One of the most fundamental metric in web analytics – Visits – has no smartness or even customizability. Conceptually a ‘visit’ should represent a user’s single ‘interaction’ with the website. With multiple sites open in tabs side-by-side, very often people could remain inactive for more than the magic number of 30 minutes despite it being the same ‘interaction’.
Example 1: When I plan for holidays, I search by myself (after lunch), show it to my wife (in evening) and then book a flight and a hotel (next afternoon in office).
Example 2: I often search some term on google, open the first 6-7 results, spend time reading tab after tab. By the time I reach the last tab (or even the second tab if I started chatting with a friend in between) my ‘visit’ is over.
Problem in example 1 might not be too difficult to solve – you can allow sites to define their inactivity period to be longer or shorter than 30 minutes for visit to be over so that they can choose as per the nature of the site.
Problem in example 2 is lot more difficult and needs some smart implementation to define a visit (or call it interaction). It can be based on couple of observations 1. The new visit has the site itself as referrer (as the site was never left) – but this ends up getting counted as direct. (eg. My session expired and I clicked on homepage logo to restart browsing) 2. The time difference between the new visit and previous visit is much lesser than the average time differences between all visits from that visitor so far. (eg. Two visits within same day would be in same interaction, if I visit that site once in 3 months)
For a proof of why this is a huge problem, check your ‘repeat visit frequency’. If a majority of the repeat visit frequencies lies within <1 day and 1-2 days, it is obvious that most of those visits are same interaction (unless you are a news site or social networking site of course )
Sorry for this longish post. Please reach me on firstname.lastname@example.org or gyanalyst@twitter for any discussion/explanation