I was thinking about Gary Angel's last post on Evaluating Internal Search Performance; he mentioned that certain web analytics packages will tell you what keyphrases are searched on "after" an initial internal search:

"..Surprisingly, one of the most interesting types of Search Term analysis is what happens after a Search! Some web analytics packages will do this analysis for you – providing an affinity report of Search Terms searched AFTER any given term. It’s a very cool idea. This analysis can help you with several key search optimization tasks. It can help you identify Search Terms where customization of the output may be most desirable; it can help you understand possible customer market baskets; and it can be used to help identify weaknesses in the Search Results returned by your Internal Search Engine."

Unless Coremetrics does this kind of reporting (and I have not been working with a fully functional version of Coremetrics yet, at IBM - but hopefully soon will be) - then I have not been working with an analytics package that does this kind of after keyword analysis.

But that's not exactly true, because Microsoft AdLabs Research offers such a tool - it's called Search Funnels, but it's not based on internal search - it's coming off of Microsoft Live.

I just came back Paris, France a couple of days ago and if I did an outgoing funnel on "Paris" it looks like Paris Hilton would be expected as something connected with Paris (which is kinda obvious, but the wrong context in this case) - something with the Eiffel Tower (or the Paris Riots - which is again, the wrong context) might be expected - by the searcher.

It's as if a keyphrase generates an expected complement - just as a color yellow would generate …perhaps it's opposite - purple, or a sound generates it's complement (I don't know music - I bet Gary Angel could figure out the music part better than I).

Let's take it up another step, keeping in mind that adCenter Labs Search Funnel is not as richly populated, strange as that seems, as your internal search engine - but still extremely useful for generating what searches may be expecting to find when they use a search term (that's the easiest way to explain this).

Digital Cameras seems to suggest that the landing page(s) have links to Best Buy, WalMart, EBay or Circuit City on the landing page (probably targeted to the specific sections of each site).  To me, that suggests that people, or, the Search Engine itself, has formed an association with "digital cameras" and large electronics and computer outlet brands - and not specific camera brands like Nikon, Canon, etc - and that's important to know, especially if your generating dynamic landing pages using Paid Search - because that's how you leverage search.

And yet, if you search for "digital camcorder" you get something slightly different:

With Best Buy seen as a clear leader in having digital camcorders - but once people get to Best Buy - they expect to see "Sears"…..Weird….or maybe..inciteful - at least, that's how I look at it.

A search on my favorite painter, Paul Cezanne, generates these expectations, if Microsoft Live is to be believed:

So, people who would search on Paul Cezanne probably want to see his biography, or look at other artists who were contemporaries like Van Gogh or Renoir (Dali was not really a contemporary, having risen to fame after Cezanne died).

However, if people just search on the last name of Cezanne - a different expectation is being expressed:

People who searched n "Cezanne" are comparing him to other great masters who emulated him such as Picasso and Matisse - see how subtle this thing is!

If people really want to tackle why their paid and organic search conversion rates are so bloody lousy - consider what landing pages your site search is matching up with the keyphrases being entered and if those pages actually have what people are expecting to see.

I think a good argument can be made for dynamically generating landing pages for both Paid Search (AdWords and Panama to start) and internal search (as well) because … with intelligence on what is "expected" you can more likely answer the needs of your searchers - and that's what I got out of Gary Angel's post on Evaluating Internal Search Performance.

However, building customer segments based on keyword analysis might be difficult to pull off unless you start off with some preconception of what kind of segmentation your likely to have or need beforehand - as search term segmentation can be very complex - with almost infinite granularity. 

Just as artists make often make a decision about what is going to be the foreground, middleground and the background (and what coloration makes makes up all three) the analyst has to decide what the limits of segmentation are before they start segmenting -or else they will be overwhelmed with too many segments - which is simply not workable - creating more confusion without solving the fundamental problem of conversion rate. 

Here's what Gary Angel has to say about that:

"..When this isn’t an out of the box report, you can replicate it by building a visitor segment based on using a particular keyword and then studying the other search terms used. Note that while this analysis is useful for most of the tasks mentioned above, it doesn’t necessarily capture any time sequence. It’s possible that the associated keywords were used FIRST!"

Like anything else, the quality of results you get with Web Analytics tools depends largely on the insight of the Web Analyst using the tools.