Click Fraud Update: Arkansas, Armageddon, etc.

We've had much debate about click fraud on these pages, so I wanted to share some of the recent work we've done on the topic over at Cherry Hill Research.  I'll try not to bore you with updates on CHR, but when we publish something that is free to all and might be of interest, I'll post a link here.

Here are some of our current conclusions regarding click fraud:  (The full write-up at CHR is here.)

• The click-fraud concern is real, and the search engines must do a better job of addressing it.  This said, the problem is manageable and will not cripple the PPC industry.
• Increased awareness of click fraud (or, more accurately, the "invalid click" problem) will likely lead to increased spending on click-stream auditing by both search engines and advertisers.  This will modestly reduce ROIs for advertisers (and thus weigh on keyword prices) and modestly reduce profit margins for search engines.
• Google's settlement in Arkansas (if approved) will be a much bigger win for the company than we initially thought.  It will preclude all future class actions based on historical click-fraud claims, including a large one that has already been filed in California.
• The plaintiffs attorneys on the California class action, in fact, called the Arkansas settlement "the worst class action settlement in history" [from the perspective of the plaintiffs.]  From Google's perspective, therefore, the settlement might be described as the best in history.
• The settlement payouts are such that plaintiffs will receive not 'pennies on the dollar' but fractions of pennies, and they will receive them as rebates on future ad spending, not cash.  As far as legal exposure goes, therefore, Google CEO Eric Schmidt might have understated the case when he dismissed the click-fraud issue as "not material."  A more accurate assessment might have been "non-existent."
• The Arkansas settlement does not preclude future legal action for claims after the settlement date.
• Because Yahoo has not settled the Arkansas class action, the California case against Yahoo will continue.  Given the terms Google got, as well as the opportunity to settle all historical claims, Yahoo might be well-advised to settle in Arkansas immediately.

Our write-up at CHR includes links to several interviews on the topic, including Ben Edelman (Harvard spyware expert), Andrew Goodman (CEO of SEM firm), and a pissed-off advertiser.  We will be posting more interviews in the next few days. 

Hope you find the work interesting and/or helpful.  Look forward to hearing your thoughts.

Comments

[King Troll]

FIRST Son !!!!!

[jared lansky]

th

[Victor]

This is great news for google...

[web20guy]

There must be a way to quantify click fraud. accurately.
we must challenge the phd crowd at google to use some CAPTCHA techniques to eliminate the automated part at least.
they will not do it. larry, sergey, eric and george will not be happy.

[John Shiznit]

Have no idea how class action lawsuit works.

Is it merely up to the judge to approve the settlement?

Can the California plaintiffs opt out of the settlement if they are unhappy with the terms?

[Henry Blodget]

No, the California plaintiffs can't opt out. They're in or they forfeit any and all historical claims. That's why the settlement is so good for Google.

[TallTroll]

>> CAPTCHA techniques to eliminate the automated part at least

Captcha can be beaten automatically. I know a guy who reckons he can break around 300k Captchas a day using some very simple software... if it became commercially necessary for Captcha to be broken on a large scale, it would be. There are a couple of serious conceptual flaws in it

[Mark]

Great research, but I'm confused why no one talks about changing the underlying advertising model (PPC) that causes this problem in the first place. Isn't it a bit like treating the disease instead of working on a cure?

[Henry Blodget]

Meaning that the alternative would be Pay-Per-Action or Pay-Per-Sale or something? The problem with that is it requires the advertiser to share data with the search engine, and it also penalizes the search engine for an advertiser's poor ability to convert customers on its own site. In other words, if Google sends an etailer a real potential buyer, and that potential buyer sniffs around the etailer's site and decides for some reason that has nothing to do with Google he/she doesn't want to buy anything, Google should still get paid for the referral. The other problem is that customers often take several visits to convert, so you can get into a pissing match over whether the referring engine should get paid or not.

[John Shiznit]

I think for the main goog site the pay per click model works fine. The only incentive for click fraud there is a companies rival continually clicking ads.

As for the AdSense network where they put ads on your site, that whole model seems flawed. AdSense seems to incenitvise click fraud way too much.

[mike]

Any commments on microsoft's new web push? Looks like they are really serious about search this time.

[Still Inside]

Just to take the PPC vs. PPA debate a step further, the paid search vendor could automatically optimize the placement of PPA ads that make them more money over time. Google's CTR algorithm already does this for PPC, and I reckon that it would work even better with PPA.

As such, you could argue that a PPA model could potentially make the paid search guys a LOT more money if they did it right.

The hard part of PPA is the accounting. The connection between the advertiser and the search partner would be FAR more complicated than it is now. Folks like GOOG, YHOO and MSFT should see that as a good thing, not a bad thing. Complexity = money when it comes to IT. At some point, SOME company is going to come along and do the dirty work of creating a better interface between the G2000 and Paid Search. I wonder who it will be.

SI

[Mark]

As a follow up to the PPC vs. PPA discussion, I think SI makes an excellent point. To expand: In the PPC model relevancy to Google=CTR because they get paid by the click. But if you asked a consumer (or advertiser), I think they would tell you that relevancy is better defined by whether the visit actually results in success (e.g., a sale or some other defined action). Incentize the intermediary on this end goal and they would quickly change their relevancy algorithm, which helps improve their results.

As to advertisers sharing this data, a PPA model would not only eliminate click fraud, it would also create a very attractive, low risk advertiser channel. May be worth the effort of working with the intermediary and sharing this data.

[Neal Lachman]

Imagine in a year or so a brandnew search concept with a brandnew advertising concept. Would that not screw GOOG up?

It's coming.

[Meg]

Correct me if I'm wrong, but is there any reason to believe that the plantiffs will accept this offer? Isn't a preposperous offer like this just typical jousting between plantiffs and defendants?

[Steve D]

interesting words:
... because the $16 billion market for online advertising has a secret. The small text-based ads that made Google a Wall Street superstar do not seem to be working as well as they used to.

``The returns have been diminishing for the past couple of years,'' said Bob Dashtizad, director of online marketing at the Intermix agency.

Web users, he said, ``have gotten more savvy and they really use search engines just to find out information, to research a product. . . . They very rarely go to the search engine to purchase.''

Dashtizad said the conversion rate -- how often a person who clicks on an ad and either buys a product, makes a donation or signs up for a service -- has gone from one new customer per every 20 or 30 clicks to one new customer per every 50 clicks.

http://www.mercurynews.com/mld/mercurynews/business/14582549.htm

[Steve D]

Google click fraud on national TV:

http://video.msn.com/v/us/v.htm?g=5ef4c128-899c-45cc-830a-ec3197c0d569&f=rssrssmoney&f=15/64rssmoney