AdMoolah News and Views

Google has announced the beta of a new AdSense referral program. They have announcements for the beta test from both the publisher and advertiser side of things.

The program allows advertisers to offer publishers a pay-per-action model. This is more like an affiliate program, publishers only get paid when a visitor you refer actually takes some action such as signing up for a newsletter or purchasing a product.

This program takes Google a step beyond what their competition, like YPN and MSN adCenter is doing, and puts them in direct competition with companies like Commission Junction. It will be interesting to see how the competition reacts.

Right now it looks like publisher need to select individual ads to be displayed on their site. I would guess that at some point Google will combine the contextual aspect of their AdSense ads with the referral program and allow ads to be automatically selected for publishers. This is the approach that companies like Turn.com and AuctionAds are taking. Update: I found out Google does allow you to select a keyword, and Google will automatically rotate through a variety of ads for that keyword.

You can sign up for the beta as a publisher, or as an advertiser. I have signed-up. If I get accepted and find out more details, I’ll be sure to let you know.

There have been reports of Google testing a large variety of different ad formats over the last month or so. I’m sure they are testing these to see which ads get the best click through rates and the best conversion rates. Even though Google is the leader in the online advertising space, it’s good to see them still trying different things to tweak the system. This makes it better for both advertisers ad publishers.

Here is a summary of their recent experiments. I’m sure there are more of them out there too.

New Google Logo tests. Google has been experimenting with different logos on the ads for a while, this is the latest incarnation. From SEO Roundtable

This is a tabbed ad. Lot’s of potential here to do all sorts of interesting things. These kind of remind me of Chitika ads that are tabbed. See more screen shots at Stanley Shilov’s Blog.

This is being called a Horizontal Image ad. Similar Vertical Image Ads have been spotted in the past as well.

AdSense ads where the headline text is in italics have also been spotted. This is a pretty subtle change, but I’m sure Google can pick up even the smallest change in the CTR between ads.

This is also a good lesson for publishers: keep experimenting. If Google is still experimenting with ads after all these years, publishers should be doing the same thing. Try different ad and affiliate programs, try different colors, formats and placements. You can never be sure that you have the best combination for your site.

A new ad network launched today called AuctionAds (aff link). It is a joint venture between ShoeMoney Media (Jeremy Schoemaker’s company) and MediaWhiz (who also owns TextLinkAds). The system allows publishers to display EBay ads based on keywords.

The system seems very easy to use. I signed up and in just a few seconds managed to get the JavaScript code to produce an ad like this:

They also have an referral program that pays 2%. Referrals are made whenever someone signs up for the system after clicking on the “Ads by AuctionAds” in the ad.

Payment for AuctionAds is done via PayPal with a minimum payment of $10.

I have a feeling this advertising option will be useful on sites that sell consumer goods. It seems like this would make it a direct competitor to Chitika. Overall it looks like a good option for publishers who want to experiment. So if you want to sign up, click on the “Ads by AuctionAds” in the above ad

Students in an Iowa State University statistics class recently used data from AdMoolah in a group project. Hadley Wickham, who teaches Statistics 480, had the students collect data and then attempt to answer questions based on the data. Two of these papers have been published on the page outline the requirements for the project.

Some of the students had some interesting questions and conclusions.

One paper attempted to answer the question: “Is there a difference in earnings between categories, sub-categories, language and page view/page rank?” The conclusion reached was:

The analysis of the Google AdSense data resulted in more questions than answers. The main dependent variable used in the majority of the sectional analysis was average earnings. Earnings were found to be dependent upon seasonal effects (through quarter of year)and upon assigned PageRank score of the website. Increased earnings could also be captured if a website was published in either English or Danish languages.

Read the full paper here.

A second paper also had some interesting analysis. My favorite part of this one was an analysis of how more page views per visitor effects results. They found “a negative relationship between the number of times the page is viewed and the amount of earnings”. That is, having more page views makes your average eCPM go down.

Read the full paper here.

I think this project was a great idea. These are the kinds of questions I had in mind when I started AdMoolah, and these student did a good job of providing some rigorous analysis of the data. I’d like to thank Hadley Wickham and all his students for the work they put into this.

A careful reading of Google’s announcement about click fraud reveals that they really say nothing about the levels of actual click fraud.

Google was careful to comment only on the level of invalid clicks that they actually catch, not the level of click fraud itself. Here is what Google actually says
1) It’s automated invalid click detection find less than 10% of the click are invalid.
2) Manual reviews of invalid clicks reported by advertisers accounts for fewer than 0.02% of all clicks.
3) What they are talking about is invalid clicks, not fraudulent clicks. Fraudulent clicks are a subset of invalid clicks.

So, lets say there are 10,000 click in the system. Google will automatically filter less than 1000 of these clicks as invalid. Advertisers will report some number of clicks as invalid. After manual investigation, Google declares fewer than 2 of them are invalid clicks. But in actual fact, there may have been 2,000 invalid clicks in the system. Google and the advertisers just never noticed them. I’m not trying to say that the actual level of click fraud is 20%, it’s just that it could very easily be somewhere above 10%.

A lot of headlines about this story are very misleading. For example Danny Sullivan’s headline on SearchEngineLand is “Google: Click Fraud Is 0.02% Of Clicks“. Google never makes any claims about click fraud, 0.02% number really is really the number of invalid clicks found after manual review as a percentage of all the clicks in the system.

Search Engine Roundtable’s headline was “Click Fraud is 0.02%, Invalid Clicks 10%, $1B Lost To Click Fraud Yearly“. Again Google claims nothing about click fraud let alone 0.02% click fraud. This headline makes it seem like almost no invalid clicks are fraudulent clicks, which again is wrong. Google said nothing about what percentage of invalid clicks were click fraud.

Again, to be clear, Google never argues what the actual level of click fraud is, just the levels that they identify as invalid clicks.