| |
Lesson 8 –
Examining Subsets of Traffic |
|
|
Filtering Visits
While web site analysis reports contain various types of data, they
are all derived from a simple set of basic data elements that exist in your log
files. When applying filters, you filter one of these basic data types:
Referrers (that start visits), Browsers (or other user agents), Requested page
or file, Authenticated Users (if your site requires a login), Host (including
domains, and country of origin of the visitor), or a Cookie that the visitor
carries. Each of these data types is discussed in detail here.
Filtering by Referrer
|
|
Referrer filtering can be used to see where traffic from a particular search
engine, advertisement or other link goes on your site. When you set up a
filter for a specific visit initiating referrer (or set of such),
you can use all the path analysis and related reports covered in Lesson 7 - Determining Visitor Behavior Patterns to
analyze the traffic patterns of visitors from particular leads.
|
| Figure 1. This sample Referrer Report shows a
recent increase from a widgetnews.net story. |
|
Figure 2. From the Exit Point report it
appears these visitors are not likely buyers. |
Referrer filtering can also be used as an investigatory tool. For example,
perhaps the marketing manager of our widgetmanager.com site sees a peak in
traffic one day and, using the Referrer Report,
finds it comes from a particular news story at widgetnews.net (see Figure 1).
The manager wants to know whether they should place an advertisement on
widgetnews.net to attract more visitors on a regular basis. She sets up a filter
to look at just the traffic generated by that story. By looking at the Exit Point report she realizes that the significant
majority of those visitors ended at a links resource page on widgetmanager.com
that provide links to widget manufacturers (Figure 2.) As widgetmanager.com
sells a tool to manage widgets (not widgets themselves) she does not think this
will be a lucrative lead source – these visitors are likely first-time
buyers and not yet in the market for a widget manager.
Filtering by Browser
Browsers (or user agents) can be filtered to analyze traffic from particular
types of users based on the software or hardware they used. As previously
mentioned, looking at traffic patterns of wireless users can help you decide
whether to launch a wireless-specific version of your site and where to focus the
efforts in that design or to improve the effectiveness of a wireless interface
to your web site. Any user agent can become an important subset of your traffic.
Widgetmanager.com distributes a custom software tool that customers can run on
their own computers to help in manging their widgets. The program connects to
the widgetmanager.com site for registration, to check for updated versions of
itself and to download and install updates when they are available.
|
Figure 3. A filtered Page Requests report can
highlight usage of custom user agents. |
By configuring a filter to look at just the traffic generated by this
software program (a particular user agent), the developers can look at the
number of hits to ‘register.html,’ which happens once for each copy,
and ‘update_check.html,’ which is requested each time the software
is run, to see how often, on average, each copy of the software has been run. In
Figure 3, the number of update checks, 17,632, divided by the number of
registrations, 4,408, gives an average of four runs per copy sold. The CFO can
also compare the registration hits to the sales records, according to the
financials, and see if there have been a significant number of pirated copies or
other inconsistencies in the registration system.
Request or Content Filters
|
|
|
When building a request filter, you often use a wildcard
pattern to find details of requests to specific sections of the site, say a
directory or a file type. This sort of content filtering is similar to the
Groups capability in Summary that we discussed in Lesson
5 - Revenue Modeling. Selection of content depends a great deal on the
organization of your site. For example, if each section of your site,
‘products,’ ‘services,’ and ‘company info,’
corresponds to directories in your URLs, then you can filter on these
directories to cover each particular zone. This kind of zone filtering
can also be used to match content with departmental divisions withing your
company, e.g. news, sales, marketing, etc.
|
Figure 4. A filtered Other Requests
report can show users’ CGI actions. |
Another approach is to filter by content type, based on the extension of each
file name. You can look at requests to all *.cgi (or similar
extension) files to see how visitors use the dynamic parts of your web site. If
you have a particular CGI script for each action that a visitor can perform,
such as Figure 4, you can use the Other Requests
report to see what actions your visitors are taking on the site.
|
|
|
You can even use content filtering to match specific reports to specific
individuals within your company. If your developers and designers are given
particular domains of responsibility, you can make a filter to gather reports
for each showing the traffic for his or her area of control. We talk more about
making reports for particular viewers in Appendix A -
Making Reports More Usable.
