I was in Seattle last week for Jill Whalen’s “High Rankings Seminar,” which was a fantastic experience. I lived in Seattle for seven years back in the ’90s, and got to experience the early days of the Internet from a vantage point very close to the University of Washington.
One of the cool things that the UW does is broadcast lectures and presentations on their own “UWTV” cable channel, which I hadn’t seen since 1997 but was the first thing I looked for on TV when I got into my hotel room. Yep, it’s still going, and as luck would have it I was able to catch most of a presentation by the UW’s Daniel Weld on adaptive user interfaces.
Okay, that’s a bit geeky, I admit, and you may be having a hard time seeing the connection to search engine marketing.
Well, one of the really interesting things Weld talked about was others’ work on using relational Markov models to predict user behavior on web sites. This isn’t really brand new (citations go back to 1999, I believe) but it was the first time I’d seen anyone connect the dots.
What’s interesting to me about this work is that I’ve spent a lot of time over the past year analyzing the performance of three different types of landing pages for PPC campaigns:
[list=1]
[*]The site’s home page or a category page.
[*]A dedicated landing page for specific search terms.
[*]An “ask” page, which offers the user multiple choices in an attempt to segment the audience better. If you want x, click here, if you want y, go there, etc.
[/list]
Weld showed a very interesting graph, showing a similar problem, where designers of ‘smart’ user interfaces have to anticipate whether it’s best to deliver the application’s default behavior, take some automated action, or prompt the user for a choice.
The effectiveness of each possible decision, as with PPC campaign landing pages, depends on the degree of certainty we have about the user’s intent. When we have a high level of confidence about what the user wants, an application can take actions automatically, or a website can serve up a dedicated landing page. When there is very little certainty, the best choice is to deliver the default behavior, or perhaps the site’s home page.
In between, where we have a bit of an idea about what the user wants, the best choice in application design is to prompt the user for a decision. Similarly, I have found that our best results for many search terms will come from using an “ask” page to let the visitor identify themselves.
What I got from Weld’s presentation was a possible method to help us better anticipate what visitors to a specific page will want, based on where they’ve been.
A bit of a side trip, but I hope that some of you will find it interesting.
Dan Thies
