Beware How You Count Events!
I am aware that a large segment of the HIPPO community advocate visit based metrics, but honestly, one can get into trouble – big trouble – if one blindly adheres to “Visits are what count.” and apply visits like ketchup to every metric entree. In web analytics there are three basic tempos – page views, visits, and visitors (actions over multiple visits). Page views represent the raw vibrance of your web site – lots of visitor actions that are difficult at times to summarize and trend. So to gloss over the visitor churn and their propensity to reload or revisit pages, we often apply visits to metrics to put a wet blanket, metaphorically speaking, over the fire that raging in the page view traffic.
This is called serialization where we attempt to remove false or duplicate counts to report only the real counts. By selecting visit instead of instance counts or serializing events by session ids, we eliminate most of the false counts at sacrifice of some of the true counts. Though this approach helps stabilize metrics, it can often make the metrics insensitive to changes in business decisions and visitor responses to those actions.
Don’t Curb Your Enthusiasm!
For example, imagine I am a business that depends upon customers placing items into shopping carts, and checking out shopping carts to make purchases. If I only counted one shopping cart and one purchase per visit, clearly I would get different metrics than if I allowed my shopping carts and purchase metrics go free. If I had promotions that increased shopping cart activity and purchases, my visit normalized metrics would show a less enthusiastic view than what may be raging (under the visit blanket) on the actual web site.
OK – so this may seem contrived. Who would not count every purchase? On the other hand, that should be our attitude in everything we count. Why would I not want to count the event every time it occurs regardless if it happens multiple times within a visit or once at the end of many visits?
We Do It All the Time!
Another example, why do I want to count just entries that start a visit and not every time the visitor comes to the site from an external referrer? The latter can occur multiple times within a visit or not occur again till many visits (These are called “Return Visits” or “Repeat Visits” not to be confused with new and returning Visitors … duh!?!). I may be interested in how many visits include a sign-in, but why would I not also be interested how many times a visitor signs-in during a visit? You may assume only once, but does your data confirm that assumption?
Blame It On the Solution Designer
For the solution designer, the question is then “how can I tell if the event is real and not a reflection of visitor navigation churn?”. This involves establishing a “serialization” strategy in the client side instrumentation or for Omniture’s SiteCatalyst adding a serialization key to the event. Too often a session cookie or session-id is too readily available to apply to collection essentially making the count a visit based metric.
What is required is that the designer take the extra effort to understand the real event being counted, and expand his repretoire of implementation tools to either declare the event only when it is true or find and establish a more appropriate serialization key. For example, may be a transaction key for a service task would be more appropriate than a visit id.
The Pain Can Be Real!
A client of mine is going through the pains of having their Key Performance Indicators transition from visit based to transaction based metrics dramatically changing their view of the business but at the same time abruptly breaking from the historical data that has been collected for years. Like a blind person who has suddenly been given sight, there is an urge to pull out their eyes and return to their old “vision”.
Stay the Path: Real is Good!
It is too early to reap the benefits of the new vision, so we have to proceed assuming that with a vision that reflects more accurately visitor behavior, we will be able to more accurately measure effects of our business decisions. Whether it is marketing campaigns, changes in web design, or attempting to understand what the customer is saying (VoC), without true visitor based measures that are sensitive to visitor based actions, it is impossible to optimize performance. The visit based metrics were like a wet blanket on the performance measures we needed to access these changes.
I’m No Einstein but …
All I am saying is perhaps a little more thought should be taken in setting up and serializing events. Quick and non reflective solutions can truly lead to pain down the road. Though it is our desire as analyst to make the trends evident and insights pop out from simple reports, we must still keep in mind as Einstein would say: “The solution should be simple but not too simple.”