The problem with that theory is that sophisticated analysis can only yield helpful information if the original data is accurate. If the original data is flawed, it?s obvious that any resultant analysis will magnify those flaws.
How are Web stats flawed? Let's say that an E-Commerce execs wants to know the products that the site's audience is most attracted to, on the theory that the retailer will order more of those. The problem is that the placement of those items?and the length of time they are permitted on the site's most prominent pages?will likely have a huge impact.
It's almost impossible to give every product the optimal page placement. If the site decides to juggle the placement, then we have to factor in questions such as "How many hours did each product stay on the best spot?" and "What were those hours? Did we factor in that from 2-3 AM is worth a lot less than Noon to 1 PM?"
After that, there comes the question of how accurate the initial data was. Any company that has switched Web analytic packages or, better yet, runs more than one at the same time on the same site, knows how varied their results can be. Even this assumes that the people analyzing the data know what they're doing, that they're differentiating number of visitors from number of visits and clicks from pageviews. (Let's be nice and assume they do know what they're doing.)
Two recent stories bring this issue into focus. The Wall Street Journal on Thursday reported very real questions about traffic claims about some popular Internet videos. Take a peek at that story and then accept the frightening reality that the Web traffic metrics you're using for every projection are probably not a heck of a lot more accurate.
A somewhat different example speaks to understanding the source of major traffic. One of the most bizarre Web site stories in recent months comes to us from the Conan O'Brien Show and a story in The International Herald-Tribune did it the most justice. Here's the best part: O'Brien did a sketch about overlooked sports mascots, including a Webcam manatee.
"At the end of the skit, in a line that O'Brien insists was ad-libbed, he mentioned that the voyeur was watching www.hornymanatee.com. There was only one problem: as of the taping of that show, which concluded at 6:30 p.m. Eastern time, no such site existed," said the story, "which presented an immediate quandary for NBC television: if a viewer were somehow to acquire the license to use that Internet domain name and then put something inappropriate on the site, the network could conceivably be held liable for appearing to promote it. In a pre-emptive strike inspired as much by the regulations of the Federal Communications Commission as by the laws of comedy, NBC bought the license to hornymanatee.com, for $159, after the taping of the show but before it was broadcast."
Before we get into the Web traffic implications of this story, let's just sit in silent awe of a network that can be so Web-savvy and psychotically paranoid at the same time.
Within a week, NBC claims, the site has received some 3 million hits and thousands of videos have been submitted. (Note: If NBC can create a site from scratch, create the content, test the programming and have it approved in a few days, why can't E-Commerce companies that supposedly are trained to do this? But, as is my nature, I digress.)
Where did those three millions hits come from? The people who happened to be watching that morning at 1 AM? Time-shifting Tivo users? People who read stories about the incident? Or, most likely, is this a relatively smaller number of people who are hitting the site repeatedly?
For NBC's purposes, they don't really have a mission-critical need to know that. But E-Commerce site's do, whether it's product purchases, inventory projections or where to focus marketing dollars to generate more traffic, the assumption that their Web stat analysis is accurate shapes tons of online decisions.
I'm not suggesting that there's a better alternative today, but I am advising extreme caution when someone places too much faith in Web analytics. In short, Web stats rarely mean what they appear to mean: unless it happens to support your position.