Google Analytics Premium, Enterprise, and Disruptive Innovation

Google Analytics Premium was announced today. Finally. It wasn't really a secret.

What do you get for an enterprise fee?

Dedicated support, a number to call, no data caps, some attribution modeling (nice), and now, 50 custom variables.

There's good literature around disruptive innovation in web analytics, with a very specific vocabulary and model build around.

Is this disruptive or incremental?

• The increase to 50 custom variables is purely an increment on an established dimension.

• Enterprise support is incremental from the previous version of support. They did have a type of support. It was vague. But it was there. So that's an increment.

• Increasing data caps is incremental.

• The guarantees are incremental. There was protection in the past. There's more protection now.

• The attribution modelling is a new. It could be disruptive.

Is attribution modelling disruptive?

It's certainly a new competitive dimension. It's likely to be far more transparent than most MMM firms. And it's a natural extension of Google's core competency. It helps people understand why things are happening. Original features that assists people in making better decisions down the line could, potentially, probably, be considered disruptive.

Watch this feature.

It's squarely in their sweet-spot.

Industry

Fan boys abound. Omniture Fan Boys will point out a superior report builder and data slicing capabilities. Coremetrics Fan Boys will point out usability differences and a superior eCommerce support. Webtrend Fan Boys (you tireless people) will argue it's the best from a compliance standpoint for government. Google Analytics fan boys will scream 'game changer!'.

Google Analytics was always a reasonable alternative to Omniture and Coremetrics. (I did my first enterprise install of GA in 2008.) There was just a lot of discomfort within enterprise on support. It persisted. It drove preference. The last of those specific enterprise concerns are now addressed.

In effect, the incremental improvements are along known primary dimensions, and allows them to compete better.

You, The Customer

There's value if you're an enterprise manager.

For 95% of the user base, the release isn't aimed at you.

When firms compete, you win.

Towards Disruption

The Adobe Vision was that designers would value analytics, and that synergies would be realized through integration.

The Google Vision is to assist people in making better websites, better experiences, and better results through data. They believe in democratic data access.

The IBM Vision is sense making.

Each vision contains a bias about which dimensions are most important, and most likely to generate a sustainable competitive advantage. The vision is a grand hypothesis.

Which do you buy into?


Credit where due


Phil Mui, WAA Research Committee member and group product manager over at Google, has an excellent track record of both incremental feature improvement and some disruptive innovation. I congratulate him and his team for getting this out and launched.

Facebook GraphRank

A lot happened at the F8 developers conference last week, the most significant was changes to the Facebook GraphRank and the Social-Product Graph.

Instead of offering a single degree of freedom, to 'like' anything or remain silent, it will be possible for people to state (verb) + (noun) something. I have 'read' + 'this book'. I have 'watched' + 'dexter'. I have 'eaten' + 'breakfast'. And, I hope, I have 'bought' + 'this phone'. This goes to the notion of 'friction'.


Frictionless

The term 'frictionless' was used a lot at the conference and this has significance.

Friction is resistance to sharing information. It's caused by technology, experience interruptions, and by, yourself.

Let's start with you.

There's a pretty good model, put forward by Wendy Moe at the last INFORMS Marketing Science Conference, that people undergo two stages when interacting with social media. The first stage is considering if they're going to say anything at all. The second stage is how they'll moderate their opinion for the audience and their peers. A lot of evidence was put forward at the 2010 INFORMS MS Conference about the 'need for self-expressiveness' amongst individuals, and youth in particular. Your subjective opinion of 'over-sharing' depends a lot on your attitude towards self-expressiveness. Another way of saying that is sensitivity to privacy, or need for privacy.

Your perception of Mark Zuckerberg's language around 'expressing yourself' is likely heavily moderated by that very same attitude. Do you feel the need to post everything you do, all the time? Do you feel the need to update anything at all?

You cause friction. Technology causes it too.

The like label on the like button is such a technology. The verb 'like' forces an editorial or an endorsement, and, applying that Wendy Moe model, if somebody were to share an opinion, it's going to be moderated. You have only two options. Click like, and generate a social endorsement, or don't click like - which generates no social signal whatsoever.

Reporting facts, without the editorial, ought to increase the volume of sharing. They can certainly increase the volume of data volunteered by reducing the self-censoring, self-moderation step.

The addition of objective 'reporting the facts' verbs will contribute a large amount of data to the social graph. And that's not trivial. The volume of information about what somebody experiences will increase, and it will be organized through the graph.

Changes to the allow technology are another factor. The photo below caused Mark Zuckerberg's heart to sink. It's easy to see why. If Super Mario Bros was released during the era of Facebook, this is what you'd probably see. What a terrible experience.

The publish allow is common. It causes friction, and, it ruins many user experiences. It has its origins in privacy concerns. It's at this point that a 'utility' becomes an 'unwanted technical solution'.