Filtering Authenticated Users
|
Figure 5. Widgetmanager.com’s
Authorized User report shows
excessive traffic from two accounts. |
If your web site has a login-required section, you can filter by authenticated
user name (using “?*” to match any non-blank entry) to see how
members travel around the site. If your user names include some organizational
data, you can use wildcards to collect traffic reports of particular groups of
users, based on their user name. You can also use a user filter for
investigative analysis. The widgetmanager.com site requires a login to purchase
and download the widget manager software tool, submit bug reports or get
technical support. The NOC Manager notices an excessive number of hits from the
User Report (Figure 5) by two particular users, ‘sudo’ and
‘ptolemy.’ By creating a filter for only these two users the NOC
Manager notices that these “users” have downloaded a large number of
copies of the software tool. He suspects that the user names and passwords have
been shared or leaked to public lists where people download pirated software.
The CFO confirms that the registration count on the web site exceeds the actual
revenue. So the NOC Manager removes the two offending accounts from the system
and sends a note to customer support to contact the original registrants (if
they provided legitimate contact data) and discuss the issue with those
individuals.
Host Filtering
|
|
|
Host filtering, like request filtering, is usually done with wildcard
patterns. The only time you would want to look at a single host would be to
examine unusual access patterns, generally of troublemakers. This kind of
investigative analysis is covered in detail in Lesson 12
- Investigating Troublemakers. Host information includes domain and country
information, so you can find patters of specific users from particular
companies, Internet access providers or countries (with some limitation –
many international ISP’s use .com or .net domains now
rather than country specific ones like .uk or .jp.) A
common practice is look at just the traffic from *.aol.com
hosts. This tells you details about how AOL users use your site. If you have a
section of your site that is dependent on visual information, this can be
especially relevant. AOL will “compress” graphics from your site
beyond the compression you have already applied, reducing the quality of the
images that visitors see. If your AOL reports show that AOL users are
frequenting your product images, for example, you may want to add a note to the
site instructing them how to disable this “compression” so they see
your product shots as they were meant to be seen.
Filtering by Cookie
|
|
|
Figure 6. Cookies tracking promotion responders
show interesting patterns in repeat traffic. |
Cookies can be very useful when you want to track user information, especially
across sessions. By adding cookie filters, you can see traffic patters of almost
any subset of users you choose to follow. Let us assume widgetmanager.com runs a
Christmas special offering 10% off one of their products. They have a special
web page that they send users to who click on one of their web advertisements or
that users can type in when they see the ad in print media (this kind of link
tracking is also covered in Lesson 4 - Advertising.)
Every user who goes to that page gets a cookie that tags them as having
responded to the promotion. Now the sales manager sets up a filter showing
traffic patterns of visitors who have the promotion cookie. Looking at the
Weekly Report for the last six months (Figure 6) she discovers that while the
majority of hits from the promotion was the week before Christmas, there was
also a significant spike in traffic for the week including the 14th of February.
Apparently this particular widget manager product is not just a popular
Christmas gift, but also a romantic Valentine’s gift. It looks like widget
managers (at least for existing customers) seem to signify true love! This
information allows her to capitalize on this pattern the following year, by
sending a Valentine’s promotion to customers who purchased the Christmas
special.
Multiple Filters
|
Figure 7. Without enough filters, unimportant data
can obscure the important details in reports. |
You can apply more than one type of filter at a time. In the example on
browser filtering, we suggested adding a filter to track traffic from the widget
manager software tool. If widgetmanager.com releases frequent software updates,
it is possible that there are a large number of requests for each update and
they fill up the reports, making it hard to find the registration page hit count
(see Figure 7.) By adding a request filter to match just the registration page
and the update check page, the site manager can "clean up" the report to just
show the information that answers the original questions raised by the
developers and CFO.
|
|
|
|
Filtering by Referrer