E-commerce Insights: All You Should Know About Click Fraud
Catalogers and other search advertisers are justly concerned about click fraud. Click fraud is when a person (or computer) imitates a legitimate user clicking on a pay-per-click ad, without actual interest in the ad’s target.
Like Justice Potter Stewart’s definition of pornography — “I know it when I see it” — click fraud escapes precise definition. To know when a click is fraudulent, one needs to know the clicker’s internal motivation for clicking or be able to prove the clicker was an automated ’bot. Most experts agree that few individual clicks are “good” or “bad.” Instead, investigators assign quality scores that indicate the probability that a click is legit.
Rare or Rampant?
Specialized companies have sprung up to help retailers track click fraud. Such firms have reason to overestimate click fraud prevalence. Last August, Google analyzed some third-party click fraud reports and found serious errors. (See How Fictitious Clicks Occur in Third-Party Click Fraud Audit Reports, at www.google.com/adwords/ReportonThird-PartyClickFraudAuditing.pdf.)
“It’s unclear how exaggerated click fraud estimates are developed, but they very likely reflect self-interest to hype the problem,” Google spokesperson Barry Schnitt tells me. “The reports often include clicks we’ve already filtered and aren’t charged to advertisers.”
Conversely, search engines have motivation to underestimate the problem. Google CFO George Reyes recently labeled click fraud “the biggest threat” faced by the Internet economy. What’s more, Google’s 2006 annual report warns, “If we fail to detect click fraud or other invalid clicks, we could lose the confidence of our advertisers, thereby causing our business to suffer.”
The key issue is advertiser confidence. Since advertisers can’t audit how engines actually charge for clicks, maintaining advertiser trust is essential.
Click fraud detection companies employ different strategies. For example, ClickForensics.com, a research firm created to develop an industry solution to click fraud, stresses the importance of pooling data. Authenticlick.net relies on data-mining techniques. And the Web analytics firm ClickTracks.com opts for human investigation of unusual cases.
+ behavioral, and
+ economic factors.
Credit Card-like Models
Michael Leonard, CEO of the recently launched Authenticlick, described how his company rates clicks “using models similar to those used in the credit card industry for credit scoring.”
To build scoring models using CHAID (chi-square automatic interaction detector) trees, he notes that Authenticlick combines tens of thousands of metrics, including:
+ Internet provider patterns,
+ time of day,
+ landing page and
+ country of origin.
Michael Stebbins, vice president of marketing at ClickTracks, describes how ClickTracks creates baseline statistics for each advertiser. Consider these metrics:
+ time on site,
+ conversion percentage,
+ campaign cost, and
+ IP country of origin.
Clicks that deviate from the norm in too many dimensions, he says, are brought to the attention of the advertiser for human intervention.
“Trimming poorly performing ads can provide more than 10 times the benefit as worrying about fraud,” Stebbins says. “It’s really all about the marketing. Cut out your bad ads!”
Most criminals who perpetrate click fraud do so for cash. Understanding money flow is essential to understanding click fraud. Differentiate between core search and contextual search.
Core search consists of the familiar text ads that appear on Google and Yahoo! search results pages. When a searcher clicks on one of these ads, the engine charges the advertiser and payment flows from advertiser to engine.
Who benefits from a fraudulent click on your core search ads?
The engine earns a few pennies, but the major engines wouldn’t defraud you. The risk is too great for such small return. However, the fraudulent click still harms you and in so doing, it helps your competitors.
Legitimate competitors have too much to lose by perpetrating click fraud themselves. But a smaller competitor might click your ads to drain your budget, run you off the page and degrade your economics.
And while your legitimate competitors are unlikely to commit direct fraud, they might unknowingly employ a disreputable agency or affiliate.
Core search goes beyond ads on the search engine’s own pages. Engines also syndicate search ads to distribution partners.
For example, Google powers AOL search. While syndication to reputable partners shouldn’t concern advertisers, whether intentionally or by accident, core search ads sometimes are syndicated to less reputable sites. ClickTrack’s Stebbins described a situation where, for a brief time, his non-content Yahoo! ads were syndicated to a spammy toolbar. So, be sure to watch your referrer strings and follow up on inbound paid clicks from unexpected sources.
How prevalent is click fraud on core search?
ClickForensics’ Cuthbert estimated 14 percent during the third quarter of 2006. Authenticlick’s Leonard suggests that 5 percent of clicks probably are bogus. ClickTracks’ Stebbins also cited 5 percent prevalence. And while Google’s Schnitt didn’t provide a statistic, he assures that “the number is very small.”
Contextual search is a totally different beast than core search. In contextual search, engines syndicate search ad results to hundreds of thousands of publisher sites, matching ad keywords to page content. The engines, in turn, share revenue from the click fee with publishers.
For example, Google shares back about 80 percent of click fees with publishers, with most of this money going to major media sites. Typical AdSense publishers are believed to receive 10 percent to 20 percent of the advertiser’s click charge.
Content networks are wonderful for legitimate publishers. Search engine payments support a slew of interesting Web businesses. Such businesses provide free services to consumers.
Criminals also migrate to content networks by creating bogus pages with content ads for costly phrases, and generating false clicks on those ads to steal money from advertisers.
Sophisticated criminals use viruses and spyware to turn infected machines into zombie click farms. Stebbins noted that criminals used to target $3-plus clicks, but now fraudsters are going down market and are targeting $0.75-plus clicks to avoid detection.
How prevalent is click fraud on the content networks? Stebbins estimates as much as 40 percent. Leonard quoted a prevalence of 25 percent. Most experts agree that click fraud is far more prevalent on the content networks, with fraud rates perhaps as much as five times higher than on core search.
Reducing Your Exposure
What one simple action can catalogers take to reduce their exposure to click fraud?
Avoid contextual advertising. And while avoiding content will reduce both your traffic universe and your cost, it probably won’t harm sales. If you do decide to run content:
+ track those campaigns separately;
+ test carefully; and, if possible,
+ restrict your ads to selected publisher sites individually by specific URL.
There are good clicks to be bought within the content networks, but buying them takes great care and attention. Watch your conversion metrics; when you see content ads not selling, turn them off.
May all your paid clicks be valid, and may those clicks yield profitable sales.
Links of Interest
Shuman Ghosemajumder’s Blog
Google Click Fraud Engineer
How Fictitious Clicks Occur in Third-Party Click Fraud Audit Reports
The Lane’s Gifts v. Google
Report by Prof. Alexander Tuzhilin
Click Forensics Blog
Alan Rimm-Kaufman is president of the Rimm-Kaufman Group, a paid search marketing firm serving catalogers and direct marketers. Reach him via his online blog at rimmkaufman.com/rkgblog.