In the midst of the controversy over click fraud, Google takes the initiative and reveals a part of its system to fight this scourge likely to shake the entire sponsored link market.
The research firm Outsell Inc. had set fire to the powder by denouncing last July a rate of click fraud on sponsored links of 15%, equivalent to a deadweight loss of 800 million dollars for American advertisers. On the sidelines of this frightening figure likely to call into question the entire economic model of Cost Per Click (CPC) advertising, Outsell Inc. stigmatized the opacity of the anti-fraud practices of the major agencies, Google in the lead. The leader of the sponsored link, aware of the risk that these accusations pose to its activity, is now back on the offensive and unveils a part of its system to fight against this scourge.
"Click fraud is certainly a major risk, but we are doing what is necessary to ensure that it remains a minor problem", specifies Shuman Ghosemajumder, product manager at Google and specialist in the issue. Under the term "click fraud", Google designates all clicks made without the intention to purchase or search for information. Its objective is therefore to identify these malicious clicks so as not to charge AdWords advertisers for them and to maintain their level of return on investment.
The task is more difficult than it seems since real networks of organized clicks have been set up, as well as sophisticated fraud techniques. Among the most frequent, we can observe the multiplication of advertising link farms, but also click pyramids with a cascading redistribution of income or the infection of PCs of honest Internet users by Trojan horses which will replace normal advertisements with pirate inserts.
These are all illegal practices that put Google at risk of seeing its advertisers lose interest in its management because of the decline in the effectiveness of advertising campaigns. Google has therefore developed a filtering procedure based in part on statistical analysis applied to the law of large numbers and reinforced by technical countermeasures to detect fraudulent clicks. A procedure based on 4 successive steps.
A first automatic filter is applied in real time to all clicks recorded on the AdWords network. This level 1 filtering would make it possible, according to Google, to detect nearly 95% of fraud attempts thanks to the use of several hundred cross-criteria, such as the geographic location of the click or the browser used or the IP address of the Internet user.
A first level of filtering detects 95% of fraud This filtering also proceeds from a statistical analysis allowing Google to identify abnormal peaks of activity by crossing each of the criteria for identifying clicks. "We keep the list of these criteria secret and our data-mining tools have been developed in-house. We also do not use data from our Google Analytics tracking service, which remains at the discretion of the users", specifies the product manager.
The second and third level filters specifically concern publishers of partner sites in the AdSense network suspected of carrying out illegal practices. Automatic alerts proactively warn advertisers of a sudden drop in results, for example. They are then able to manually request the cancellation of certain clicks. "We are closing AdSense accounts every day and automatic tools make it possible to prevent sanctioned individuals from opening a new account under a different identity," explains Shuman Ghosemajumder.
The fourth level of protection allows concerned advertisers to request an investigation of contentious cases. "The majority of these requests for investigation do not in fact relate to cases of fraudulent clicks actually invoiced. These are often counted automatically from level 2", indicates Shuman Ghosemajumder.
The results of this filtering communicated by Google do not confirm the results of the study carried out last July by the Outsell firm. “Less than 10% of clicks recorded are detected as invalid,” says the product manager. These invalid clicks are therefore not billed to advertisers, which further reduces their potential deadweight loss. "Outsell's study was based only on a panel of advertisers who were asked to estimate the level of fraud. These results are therefore not scientific", concludes Shuman Ghosemajumder
