Re: online targeting

From: Clint Ballard <clint_at_accelerationsw.com>
Date: Tue, 2 Feb 1999 06:02:24 -0600 (CST)

TIM LEE WROTE:
> Impressions CPM CTR* CPV** CPS#
>Targeted 40,000 $25 0.25% $9.80 $143
>Untargeted 200,000 $5 0.46% $1.18 $500

>The results show that targeted banners have higher price,
>lower click through rate, higher cost per visit but lower
>cost per sale.
...
>More interestingly, this set of comparisons provided several
>unexpected results that need more detailed research before
>we make any conclusion. To understand the targeting
>mechanism, we need to understand: 1. Why targeted banners
>generated lower click through rate? 2. why targeted visitors
>have lower click through rate but higher conversion rate?
>
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Targeted had a 6.86% conversion rate and untargeted had
0.22% conversion rate. The answer to Tim's question 2. above
is because the targeted visitors had over 30 times the
conversion rate, showing that the premium paid for the
targeting was well worth the money. Note that it could just
as easily had 5 times the conversion rate (equal to the
premium), but then it would have cost more per conversion
due to the lower CTR. Then the question is if the residual
effect of having the banners seen by 39,999 targeted people
versus 199,998 untargeted people compensate for the higher
cost. In order to answer that we need to be able to
calculate the theoretical value of the targeted and
untargeted impressions, but before we can do that, we need
to be able to know how many different people actually saw
the banners. To my knowledge, nobody has actual answers to
these questions and this is not the thread to deal with
these thorny issues.

When doing online campaigns, you need to know the specific
results (all variables) of the specific creative on specific
audiences. This applies to all forms of traffic generation,
from banner ads, text links, email ads, etc., so it is a
very important concept. Generalities, like a conversion
rate better by a larger factor than the premium are not
precise enough to evaluate things properly when calculating
cost per conversion. The breakeven point in this case would
have been a 5 times conversion rate, so getting 30 times
means that we are getting 6 times the value, right? Well
actually, no, and this leads us to Tim's question 1.

The fact that there are two events that have to happen in
order to measure the cost per conversion means that we
cannot take the CTR for granted. In my experience, things
never seem to go where I would have predicted, so I have
stopped trying to guess what the Internet will generate and
just measure what it does. The implicit assumption that the
CTR would be better to Targeted is something that has no
experimental basis, and therefore it should not be a
surprise that in this case it turned out this way. If the
creative was different (or the targeted audience was
different), it could easily have had a higher CTR. Once you
start changing the creative, then not only does it change
the CTR, but also the conversion rate. All of this doesn't
take a Math degree from Caltech to understand, the key point
is to treat the CTR and conversion rate as independent
variables and measure them to verify what they are, rather
than assuming ANY type of relationship.

I would be happy to conduct some controlled experiments and
publish the results to the group, but I need some help to
set up the tests. We have two types of traffic coming to
our websites, one group would be considered "Targeted" for
companies trying to promote Window's software downloads to
advanced users. The other traffic has a significantly
larger population and contains the first group, so it would
work as a reference "Untargeted". I can think of three very
interesting types of experiments, the first one would be to
see what the difference in conversion rate the identical
creative will have for a couple of different types of
downloads. We can also see the difference in conversion
rates of different creatives to the same audience. By using
keyword based text/icon links to generate clickthroughs, we
could run all of these experiments in parallel. A two week
test should generate about 2,500 clickthroughs for each
sample point, so the measured conversion rates should be
statistically relevant.

In order to be able to generalize and produce useful results
to the group, I need several volunteers who have Windows
software downloads that they can measure the conversion rate
on. Volunteers, please email me and we can publish the
results in mid-February.


Sincerely,

Clint Ballard
C.E.O., Acceleration Software, clint_at_accelerationsw.com
"http://www.clicksales.com", pay-per-click advertising with a 100% satisfaction
guarantee


========================================================================
------------------------------------------------------------------------
       This week's Online Advertising Discussion List sponsor:
                                Conducent

   Don't Buy Hope! Get a 94% User Response Advertising with Conducent!
     Place Dynamic Ads in Downloadable Software. Invest in Results!
 http://www.conducent.com/advertisers.shtm or eisenstadta@conducent.com

------------------------------------------------------------------------
========================================================================
Online Advertising Discussion List To Unsubscribe send UNSUBSCRIBE
http://www.o-a.com/ to online-ads-request@o-a.com



Received on Tue Feb 02 1999 - 07:22:53 CST