Why Google owns your lunch: The "I'm feeling lucky" button

Google is a force of nature in global business today, in the same way that the Velociraptors ate everything in Jurassic Park, and Napoleon conquered everything non-frozen in Europe.

So I am bemused at Valleywags's credulous acceptance of Sergey Brin's statement that the "I'm Feeling Lucky" button costs Google $100+ million a year because they miss out on the advertising on the subsidiary pages when users click that button.

Yeah, right. Aren't you Valleywag guys supposed to be snarky?

News flash, Silicon Valley: Google is composed of some of the smartest machine-learning dudes to ever walk the planet. They have transcended their humble beginnings in web graph analysis to meet their true calling, which is to capture and analyze every single bloody user click, everywhere, ever, that they can get their hands on.

Within that context, if you think that the "I'm feeling lucky" button still exists because of some avatistic non-commercial motive within Google, get a clue. While a friendly grad-student impulse may have been the original source of the button, that button has long since proven its worth on a simple, highly commercial scale: query term type differentiation.

There are, broadly, three classes of queries on the Internet: discovery/information, navigation, and transaction. "Navigation" is of the type: "I need to know the URL for IBM, please take me there." A considerable amount of query-response and relevance angst is expended by most search engines on differentiating the three types of user intent, because users in each mode have vastly different expectations. Google, unlike others, gets at least half of this differentiation for free, because of that magic, friendly, 'Lucky' button.

Sergey, that's worth $110 million and more, and you know it.

That crucial differentiation makes Google search better; and better = more profitable. Scoreboard!

How and why does this work?

Because there is a strong predominance of searchers with navigational intent who click the 'Lucky' button. Users with informational (discovery) or transactional intent, on the other hand, tend to click the 'search' button. That distinction allows Google to build a dictionary which correlates a certain set of search terms with likely navigational intent -- for free, in machine learning terms. Voila! Google take a set of searchers with inherent low monetization potential -- after all, they only wanted to find their way to IBM's website -- and makes its insanely profitable search engine even better.

money...
...money...
...money

Sergey may pitch this as some charitable instinct on Google's part, but we know better.

Rock On, Google Borg!

Footnote: I talk a good game, but I only vaguely understand this stuff, and I am thrilled and humbled on a daily basis by the brilliance of the people who work with me and help me understand how much this sort of science and engineering matters. Thanks, Z Team!

Waiting for Virgin America...

I tried three times to do a flight search on Virgin America.



No dice.

JetBlue just got over $1000 of my business, because their web site works.

Web sites matter, people. Site performance and search performance are two of the most critical customer-service metrics you can measure, across an incredibly wide range of retail and commercial endeavors. They're hard to do well. Which is why they matter even more than you think -- because you can differentiate from your competitors, and delight your customers.

Higher-Performance Computing: Improving the Google File System

For those of you interested in high-performance computing, Doug Judd, our principal search architect at Zvents, has an awesome new post on the Zvents developers blog called "The Google File System - and How It Can Be Improved." Doug is leading our forthcoming Hypertable open-source project, which builds on top of distributed file systems like GFS, HDFS, and Kosmos, and he's got plenty of interesting thoughts on why GFS works well, and where it can be made better.