Much ado over click-fraud statistics

Click Forensics has determined that the average click-fraud rate for Web ads was 15.8 percent for the quarter, but the companies serving those ads dispute the numbers

The battle between advertisers and online search networks over the pervasiveness of click-fraud continues to grow more heated with researchers claiming rapid growth of automated ad impressions and outside observers noting an overall lack of transparency in the space.

According to a research report issued on July 19 by Click Forensics -- which markets services used by advertisers to analyze online advertisement traffic -- the average click-fraud rate for Web ads was 15.8 percent across the industry during the second quarter of 2007.

The result indicates a 14.1 percent increase in click fraud when compared to the same period last year and a 14.8 percent gain over the first quarter of this year, based on the company's numbers.

Further, Click Forensics contends that the fraud rate within pay-per-click networks run by industry leaders, including Google and Yahoo, accounted for 25.6 percent of all traffic during the second quarter, representing a 21.9 percent gain over the same timeframe in 2006 and a 19.2 percent leap in fraudulent hits compared to Q1 2007.

Contributing heavily to the significant gains in contested impressions, Click Forensics maintains, was the increased use of hijacked botnet PCs to manipulate online traffic. While not offering concrete statistics on the trend, Click Forensics' research showed that such activity roughly doubled over the second quarter.

Market watchers have been trumpeting massive growth of the online ad sector in 2007, and rightfully so, said Tom Cuthbert, chief executive of Click Forensics.

However, if the online space is going maintain credibility with advertisers going forward, Cuthbert said, those who control the largest networks must begin working harder to track and eliminate inflated results.

New York-based market research firm eMarketer has predicted that the North American online advertising segment alone will grow by more than $3 billion in 2007 to reach $19.5 billion by the end of this year.

"The problem only appears to be getting worse within the content networks run by companies like Google and Yahoo, it's very concerning to advertisers, and the botnet issue appears to be fueling this growth," Cuthbert said.

"A big part of the problem is that the content networks themselves aren't doing enough to look for traffic coming from automated sites and parked domains that only exist to create fraudulent clicks," he said. "We see a lot of this low-quality traffic occurring inside those networks."

Click Forensics gathers its research data from Click Fraud Network, an effort it sponsors that aggregates information provided by the company's advertising customers about the advertising traffic that they receive.

Yahoo representatives did not discredit the Click Forensics report but defended that the company is weeding out fraudulent ad hits and working to ensure that its customers only pay for legitimate traffic.

"We've dedicated significant resources to delivering high-quality traffic to our advertisers and have identified and not billed for 12 to 15 percent of the clicks on our network [as a result of click-fraud]," said Reggie Davis, vice president of marketplace quality at Yahoo.

The company is also launching new programs to help offer its advertisers a better idea of what they're paying for, Davis said.

"Yahoo is also actively pursuing numerous new quality initiatives that provide advertisers with more control over and visibility into the quality of their traffic," he said. "We've recently launched new features and functionality for advertisers like quality-based pricing and enhanced geo-targeting tools, and we plan to introduce additional controls like domain blocking in the coming months."

Google representatives contend, however, that Click Forensics is manipulating its numbers to make the rate of click fraud appear much higher than it actually is.

A good deal of the traffic being counted by Click Forensics as click fraud is already identified as such by Google's own filters, the company said in a statement, which means its advertisers never get charged for those hits.

"This estimate counts clicks Google does not charge to advertisers as fraudulent, so it is not actually a click-fraud estimate," Google said in the statement. "It is also worth noting that in all of 2007, only two advertisers have contacted us regarding click fraud data from Click Forensics, and in both cases, we found that the suspicious activity was not charged-for in the first place."

Media representatives at Google labeled the Click Forensics report as an intentional effort to generate fear and drive interest in its own ad performance tracking services.

The company has a firm grasp on click fraud and is devoting more resources than ever to combating the problem, the Google representative said.

One of the biggest questions facing Click Forensics in defending its research is just where its numbers come from and how it quantifies the activity of botnets, which are often comprised of larger numbers of machines that spread traffic out as to evade discovery by behavior-monitoring filters.

Cuthbert said that Click Fraud Network has attracted a large enough group of advertisers to garner accurate averages in traffic activity and countered that the firm has access to such data as the behaviors exhibited by bots once they arrive on its customers' sites that Google will never be privy to.

"We use people from Symantec and Yahoo and eBay to help us determine these findings and approach the issue in a very scientific way that utilizes heuristics for identifying traffic for botnets and pay-for-click sites," Cuthbert said. "Our customers are drawing data from across multiple networks, which allows us to compare results; we get that behavior information about what a visitor does after they get to the site which tells us a lot about who they might be."

For instance, when Click Forensics notices hits coming from a particular source on a regular pattern -- such as a new click every 2.5 seconds -- it's easy to tell what traffic might be generated by bots, he said.

Cuthbert also noted that his company's methodology was examined by Dr. Alexander Tuzhilin, a professor at New York University who examined Google's click fraud practices and wrote a report used as part of the $90 million click-fraud settlement the search company reached with advertisers in March 2006.

Other click-fraud experts, such as Ben Edelman, an assistant professor at Harvard Law School who has done extensive research in the area for a number of years, agreed that the problem continues to grow and that it's hard to believe which camp has more accurate numbers.

However, the researcher's own recent experiments don't always gel with Click Forensics' observations.

Edelman said that a system he has employed to monitor click-fraud activity on advertising networks by attempting to carry out the behavior has been able to scam just about every popular system, including that controlled by Yahoo's Omniture group.

However, the "robot PC" he created to carry out the click fraud wasn't able to create false impressions on Google's AdWords, which impressed the Web economics expert.

"The core problem is that it is still hard to determine whether the supposed clicks occurred, the search engines will always dispute published figures, and advertisers will be left uncertain who to believe," Edelman said. "In my own work using the robot, I've observed click fraud through almost every network, but not AdWords, not that it means it doesn't happen."

One of the problems in figuring out to what extent click fraud is influencing ad traffic is that neither the search networks nor their advertisers are particularly open with the data they retain about user behavior, he said.

People submit complaints to the search companies on an individual basis, but little comes of the reports other than automated responses, according to Edelman.

One approach that could help address the issue would be if companies like Google and Yahoo created message boards where advertisers could compare notes and draw conclusions about their shared experiences, the researcher contends.

Just as Apple has allowed its customers to vent about product problems in its public forums, the search networks could prove their diligence in fighting click fraud if they allowed customers to share information in public, the expert suggested.

"A different way to run the traffic auditing and review process would be to add this transparency, to allow crowds of advertisers to expose their problems and share complaints, that would allow much less leeway for the search providers to bury their results," said Edelman. "It's clear why they might not want to do that, but they could do it if they were really confident that there's no problem to be exposed."

Faced with the proposition, Google representatives said that the company would consider such an idea, but indicated that for now, it harbors no plans to introduce such a program.

"Without that type of transparency, the extent of the click-fraud problem will remain tough to nail down," Edelman said.

Copyright © 2007 IDG Communications, Inc.