Accuracy
Web analytics systems are not very accurate. For the most part, this is because the systems measure visitor behavior in a relatively crude fashion. For example, few systems are capable of determining how long somebody spends on the last page of their visit. Time spent on the last page is never included in any visit duration. This means that every single number you have ever seen relating to visit duration was wrong -- if you regarded it as a measure of how long people spend on the site. If, on the other hand, you regarded this figure as no more than the time between the first and last page requests, you don't have a problem. But how many of us do that?
Wake turbulence is a particularly widespread problem. Many people leave a site by repeatedly clicking their back button to reverse their path. Thus, many visits end with a series of one- or two-second pageviews (called "wake turbulence"). Most analysis tools don't even recognize this problem, let alone deal with it. This increases the average number of pageviews per visitor, and reduces the average page duration. It may account for up to a third of all pageviews on most sites. This means the average number of pages per visit is being over-estimated. It also further reduces estimates of how long people are reading pages for.
In my view, most web analytics systems report an average visit duration that is half the reality. This near-universal underestimate has skewed web design, because designers have a fundamentally inaccurate understanding of how long people spend on web pages. This, in my view, is one of the reasons why we have such appalling conversion rates -- we are not designing sites for the way people actually use them.
People also resist being tracked by using cookie cutters and anonymous surfing tools, some of which make one person seem like many. At the same time, hacker systems, disguised as legitimate browsers, are spidering millions of web pages each second. I do not know of any web analytics system that seeks to filter these visits, or even acknowledges these issues, yet they may account for 10 percent of traffic on some sites.
So we really don't know how long people are spending on our websites and we are making broad guesses about their behavior, which are clumsy at best and wildly inaccurate at worst. Is it any wonder then that most sites consider
themselves lucky if they can manage to squeeze a sale out of one visitor in 50?
Consistency
Web analytics systems often have internal inconsistencies within their data. This means that numbers presented in one report may not match the same numbers in a different report inside the same system. Different programmers, most of whom have only a rudimentary understanding of web analytics, have written different reports.
I've built web analytics software from scratch. I've worked with teams of developers struggling with the intellectual challenges of how to encode the huge amount of incoming data in a way that makes it accessible and flexible for reporting systems. There are no best practices or customary solutions you can look up. Unless you've worked with database design, it is difficult to comprehend just how complex creating a database can be and how fundamentally the structure of that data can affect what you can do with it. Because web analytics software was not created by people with a background in web analytics, different portions of the code, written by different people, often process the same data in different ways and thus create different numbers for the same thing.
<< Previous page | Next page >>
