How to Reduce Default Loss
What is “Default Loss”?
If you have been working with ad networks
for a while, you are familiar with the lost impressions that have been termed "default loss". The term
comes from where the impressions are lost — when one network “defaults” (has no ads to serve), and forwards impressions
over to a second network usually of your choosing. Unfortunately, the second network does not see nearly as many
impressions as the first network purports to have defaulted on, and so are “lost”. Existing publishers have observed this
loss to be as high as 30%, which means that when the first ad network has a low ad inventory,
more than 30% of a publisher's impressions go unpaid.
Understanding Ad Inventory Loss
Given a network chain of A->B->C, We've
observed that there are three kinds of default loss — one is that A fails to send the code over
to the client for B, the second is that B fails to serve when it receives default code, and the
third is that the client (user) fails to GET the code, whether limits on redirects or whatnot.
What I think
that few publishers realize is that default loss breaks down into two separate kinds of loss — Service Loss and
Forward Loss
Service Loss is the proportion of impressions that are given to the ad network which it fails to
see, and therefore does not count, and Forward Loss is the impressions that the adnetwork fails to forward over to
the next ad network in your advertising chain.
For instance, if you have a 1 ad network chain, then the
impressions lost are at these stages.
*1 = Service Loss.
*2 = Forward Loss.
YOUR SITE —>
*1 [ Ad Network ] *2 —>
As this information is not readily visible to publishers, with a number of
the networks in their advertising “chain”, there appears to be this one large default loss number which is the
aggregate of the number of losses:
YOUR SITE —> *1 [ Ad Network A ] *2 —> *1 [
Ad Network B ] *2 —>
A site with two ad networks would see a “default loss” by network B
that is the sum of lost impressions leading up to B (A*1+A*2+B*2).
Therefore:
Service Loss = (1 -
Ads Seen by Network¤Ads given by YOUR SITE).
Forward Loss = (Ads that can be seen by next network¤Ads
Reported as Default).
Default Loss Measured
We've ended up spending a LOT of time on reducing default loss.
I mean, nothing is more frustrating than watching 30% of your impressions go unpaid for even the smallest amount.
The first step is pricing it in. While we have figured out a way to manage and reduce the
loss itself, as an independent entity, your first step is to measure and “adjust” the prices that you see
from the networks by their rates of failure — both service and forward.
For instance, if a
network loses 10% of your impressions on service, then if they are averaging $1.00 CPM, then their actual (real) CPM
is $0.90. So between two networks, paying the same, you'd prefer to send your impressions to the network that
sees (and pays you) for more.
Several weeks ago (2006), we measured the service rates from our server to the
various networks — these are the relative service loss numbers that we saw:
Service Loss.
Burst: ~7%
FC:
~6%
Casale: ~10%
TF: ~9%
Now, pricing in forwarding loss is more difficult, and in fact doesn't
really solve the problem of loss, so at this point, we seek a way to lower it altogether. But
for consistency, we measured the forwarding loss from the various networks, and this is what we saw:
Forward Loss.
Burst: 33%
Casale => 23%
Fastclick: 18%
(we don't have it for TF yet.)
To some
of you, these numbers should come as no surprise. While overall we've found that Burst! pays the highest, they
also fail to forward over more impressions to other networks than others. If the Fill Rate of the network
is high, having a high forward loss isn't too big a deal, but if the fill rate is low —
say 1 impression per person, or less, then you can be losing greater than 33% of your ad inventory right
off the top, due to your chain ordering.
So formally, the total default loss of a single network is:
DL
= SL + FL x (1-fillrate)(1-SL)
With a fill rate of 0 (no ads), the default loss of a
network with a service loss of 5% and a forward loss of 33% is: .05 + .33*.95 = 0.36
—> 36%!
FYI: fill rate = (ads served)¤(total ads seen by network)
So your first steps to reducing default
loss is to measure the average “fill rate” of all of your networks, and the above information will tell you
the percentage of inventory that you are losing in each stage (and thus the real price that you are
getting).
Our next article will be “how to RECLAIM” default loss — or “PREVENT IT FROM HAPPENING”. Stay
Tuned!
Beracah Yankama