suppliers onto the network if there is a real threat that business is lost because certain business is bid out through the platform.

The message that must be sent by the enterprise — “Sign up, compete, or lose the business.”  Suppliers will do funny things if you make them break a sweat.

The fact that networks overlap, customers can benefit from other customer’s suppliers, and suppliers can serve other supplier’s customers, means that local network density can translate quickly to a real global network effect.

To circle back, this is the Pollenware example., in order to have the best auction of your accounts payable, you want everyone on there in your network to which you own an accounts payable.

The Cold Start Problem and Local Network Effects

Facebook has a billion members.  But if your couple hundred friends are not members, the wonders of network effects are purely theoretical.

Similarly, take an instant messaging network.  On an instant messaging network, a user wants to communicate with only a small fraction of the actual and potential user base of the network.

Here is one definition of local network effects:

For example, a good displays local network effects when rather than being influenced by an increase in the size of a product’s user base in general, each consumer is influenced directly by the decisions of only a typically small subset of other consumers, for instance those he or she is “connected” to via an underlying social or business network (instant messaging is a great example of a product that displays local network effects).

As the user benefits when his or her “local network” joins the network, that user has an incentive to get his or her network to sign up.  This provides a potential solution to the cold start problem  – how does one jump-start network effects?

Local networks lead to global network effects. Local networks interlock and overlap; they are not isolated.  Local networks of friends on Facebook or IM network interconnect with other local networks.

The key lies in nurturing that interconnection.  By empowering and encouraging the local network, the global network can rise.

Network Effects in Data

Tim O’ Reilly wrote one of the early great posts on the increasing importance of network effects in data.  Data network effects make products better as a function of more users contributing data to the product.  Here are a couple of key excerpts from O’Reilly’s post:

Ah, I say to myself: Nick only sees first order network effects, what you might call endogamous networks, those that require the user to be part of the tribe. Thus, phone networks, and networks like Facebook.  But the internet is an exogamous network; its benefits increase by the extent to which it reaches out to new groups, increases cross-breeding, and thus the total robustness and variety of the gene pool. This is why links matter, why web services matter, because they extend the reach of the network. Understanding the benefit of exogamous networks requires a more subtle calculus than Nick is applying. It’s not necessarily that you benefit directly from belonging, but the fact that you belong allows others to harvest the benefit of your participation.

The next layer of attractive profits will accrue to companies that build data-backed applications in which the data gets better the more people use the system. This is what I’ve called Web 2.0.

It’s network effects (perhaps more simply described as virtuous circles) in data that ultimately matter, not network effects per se. Nick probably wouldn’t think of Nuance as a network-effects driven company, but it is, because their applications and services depend on data that gets better the more people use it (or have their data harvested in one way or another.) More speech in more circumstances and more domains makes Nuance better for the next user. No user thinks that Nuance is better because of them (and because many of Nuance’s products are standalone, this is in fact true.) Yet Nuance, like Google, has figured out how to harvest data contributed by millions to build a better product.


Data Network Effects in Enterprise – A Converse Coase Theory

Here are two comments that I wrote soon after attending the Fred Wilson talk on network enterprises.  These explain the commonality between data network effects and the recurring theme on this blog of intermediating talent through networks rather than enterprises:

Like the original post, this talk was EPIC. Not only learned a lot, but left me with a lot to chew on.

What I took away is that the key to network efforts is getting users to contribute “data” to the network to create something better for the user that is also defensible for the platform owner.

In consumer, it was about getting users to share “data” that they didn’t know they wanted to share. Facebook (posts, photos, etc.); 4Sq (location); Twitter (micro-thoughts).

In enterprise, it is complicated by the dominant default assumption, that “data” should stay within the four walls of the enterprise.

Finding ways to overcome that can yield so much market surplus, because it is in many ways, such a self-evidently wrong assumption.

I have been thinking about it as the converse of Coase — while the creation of the firm solved inefficiencies associated with transaction costs, it created inefficiencies related to “data” flow across enterprises. Networks in enterprise are the way to correct some of those inefficiencies.


But in reaction to your talk, one thought that I had was that the notion of data network effects is the other side of Coase’s Theory of the Firm.

An overlooked potential corollary of his theory that a firm-based economy is a response to contractual transaction costs is that there is also an overlooked cost to efficiency in predicating an economy on the firm.  A firm’s first instinct is to keep things within its four walls, so knowledge flow is inhibited across the economy.

By finding ways to devolve enterprises into networks as an extreme case or force enterprises onto networks as the more common case, you fight that deadweight loss in our economy by transferring knowledge across firms across the economy.  Sort of like the pre-calc teachers finally sharing lesson plans with each other via edmodo or patients potentially sharing data among themselves bypassing drug companies and governments.  In the non-network world, this is what a lot of professional firms like McKinsey do — take knowledge and spread it around.

So perhaps, part of the network effects thesis in enterprise is freeing knowledge by using networks to group people in ways that are in addition to or in replacement of the firm.

User Contributions, Network Effects

For defensibility of an enterprise business model and avoiding the software morality play, the key in Fred’s view is gaining network effects in data contributed by the users of the service.

Data contributed by users is key.  It is not sufficient to pull data from internal or public sources; instead each user must send back data.  Usage then potentially builds an insurmountable lead in the network, since an entrant cannot catch up merely by replicating the service and pricing it cheaper.

So what are concrete examples of data network effects?  Some examples of the data contributed by usage come from the USV platform:

  • Work Market: Company customers put their entire “contingent labor” or freelance workforce onto to the platform, creating a supply platform accessible to all customers.
  • Pollenware: Company customers bring their suppliers onto the platform in order to bid on accounts receivables.
  • Return Path: Return Path provides email software to heavy email senders in exchange for the information as to who are good or bad senders

In a data network effects model, every new user makes the network more valuable through the contribution of some sort of data.  Every new user makes it harder for all users to get off.

The Fundamental Vulnerability of Software Businesses (Ankle-Biters and Open-Source)

Right before the holidays, I attended an amazing presentation by Fred Wilson on software and the enterprise.

The thesis of the presentation was network effects, and it underscored to me that in developing a product a key test has to be whether there is a sustainable network effect in addition to delivering value.

This has inspired me to write an ongoing series of posts on network effects.

The first question that we will take on is why network effects are important in software businesses.

Fred demonstrated his answer in the form of something he called a software morality play in three acts:

  • Act 1: A high end enterprise software company launches costing hundreds of thousands dollars in recurring fees.
  • Act 2: This value is undermine by a couple of entrepreneurs funded by Paul Graham working on the software in their garage selling it on self-serve basis rather than through a large sales force.  This venture serves as an ankle-biter to the big enterprise company causing pressure on prices.
  • Act 3: An open-source technology is launched further destroying the value of big enterprise company.

Network effects serve to protect the venture from the fundamental point of Fred’s morality play: software is easily duplicated.  Network effects are most importantly about defensibility.

More on what those networks effects are in future posts.  Fred’s talk is here.