In article <telecom24.171.17@telecom-digest.org>, <gds@best.com> wrote:
>Robert Bonomi (bonomi@host122.r-bonomi.com) wrote:
>> With web-pages, there is *no*way* to estimate how many people see any
>> particular ad. *OTHER* than to count how many times it was displayed.
>> And that is not a "reliable, accurate" number, by any means. What it
>> is, however, is the "best available" data for estimating.
> I realize this, actually. However, this "best available" method of
> estimation is what is in dispute.
So? What else is new? It's a *stupid* lawsuit. "caveat emptor"
applies to the buyer of advertising, just as it does to buyer of the
advertised product.
If the buyers "didn't understand" the _inherent_ limitations and
inaccuracies of (a) the marketplace, and (b) the particular pricing
scheme, they really shouldn't be playing in that market.
_Smart_ web-advertising buyers, have, for *years*, being writing their
pay-per-click contracts in a manner which makes it difficult for someone
to fraudulently inflate the page click-count.
>> Consider a "fleet" of 500,000 "zombie" PCs, scattered across three
>> continents.
>> Each machine, _once_a_day_, at a random time, connects to a given
>> web-page, without anybody in front of the machine.
>> Now, just _what_ are you going to detect?
> I have described many scenarios such as this where there is no
> reliable way to differentiate them from clickstreams where the users
> do not find what they are looking for at the advertiser's site, or
> decide (non-fraudulently) not to buy, or are just window shopping.
Yup that's the point. It's *impossible* to do so. Which answers your
question about "why aren't people working on better fraud-detection?"
Those who have seriously looked at the problem recognize that
detecting such actions 'in progress' chews up exorbitant amounts of
resources, and costs more than the fraud does. The 'simple' stuff --
e.g. paying only for 'unique' clicks over time -- kills off all but
the sophisticated fraudster. The sophisticated fraudster, on the
other hand, is effectively _impossible_ to so much as slow down.
> I guess I don't understand the general tone of your response. It
> seems you are agreeing with me that PPC is a poor business model.
It is what the *advertising buyers* have *demanded*. This wasn't some
bright idea dreamed up by the folks selling ad space. Fixed pricing
is much simpler for _them_ to deal with.
"Poor model" is a term without a referent. _First_ you have to define
the goals; only after you have done that, can you attempt to evaluate
the 'quality' of the method. With advertisers demanding "$/M" figures
or equivalents, to base their buying decisions on, PPC *is* the 'best
available' fit to their demands. That the best available technology
doesn't live up to their expectations is *their* problem.
> you feel that it is superior to, say, paying fixed fees for a certain
> period of time, I'd like to know why.
Fixed pricing ends up in *exactly* the same position, unless you're
paying for your ad to appear *every*time* somebody calls up a web-page
at that site.
If you're on a limited budget, and cant afford a "shown on every page"
ad, but want 'general' exposure, you can only get that by having you
ad appear on 'some' pages And, thus the question becomes "how many
times do you want that ad to appear?" whether it's expressed as "so
many thousand times", or "such-and-such percentage of the time",
> The advertisers can use information that comes from companies such
> as Nielsen NetRatings to estimate how many people use a search
> engine, and what queries they submit to it, to determine a fair bid
> for an ad buy.
*snicker*
Nielsen NetRatings lacks -- by several orders of magnitude -- having
enough reporting sources to produce estimates that are within a factor
of _ten_ to _fifty_ for all the various 'keywords' that the
search-engines selectively sell ad-space for.
> Such information is no worse than what is used to determine rates
> for TV or radio ads.
It is *far* worse, in point of actual fact.
The number of samples you have to have, to have a 'meaningful'
representation of the population, depends on a number of factors. Of
which the size of the population is only one. The number of different
categories enters into the picture, as does the relative frequency
which the population as a whole touches that category.
For a Presidential election, where roughly 120 million of 220 million
voting- age adults did vote, you can get +/- 3% accuracy with a sample
size of circa 1,500. For a population of a million, you can get the
same accuracy with a circa 500 sample size.
However, to get a similarly meaningful estimate of something where
only 1:10,000 of the population uses it, you've got to have a sample
that includes enough people that _might_ use it, before you can have
any confidence in the numbers. Say that 1:4 of the people who -might-
use this thing, use a search engine to find out about it. That means
that roughly 1:40,000 people will hit on that keyword. and the
underlying population is about 100 thousand, if you assume that 1:5 of
those who might use the thing itself, do use it.
To get a meaningful sample on a population of 100K, you need a sample
size of a couple of hundred _of_that_population_. to get *that*
couple of hundred in a sample of the 'general population' of 220
million, you need about _THIRTY_THOUSAND_ people in the sample.
Unless it is a 'scientifically selected' sample, with balanced
demographics matching the population as a whole, you at least triple
that sample size -- to 'hope' to minimize the effect of distortions in
the sample.
TV/Radio numbers are *much* easier to do relatively accurately,
because the spectrum of possible choices is much smaller.
> --gregbo>
> [TELECOM Digest Editor's Note: As for myself, I cannot really picture
> 'five hundred thousand zombie computers scattered across three
> continents'.
Your "inability to picture" is not relevant to the real world. <wry grin>
Spammers, *today*, are routinely _advertising_ access to pools of
50-100 *thousand* zombie machines. "small" pools are only 10,000
machines or so.
A large portion of PC viruses currently being spread are for the
purpose of turning machines into zombies, for spamming, DDOS attacks,
and other kinds of outright criminal activity.
> If so, under whose coordination? A gang of crackers all
> working in concert to cheat some advertiser's competitor, by running
> up his advertising bill? Seems sort of improbable to me. PAT]
Your "disbelief" is not relevant to the state of affairs in the real world.
Those zombie armies *are* out there, and *are* being used for many
kinds of nefarious activities. I don't have any direct knowledge of
their being used for click-fraud, But there is absolutely no question
that they _could_ so be used.
If the criminal gangs can put together several hundred thousand
machines to execute a DDOS attack against a web-site, as _has_ been
done more than once. *ALL* they have to do to make it click-fraud is
change the URL to point to an ad.
The _typical_ "click-fraud" scheme is to _make_money_ doing it, not to
run up the expenses of a competitor.
The way it works:
Somebody puts up a web-site with 'pass through' ads from somebody
like Yahoo. Yahoo supplies the ad content -- for people who have
bought ad space through their 'syndication' service. When a viewer
clicks on the link on that web-site, "somebody" gets some money for
it.
The scam comes in when that self-same "somebody" contrives to have
_lots_ of clicks happen to that Yahoo-supplied (or whomever) link.
They don't care _who_ the advertiser is, or what they're selling,
The idea is to run up the revenues for that "affiliate" site owner.
At the expense of the syndication seller, and the actual
advertiser, of course.
Robert Bonomi (bonomi@host122.r-bonomi.com)