1) Upstart networks
tend to rely on smaller sites, entertainment sites, gaming sites, and sites
that skew towards younger audiences as they try to gain a foothold in the
marketplace. As such they tend to
define what "works on their network" as what "works on these
sites", not what might or should work on the sites they will grow into as
they mature. This slows the maturation
process and pigeonholes them. It is at
the same time the reasonable conservative approach.
2) Most ad networks do not evaluate the risk/reward
threshold when evaluating a new advertiser or publisher (at all).
For new advertisers,
Risk is defined as "cost to test" (# of placements
tested on * reasonable and allowable CPA * 5). So if you want to test a new advertiser on 10 placements, and the
reasonable allowable advertisers CPA is $20 per lead, the cost to test this is
approximately 10 * $20 *5 = $1,000.
Risk is $1,000 in
this scenario.
Reward is defined as (# of placements tested * expected
profit margin for winning placements * chance of obtaining a winning placement
* available volume per month on the average winning placement * with the chance
of a placement being a winner * average number of months an ad/placement works
before replaced with a better option).
--Take a break--
So reward for 10 placements, with the expected profit margin
of a winning placement to be 40%, with an available placement volume of 50
units per month for winning placements, and the chance of a placement being a
winner of 20%, with the average number of months a winner remains a winner of 5
months).
So the expected
reward is (10 * .4 * 50 * .2 * 5) = $2,000.
SO in this specific case (and assuming your existing
business margin is 0%, and your historical data is as illustrated above), it
makes sense to test this advertiser. Our expected cost to test this advertiser to see how it does is $1,000,
and your expected profit given it working in an estimated number of placements
for an estimated amount of time at an estimated volume, is $2,000, for an expected
gain of $1,000.
3) The risk/reward is a continuum, not a singular formula
(so even if you get the above, you may be missing the real point). An advertiser with a $2 reasonable/allowable
CPA and 10% chance of working should probably be tested on more sites then an
advertiser with a $20 reasonable/allowable CPA with a 20% chance of working
(although this depends upon the size of sites you are testing on and the
average time a working combination continues to work).
I'm sure one of you math types (I haven't successfully
completed a math course since High School) can more effectively articulate the
risk/reward threshold for a singular scenario then I. I just wish one of the networks would get this right, with good
estimates for the unknowns for the risk and rewards opportunities, and test
efficiently as math dictates as appropriate. Until then, Fastclick and RightMedia will continue to run too many
flashing/blinking ads for low dollar IPOD giveaway advertisers and ignore
higher CPA on large site opportunities, and some of the large networks will
inefficiently continue to run $50 lead ads on small sites where it makes no
risk/reward sense, while continuing to aggregate small sites with little or no
performance correlation together into one data set as a publisher to justify
testing the high expected CPA advertiser with the low chance of matching
profitably.
--I expect I can clean up and re-word this post to be
a bit clearer in the future, but I wanted to publish it in the hopes of receiving
some helpful feedback sooner rather then later.