Arguably, if you have an app from Netflix, or from RDIO, on Facebook, what you listen to will be added to your social-product graph seamlessly.


The Benefits To You, The User: Relevance

This information is arranged on a product-social graph. A graph is a mathematical representation of vertices and edges. You're a vertex, and you're connected to your friends and the things you like. You will also become related to the things you've read, games you've played, songs you listened to, and so on.

All of that information can be used to filter the newsfeed to surface relevant information. It should also enable data scientists to help you discover new music, better TV shows, and just in general have a better time with the long tail of content. The objective is better experiences.

You've been the beneficiary of such technology long before Facebook was ever created.

Consider how Google uses the web graph to generate relevant experiences, specifically, how to find content. Google's key insight was to add information to webpages based on how they are connected to one another. They use this information, among many other pieces of information, to make recommendations to you on search terms. Google returns hundreds of pages of results for some words. Most people click on one of the top 3 returns. People are generally satisfied with the results returned by Google. Google's algorithm is called PageRank, and it operates on Graph Data.

Facebook, too, uses an algorithm to optimize your newsfeed. It's called GraphRank, and it will operate on more complex information. In theory, the more you share through Facebook, the better GraphRank will become, personalized to you.

The aim is to compete on relevance.


Competing on Relevance

GraphRank will be taught through user control. Users will decide who their close friends are and who their acquaintances are, and list them as such. Even though most people 'don't make lists', many people will make lists if they figure they'll get a return on that effort. Users can easily decide to censor those that share too hard or share too much. It may very well be the case that there's no such thing as sharing too much, it may be the case of sharing not enough relevant information.

Brands, too, will have to compete more than ever on relevance. The degree of user control means they can slap a brand with censorship without 'unliking' them. The flip side is that never before has there been so much intelligence to understand an audience (non-PII). This linkage between analytical intelligence and communication strategy is better than ever.

It used to be that being boring was a good way to reduce risk. Being boring may actually be more risky. Nobody is going to be rewarded with endorsement and engagement by being boring. But what constitutes 'interesting', what constitutes relevance, varies by audience.

What constitutes 'interesting' can be observed directly from the product graph, especially with expanded linkages and data.


Privacy and Utility

When the product is free, you are the product.

Whenever I publish a web page, a blog posting, or put up a site - it is scanned and indexed by Google. Google helps people find my site. I am the product. But I get utility. I have the ability to suggest to Google not to scan my website (blogger excluded), however, I have to do something to opt-out, instead of doing something to opt-in. Of course, I run the other way. I use Google Analytics because it's worth the value exchange.

Whenever I publish an update to Facebook or exhibit a valuable behavior, it is scanned and indexed by Facebook. Facebook helps people find things I like, and developers who create new apps will help me discover new content. I am the product. But I get utility.

So long as Facebook educates people, and people educate themselves, about the privacy they are exchanging to Facebook in exchange for utility, we have a value exchange. There is consent from the user. There is consent from Facebook.

Users ought to have control over what they share, and what they don't. Utility will adjust as a result. Tradeoffs.


Conclusion

A lot happened at F8 to GraphRank.

There will certainly be a reduction in friction. There will most certainly be an uproar around privacy and informed consent. There will most certainly be an explosion of really awesome, useful, functionality forthcoming to Facebook.

Random Finds You

The longer you look for something, the more chances random has to find you.

To demonstrate:

I drew the blue lines in a pseudo-random way using Google Correlates awesome Search By Drawing function.

That is to say, it's impossible for anything that I draw to be truly random. I'm biological and my brain-hand coordination is nowhere near the same degree of randomness as background radiation.

However, I drew the lines without a conscious design. It wasn't governed by a model or a theory, like the one that powers the S-Curve.

I made 5 runs. Google returned 3 correlations, pictured below.













There are a few pretty good R's here - all exceeding 0.66.

Moderate correlations can be erroneous in very specific ways - missing small peaks or having slight phase variances. And it's possible that Google Correlate won't return anything with a correlation 0.6000 (the lowest I managed was 0.605.)

Google didn't return correlations for two attempts.

I drew those lines without a specific theory in mind. Yet, correlations emerged. I suppose, after the fact, you could construct some sort of theory to explain why a random doodle works. But it's not informed by a theory.

A lot of pure data discovery generates seemingly random findings, and there are methods to assess variance. It involves cutting your data into two, exploring your theory on the training set, and validating any emerging theory on your testing data set. That reduces the risk. It doesn't eliminate it altogether.

Such studies have been executed on massive health files to turn up correlations between astrological sign and breaking bones. In fact, in much the same way that buying thousands of lottery tickets increases your chances of winning, attacking a dataset millions of times with varying functions can turn up significant correlations that survive the testing data set! Over-training ahoy!

This doesn't invalidate exploration.

I'd argue that it's far less risky to go into a dataset with a theory / model in mind. The likelihood of getting caught in a random trap of your own imagination is less likely.