ClickForensics CEO Paul Pellman discussed the report, its findings and methodology.
AdExchanger.com: Why do you think social networks have a lower average click fraud rate than other CPC traffic sources?
PP: There is inherently less motivation for click fraud on a social network. The typical incentive for click fraud is to generate fees for a publisher (Simple example: I write a blog, put AdSense ads alongside, and click on the ads all day so I make money). On a social network the only beneficiary of a paid click is the network itself, so there's no motivation for others to perpetrate click fraud. Also, social networks are closed networks that require stricter audience validation via user logins, so fraud would be more difficult to effect.
Can you give a use case or two on what publisher collusion fraud looks like?
Publisher collusion fraud occurs when site operators agree to help each other out by clicking on ads not on their own site, but on their friend/partner sites, generating fees for the friend/partner. That's where the name comes from, anyway. More commonly the collusion occurs when a single operator registers dozens or hundreds of domains and then clicks on the ads across all his own domains via some form of automated clicking software or botnet. The clicks look "random" because they're spread across many domains and originate from many different computers. In the picture below we provide a simple overview. Triangles represent visitors and circles are sites visited. Circles surrounded by triangles represent normal activity, many unrelated visitors to a single web site. But triangles surrounded by circles are problematic, a single visitor to dozens of interrelated sites. Combining this type of cluster analysis with volume and frequency data, we can identify suspicious activity.
Please discuss the click fraud detection methodology that was used to create this report. Could the increase seen this year over last year be attributed to increasingly sophisticated detection rather than more fraud?
Click Forensics examines PPC traffic from over 300 ad networks, top search engines, advertisers and publishers. We utilize advanced predictive modeling and heuristics to analyze dozens of attributes of every click to detect fraud and invalid traffic. We've been finding sophisticated types of fraud (collusion, botnets like The Bahama Botnet, etc.) for some time. One of the reasons for the increase we saw in Q1 was due to the fact that fraudsters are using these kinds of sophisticated attacks more frequently because they work. Collusion fraud is a great example. It's a special form of fraud that requires advanced cluster analysis and pattern recognition to find. Activity looks very random and patterns are difficult to detect and identify. But our collusion detection algorithms do just that.
By John Ebbert