Why Google click fraud is NOT 0.02%

It is no wonder Shuman Ghosemajumder, Google Business Product Manager for Trust & Safety, invests in lengthy and frequent public missives on how Google “protects you against click fraud,” his almost every Google backed word tends to be cited and reinforced in the media as irrefutable truth!

Perhaps the most misinterpreted element of the present Google click fraud proclamation is the Ghosemajumder affirmation:

Because of the broad operation of our proactive detection, the relatively rare cases we find of advertisers being affected by undetected click fraud constitute less than 0.02% of all clicks.

In taking Google at its ambiguous word, many in the media simply reiterated the Google 0.02% stat in relation to “click fraud.”

For example, Danny Sullivan headlined “Google: Click fraud is 0.02% of clicks.”Is Sullivan’s headline accurate? A precise reading of Ghosemajumder suggests not.

Google itself qualifies the 0.02% stat:

Put another way, for every ten thousand clicks on Google AdWords ads, fewer than two are reactively detected cases of possible click fraud.”

Google is NOT affirming that for every ten thousand clicks on AdWords ads, fewer than two are fraudulent clicks.

Google is actually saying that for every ten thousand clicks on AdWords ads, the number of advertiser submitted invalid clicks claims that Google agrees to reimburse the advertisers for amounts to less than two.

Jakob Nielsen also refutes Sullivan’s analysis of Ghosemajumder’s calculation, in lengthy comments discussing Google’s “math mistake”:

As I read the article, the 0.02% number refers to the number of fradulent clicks discovered through client-requested investigations divided by *all* the clicks on Google. This doesn't mean that 0.02% of clicks are fraud that was not discarded automatically. For that to be true, *all* clicks would have to go thorugh the manual investigation.

In other words, the percentage is calculated by using different scopes for the numerator and the denumerator. The type of classic math mistake one would not expect from Google.

The only valid percentage is the number of manually-discovered fraudulent clicks divided by the number of clicks in those campaigns that complained and were investigated. Presumably, most campaigns don't complain and thus are not investigated, meaning that they hide some additional clicks that would have been found through manual investigation.

The true percentage should be derived by taking a random sample of campaigns, whether or not they have complained, and investigate them manually.(Only investigating campaigns that complain might bias the sample, if you assume that some companies have the ability to estimate the extent of fradulent clicks in their campaigns.)

All of this, of course, assumes that the manual check is in fact capable of identifying the fraud.   

Earlier this morning I dissected various other components of the latest Ghosemajumder public proclamations to conclude “Beware Google $1 billion click fraud PR campaign” and “Google: Stingy with click fraud refunds?”