NONE: ONLINE-ADS>> counting impressions and click-throughs
ONLINE-ADS>> counting impressions and click-throughs
Cliff Kurtzman (cliff.kurtzman_at_tenagra.com)
Wed, 12 Feb 1997 02:29:24 -0600
This post explores some very technical aspects of counting ad impressions
and click throughs. The purpose is to illustrate some of the difficulties
and ambiguities in measuring page impressions, banner views and
click-throughs. Small changes in assumptions and counting methodologies
can have big effects on the numbers you come up with and report to
advertisers (or are reported by advertisers).
I'm using real data collected on one of our sites over a 1-day period
(January 30th, 1997).
Here is the scenario. We are running a fixed banner ad on a web page on
one of our sites. The banner is hyperlinked to the advertiser's site. In
conjunction with the banner is also a text link to the advertiser's site.
(Actually, each time someone clicks on either the banner or the text link,
it registers a hit on a URL on our server, which then redirects to the
advertiser's web site.)
The page with the banner is an HTML page, and is not generated by a cgi.
(Pages generated by cgi's are not subject to caching). This particular
page has a META HTTP-EQUIV="Expires" tag on it to perhaps reduce the
likelihood that systems will cache it and hence give our server a more
accurate access count. I don't know how many systems and browsers actually
recognize this tag. In any event, the bottom line is that caching causes
the true number of accesses to our page to be under-reported, but I suspect
not by more than 10%.
Over a several day period we noticed a jump in click-throughs on the ad to
about double the normal level even though the the page traffic was pretty
much constant. This caused us to examine what was happening in detail, and
to turn up some very interesting results.
Our log file showed that the page on which the ad resided was loaded 1603
times. This includes:
992 full loadings of the web page (the server logged transfer of the full
11053 bytes)
16 aborted (partial) loadings of the page (either 0 or 8192 bytes were
logged as transferred)
595 page accesses that show up as "304 0" in our log file -- indicating
that the page was unchanged since the last time this visitor came to the
site and the page was therefore loaded from the visitor's local cache of
their gateway's cache.
There were 925 unique hosts hitting this page over the course of the day.
A session analysis on the home page hits showed that the 1603 home page
accesses were logged by a total of 1026 sessions, with a session being
defined as hits from the same IP address within a 20 minute period. The
number of sessions (1026) is close and slightly greater than the number of
full page loadings (992) which is in turn close to but slightly greater
than the number of unique hosts (925). This is as was expected.
Our log file also showed 868 banner loadings (694 full loadings of the
banner, 172 "304 0" loadings from the visitor's local cache, 2 aborted
loadings). This 868/1603 ratio is a fairly typical ratio of page loadings
to banner loadings. There are a lot of things that I think might account
for the discrepancy:
1) The visitor's browser finds the page is unchanged, logs a "304 0" in
our access log file, and then grabs the page and banner from the visitor's
local cache without the need to register a hit on the banner on our server.
2) The visitor is going through a gateway (e.g., AOL, WebTV) that provides
file caching. When the page is requested, the gateway queries our server,
finds the page is unchanged, logs a "304 0" in our access log file, and
then grabs the page and banner from its cache without the need to register
a hit on the banner on our server.
3) The visitor is accessing the page through a system (e.g., AOL) that has
not cached our page but has cached the banner.
4) The visitor clicks the link to my page several times before waiting
long enough for the page to load. This shows up in the log as several page
accesses but only one banner access.
5) The visitor accesses the page with the graphics turned off.
6) After linking to the page, the visitor hits the stop button on their
browser before the banner has had a chance to load.
Except for cases 4 & 6, the visitor has the opportunity to be exposed to
the banner or the text link with each page access, provided that they view
and read the part of the page with the banner.
So the first question this leads to is: How many ad exposures did this
advertiser receive? My feeling is that it is pretty much the full 1603
exposures, although some portion (perhaps about 580) of the banner
exposures were repeat exposures to people that had seen the banner already.
Without knowing the true magnitude of #2 above, or tracking cookies, it is
hard to precisely quantify the true number of repeat exposures.
I've had advertisers approach me that want their ad banner to load from
their server so they can independently track the number of impressions they
are getting. However, as described in the previous paragraph, it is my
opinion that counting the number of times the graphic loads undercounts the
number of true impression of their ad significantly.
This source of ambiguity can probably only be eliminated entirely by
serving the page and ad banner through a cgi script.
Now on to the click-through analysis, where things really get weird.
128 banner clicks were logged by our systems as hits on the URL that
generates a redirect to the advertiser's web site. As I noted earlier,
this is about double the normal rate for this banner. A look at the log
file indicated that 60 of these click-throughs came from a brower indicated
as NetPIM and nearly all of these hits from this browser came from hosts
with seattle.wa.ms.uu.net in their host name. This same browser caused a
similar number of hits to other pages on the site.
These hits fully account for the discrepancy in number of click-throughs.
I'm guessing this is some kind of automated downloader. It did not
register as a robot by checking our robots.txt file.
I also noted that another 21 of the registered banner clicks came from host
IPs that had previously clicked on the banner that day. In most cases,
this appeared to be people that were clicking on the banner multiple times
impatient for the referred page to load.
Taking out these 60 + 21 clicks results in concluding that this banner
resulted in sending about 47 real visitors to the advertiser's web site.
It is possible that there were some other robots that caused click-throughs
too, but l checked the browser field in every apparent click and I doubt
there were very many others caused by robots, spiders, or automated
downloaders.
Now we can have some fun with the numbers.
If I count all 128 banner clicks resulting from the 868 banner impresssions
to calculate click-through, then this advertiser had a 15% click-through
rate.
On the other hand, if we count the corrected estimate of 47 real visitors
we sent to the advertiser's site from the 1603 page impressions served,
then this advertiser had a 2.9% click-through rate. A major difference.
If you are buying ads and gauging their effectiveness on different sites
based on a click-through rate reported to you, then you need to be asking a
lot of questions to be sure you are comparing apples to apples.
Automated robots or downloaders visiting this web site and generating 81
false visits on the web page with the ad would cause only minor error
compared to the total number of page impressions served. If you have 81
false impressions of your banner a day out of 1600 at a couple of cents a
banner impression then it is no big deal.
But if those 81 automated visits generate 81 false click-throughs, then
that is not a minor error, because there are so many fewer click-throughs
than page impressions. If 81 out of 168 click-throughs are false and you
are paying for each click, then more than 1/2 of what you are paying is not
for real eyeballs visiting to your site.
This may be an extreme example, but I believe this shows that robots and
automated downloaders make the concept of charging an advertiser based on
click-through rather unworkable from a technical perspective (not to
mention not making too much sense from the point of view of the ad seller).
--Cliff
Cliff Kurtzman
President
The Tenagra Corporation
Old URL/New Web Site: http://www.tenagra.com/
281/480-6300
Internet marketing, public relations, consulting and web site design
-------------------------------------------------------------------------
This week's Online Advertising Discussion List sponsors:
WEB ADVERTISING '97: How to Bring the World to Your Web Site
Make your web ads *work*
http://www.thunderlizard.com/ola.html 800/221-2806
---------------
Q: How can I maximize my Web site's revenues from banner advertising?
A: Don't carry unsold, excess ad space. Sell out every impression.
adbot: the auction market for internet advertising (tm) http://www.adbot.com/
-------------------------------------------------------------------------
Online Advertising Discussion List To Unsubscribe send UNSUBSCRIBE
http://www.o-a.com/ to online-ads-request_at_o-a.com
