Website Optimizer is great, but Website Optimizer with e-commerce data would be better. In general, increasing conversion rates is what we are trying to do, but it has to be in context, right baby? For online retailers, a good context would be revenue. I want to know that a wining combination according to Website Optimizer actually makes money. Perhaps a "winning" combination appeals to fewer visitors, but these visitors have a higher average order size. So, a better conversion rate may not always be in line with financial goals (do I sound too serious?).
Website Optimizer will show the same combination to a visitor
for up to two years. That's ok, but lets say I see combination 1 today, don't convert, come back a full year later (and my cookies are still there), don't see the test page and then convert. Website Optimizer will credit the conversion to combination 1. Is this justified?
Perhaps, but I will just say that latent sales
and revenue attribution
are tough nuts to crack and that there may be no one right approach.
In other words Website Optimizer uses visitors
as the basis whereas web analytics packages in general (Yahoo! Web Analytics and Google Analytics) use visits
. (side note: the Website Optimizer reports show Conv/Visitors, but not according to this help file
. Maybe just an outdated file).
So, 2 years vs 30 minutes. Where am I going with this?
Well, simply put I still like visits
as a basis for calculating conversion rates (as do others
We started using Event Tracking
to display Website Optimizer experiments in Google Analytics.
Events are tied to 30-minute visits, so we are no longer using visitors as the basis, but the more I think about it the more convinced I am that it is a valid approach. The advantage of event tracking is that it is independent of pageviews, so no need to worry about fake pageviews. The only thing to pay attention to is bounce rate
, so we just send events to a separate profile.
This is what my experiment looks like in Website Optimizer:
(Click for larger)
The same experiment in GA:
(Click for larger)
I think I can work with this. I am not sure if GA and GWO data is updated at the same time, so this would be an obvious source of discrepancy. As you'd expect visits would be higher than visitors over time, but otherwise the numbers seem to be consistent and both combinations get the same number of visits. The conversion rates are also consistent.
This example also exemplifies my point about taking into account revenue. Combination 1 has a lower conversion rate, but a higher average value. At the time of this writing no winning combination was declared.
In summary, I think it is a valid approach to analyze website experiments in terms of visits and that Event Tracking is a great vehicle to do that. Agree/Disagree?
P.S. We use a JS library to make event tracking calls a bit easier. We'd be happy to distribute it if there is any interest.