Web Analytics - Puzzle or Mystery - the article

Recently I came across an old article from a popular Web Analytics blog, Occam's Razor that is as applicable today as when it was published in 2007.

In 2007, Malcolm Gladwell published an article in The New Yorker expanding on the concept of 'mysteries' and 'puzzles' as originated by Gregory Treverton, national-security expert. From Treverton's definitions:

• Puzzles - have a correct, factual answer which is unknown because we do not have enough information.
  
• Mysteries - do not have a single, correct answer. We have a lot of information and require knowledge and judgment to make an assessment of the answer.
  

I find this a fascinating way to look at the unknown. If you think your question is a puzzle, then look for the information to solve the puzzle. If you think your question is a mystery, then look for people with the knowledge and judgment to make sense of the information you have.

The poster from the Web Analytics blog argues that web analytics is a mystery and I would largely agree. In my experience, there are few puzzles in web analytics, mostly around data collection and data definitions, although many data definitions are mysteries. Interpreting and acting on the resulting data are most certainly mysteries. Yet, I've come across many people, especially those new to web analytics, or even outside it, who think it will be a puzzle - "If we could only report on 'X', we'd know what to do". I think that explaining web analytics as a mystery instead of a puzzle would go a long way towards helping people understand web analytics.

Golf Statistics Article

I grew up in a golfing family, so I've been playing and watching golf since I was a little kid.

For most of that time, I've been tracking my golf outings. Always score, of course, but often putts, penalties, sand shots, clubs used, and whether a shot felt good. This season, I have been tracking the distance each shot goes using a gps device, by club, hole, and lie.

The data isn't very useful, given the big number of variables in a round of golf, and the small number of rounds I play each year. But, it has been interesting to look at, and has helped me finetune my play in small ways. (as an example, earlier in the season I was having a problem with three-putts. After looking through some rounds, I realized my medium and short length putts seemed fine, but my long putts were consistently leaving me with medium length putts instead of short putts.).

So, I was extremely excited to see an article in Slate about golf statistics!

From the article, I learned that the PGA employs ShotLink to record each shot made during its tournaments. With that wealth of information, there are many applications. For example, individual players can determine how they rank in types of shots compared to their peers to determine where to focus their practice efforts. They have also determined that if the hole were twice the size, poor putters would benefit more than good putters.

Mark Broadie, a golf researcher at Columbia, presented at the 2008 World Scientific Congress of Golf. A related paper, with much more detailed analyses is located here.

I would love to get my hands on the ShotLink data - there are so many interesting questions waiting to be answered!

Facebook Apps Analytics via Omniture

Yesterday, Omniture "announced App Measurement for Facebook, a new solution that enables Omniture SiteCatalyst customers to measure the popularity and success of Facebook applications." (from the Press Release) (screenshots)

In March, Omniture announced Twitter analytics via its SiteCatalyst tool. (Press Release. I wrote a quick blurb here).

As a strong believer in the power of data and analytics, seeing a company with an established following making headway in the social media space is exciting.

I.B.M. Unveils Real-Time Software to Find Trends in Vast Data Sets - the Article

There was an interesting article in yesterday's Wall Street Journal about a new IBM software, to track and analyze data in real-time.

This sounds similar to a technology I have used in the past, although IBM seems to have built out the system to learn and find correlations in the data without guidance.

Real-time data analysis is often expensive compared to near-real-time or offline processing. Additionally, with complicated data, it would be a sophisticated and well-designed system to anticipate and correctly handle all data. So, a system that is self-learning would be advantageous.

I would be very interested in finding out more about this technology and its related product, System S. I was not able to find much information online, unfortunately.

Differentiating Online Ads

Recently, Google started offering behavioral ads in addition to its contextual ads. There was an interesting article in the NYTimes blog section this morning about it.

Google allows you to opt out of its behavioral ads, and also provides generalized information on why you're getting specific ads. The author of the blog suggests going further, to display exactly why the ad is displayed and why it's displayed the way it is. For example, whether the image, text, or price displayed are variable, and if so, exactly what information about you was used to determine its value.

What I found most interesting about the article was a quote from a Google representative which suggests that few consumers understand the difference between contextual and behavioral ads.

It's simple (or it can be, at least). Contextual targeting uses information about what you are currently doing. For example, search for 'Prius repair' and you'll get ads about local repair shops. Search 'Prius 2009', and you'll get ads about new Priuses. Or in an offline sense, an ad for sugary kids cereals during Saturday morning cartoons.

Behavioral targeting uses information about what you have been doing. For example, if you've been browsing for new Priuses, you may see ads for cars even when you're browsing a news site. In an offline sense, this could be like if you go to the same waitress every Saturday morning, and one week she offers you a coupon for being a loyal customer.

Of course, there's a difference between online and offline. Most people fully understand, or *could* understand offline collection methods. But 'online' is a black-box to many. I think the author's idea is an interesting one, if unlikely to come to fruition any time soon. I would probably use it, but most wouldn't. I think the most benefecial aspect would be one mentioned in the article, that companies would behave more ethically or risk public press because of watchdog agencies and the media having access to the information.

Twitter Analytics via Omniture

In addition to its usual web analytics, Omniture has started offering Twitter analytics, which I think will be a powerful addition to a company's business intelligence.

The Twitter analytics space is young, but with an experienced brand like Omniture joining the fray, I expect interesting things.