Segmentation and Revenue per Visitor
Monday, November 10, 2008
I don't know if everyone knows this -- I certainly didn't before last week, and neither did anyone here. I haven't been able to dig anything up on Google, but would love to read it if there is anything.
Basically, something you sometimes do in web analysis is say, "How did people who saw a particular page do in terms of money per visitor?" This is a common way of (supposedly) determining the influence of particular pages on success, and on determining whether a page is working. It turns out, except under controlled circumstances, this is a totally invalid question.
The way to think of it is, pick any page on a site -- revenue producing or not -- and think about all the people who came to your site. How likely is it that each person saw that page? Well, it depends on how many pages each person saw -- and that, in turn, is closely related to how much they spent. Those people who saw a lot of pages were more likely to have seen your page, and they were also more likely to have spent a lot of money.
In fact, logic says that the less busy a page is, the higher the revenue per visit will be for people who saw it, because statistically, the people most likely to see it will be those inclined to do a lot of clicking around. If it's hard to find, only people doing a lot of looking will find it, and those are the people who buy the most.
 And when you simulate, this hunch is confirmed. The graph to the right is for a website with 11 pages. Each page contains a product with the same price and the same likelihood of purchasing once you've seen it. They all have different likelihoods of being viewed, from 0.001 to 1. I ran 100K visits with varying initial interest in purchasing (applied as a multiplier to the viewing likelihood), and what you get is this curve relating revenue per visitor for the visitors to each page against the (unadjusted) likelihood of seeing that page.
To sum up: the less likely a page is to be seen, the higher value the visitors to that page will seem. But this is a figment of the probabilities involved, and the fact that high-value visits will fall into way more segments than low-value visits. It means that the value of every single segment you look at could be higher than the site-wide value per visit -- which seems like a paradox, until you get your head around this.
It also means that you really shouldn't take the performance of your shiny new page compared to the rest of your site as evidence of anything. Of course it's doing better than average: it's harder to find.
posted by henry |
11/10/2008 02:38:00 PM|
|
