Last week in my Utility Cloud Computing post, I promised to write a follow-up exploring network effects for platforms as a service (PaaS). This was sparked by an interesting exchange between Nick Carr and Tim O’reilly with commentary on the Smoothspan blog.
I think the important point to explore can be summarised as “software versus data” for the next big network effect driven opportunity.  Smoothspan argues for the software angle and Tim for the data angle.
For me this turns on a critical distinction between platforms for direct versus indirect network effects. (Take a few minutes to read http://en.wikipedia.org/wiki/Network_effect if you want a good introduction to network effects.)

The software industry is well aware of the defensibility of network effects; and no one can possible miss the incredible defensibility of the Windows empire. It is tempting to take a pattern that has succeeded before and copy it. But it is wise to consider whether certain critical factors may be different now. Microsoft built a platform business model that harnessed indirect network effects. One way to think of indirect network effect is value built by the availability of complimentors rather than by the activity of individual users. The more windows applications: the more reason for a user to choose windows, the more users, the more reason for developers to build windows-only applications. But my usage of MS Word does not affect yours.

A direct network effect arises when there is some form of shared system which means that one users activity directly effects the experience of another. Communications networks naturally create direct network effects, and the internet is the greatest communications network ever built. MS Windows, on the other hand, is not a shared system. No matter how many people are writing documents on MS Word, it doesn’t effect me in the least. This contrasts with people writing on Wikipedia—the most extreme counter example to MS Word. With MS Word, each user has an isolated copy of the same software. With Wikipedia, each user is sharing both the same software and content. Nearly the entire value of Wikipedia to me is the trail of activity other people have left on it. User activity on a shared system can effect other users because each user leaves a trail of new data as they interact with the existing data. Be that data a text message, a blog post, a web page, a video on youtube or a users clickstream on Google, it is data generated by user activity on a shared system which creates direct network effects. Web 2.0, collective intelligence and telecoms all create direct network effects based on data.

The Windows model worked because Windows was a single-homed platform. If I as a customer put windows on my computer, I could not easily also have OS/2 or Mac OS. That means that my choice of OS was heavily reliant on the world of applications that were available and going to be available on that platform. That was fair enough, it really was hard to have multiple operating systems, even down to having to learn different user interaction patterns.

MS understood that their indirect network effects hinged on the developer community because these were the complimentors.

But there are two huge differences now compared to when Microsoft started building its software ecosystem back in the day.

The first is open source. Software can be built in a modular fashion which makes it very easy to build more software through recombination of pieces (e.g. LAMP stack, apache etc). Open source software is a massive toolkit that through recombination just keeps growing. By taking a shared approach to solving software problems, it becomes easier to solve the next by drawing on the pool. This is a huge indirect network effect not limited by the friction of license cost. Because a relatively small number of people can solve general problems which everyone can then share in, any generally applicable software problem is very hard to make money from because it is getting easier and easier for the solution to be assembled from open source components. When you add utility cloud computing into the mix, it becomes easy for SaaS offerings to be attacked by open source also.

There is no general solution to the data problem. Your best guys in silicon valley can solve the problem of how to graph a data series easily, but they can’t create for me a graph of my personal calorie intake for the last week. That data is specific and localised. If  I want that specific graph, my problem is not the software required to generate a graph, it is the data problem of my calorie intake. Some data is general, but the vast majority of data is specific. This makes the data problem very different from the software problem.

The data is where the users are, the software is where the developers are.

Secondly the important thing to note is that PaaS  based on open standards are NOT single homed for the consumer of applications; especially with the rise of standards for interoperability like web services, REST, XML and RDF etc. If I use BaseCamp, it doesn’t make it any harder for me to also use SalesForce.com or EC2.

This fundamentally makes the API level lock in combined with a developer community model very unstable compared to the good old days of deployed software (e.g. Oracle, SAP, and similar models). In fact for SalesForce.com, I would say the greatest thing that draws people in is that their central business data is in the platform, rather than the wide choice of complimentors.

I don’t think software and developer communities provides the same kind of strong platform network effects today. Look around at the successful network effect players. It isn’t the developer community that has made them defensible (e.g. eBay) but the direct network effects between users. This is a very different game and one I think MS will have a very hard time really understanding. The dichotomy is that software is essential for data-centric direct network effects, but isn’t where the defensible value really is.

Putting the approaches another way:
Developers are where the software is.
Users are where the data is.

So with a direct network effect, it is the usage of the shared system by users which makes the system itself more valuable to each user through the trail of data they leave. Software, just like hardware, is required to enable data but it is the data which generate the value. You can see this exact effect with Wikipedia, Facebook, Skype, IM, Twitter and Google.

Let me expand on Google a bit. I don’t pretend to fully understand Google, maybe no one does (including Larry and Sergey). It is easy to see that the web is an open network with very strong direct and indirect network effects, but understanding how Google turns it into a profitable business is less so. Google harnesses the content and collective intelligence in the links to create relevance data but it also watches the activity of its users to derive further collective intelligence and improve the experience. The first is an open system which anyone with enough compute power and bandwidth could copy; the second is based on the large volume of users coming to Google and is closed. Only those who can generate that kind of search traffic can have access to that. So, those are two direct network effects: one based on an open network of users (people who create web content), the other on a closed network of users (people who search on Google).

Google’s great fortune was to keep (by luck or judgement) their software to themselves. Instead of selling search technology to power other search services and becoming a technology supplier, they used their best in class technology to take huge market share in search and created a global consumer BRAND. Let me say it again: global BRAND. There are no magic network effects that keep people glued to using Google for search. Maybe Google is a bit better than the others but I wouldn’t know because I haven’t tried them. It’s mass market: Want a cold drink? Coke please. Want to search the web? Google please. Google uses its infrastructure scale and its talent scale to keep wowing the public—to keep Google top of mind for amazing things like search. I switched on a 24hrs news channel the other day and noticed the map behind the presenters head had a big Google logo on it, it was Google Earth. You can’t buy that kind of brand visibility. The Google brand is everywhere. With—by far—the largest search traffic, Google is naturally where advertisers want presence because that is where the eye balls are. Yes, Google has used network effects again to maximise its advertising revenue; but that is only possible because of the huge search traffic its infrastructure scale, talent scale and BRAND give it.

So, Google creates huge value from network effects but doesn’t actually tie users in through a closed network. Google harnessed the value created by the open platform of the web, its defence is scale of infrastructure and talent in order to maintain itself as THE BRAND in the public mind for search on the web to drive ad sales. With scale, talent and brand built on an open platform—its key levers—it is no surprise that Google wants to play the same game by ensuring the mobile platform (Android) is open and the web platform stays open (Chrome). Apple, on the other hand wants to own a closed mobile platform. I can’t help but think Google will be the long-run winner.

Which do you need more: the community of complimentors or the users, their activity and data? But be under no illusion that the software game has changed for ever. SaaS & PaaS platforms are not naturally single-homed, and that was a crucial part of the old game. How are you going to create software-based network effects with out hurting your customers through lock in? And, how do you stop your community being outgrown by the open source ecosystem on open utility computing infrastructure?

My advice: it isn’t about the software anymore, its about the data.