According to a 2009 report by the Internet Advertising Revenue, 22.7 billion dollars have been spent for online advertising. Twenty two percent of this figure is for display ads. These are huge numbers, but there are very few studies about how effective these ads are and the huge amount of money spent on them. This is the summary of an article by Randall A. Lewis, David H. Reiley, and Taylor A. Schreiner. You can get the pdf of the behavioral targeting article here: Ad Attributes and Attribution: Large-Scale Field Experiments Measure Online Customer Acquisition.
The study tries to make an assessment of the effectiveness of online display advertisements in making users sign up accounts for a certain online business by coming up with an attribution model based on large-scale experimental data.
This study randomly assigns a Yahoo! visitor to a control or treatment group. The treatment group consists of four different ads based on format and placement. Two formats are considered: banner and large rectangular (LREC), while two placements are considered: Yahoo! Run-of-Network and Yahoo! Mail. In addition, these Yahoo! visitors are relevant to the advertiser in this study.
A browser cookies is used for each Yahoo! visitor for random assignment. Fifty two percent of the randomized Yahoo! visitors were sent to the control group where they saw other advertisers’ ads, while 48 percent were sent to the treatment group. Since the treatment group was divided into four kinds of ads, each had 12 percent of the targeted population. The four kinds of ads are the following:
Yahoo! Mail banner ads, Yahoo! Mail LREC ads, Yahoo! Run-of-Network banner ads, and Yahoo! Run-of-Network LREC ads.
3.7 million Yahoo! visitors were sent to the treatment group, where a total of 67.4 million ads have been received by them during January to February 2008.
Descriptive statistics show that banner ads enjoy higher frequency per user compared to LRECs, and users tend to click more on banner ads as well. Another observation is that only a small amount of users clicked the ads and signed up for accounts. Therefore there is high interest in those who didn’t click yet signed up for accounts.
In the six-week campaign for the particular advertiser in this study, 20,947 users from the control group and 19,196 users from the treatment group signed up. Comparing Yahoo! Mail and Run-of-Network, Run-of-Network advertisements performed better because it had a statistically significant and strong impact on sign-ups. Ad placement is also compared, and the study shows, that both LREC and banner ads performed well in Yahoo! Mail. Run-of-Network ads performed well too.
As a summary, Yahoo! Mail advertisements did not produce a statistically signficant effect, and is outperformed, at least for the advertiser in this study, by Yahoo! Run-of-Network.
A rough calculation on cost effectiveness in this study showed that the advertiser would probably regain the costs of these advertisements after four years. However, results in this experiment will cut that value in half because it will encourage the advertiser to focus on the more effective Run-of-Network based ads.
A lot of factors affect the effectiveness of online advertising. The factors considered in this study show that more users respond to ads and actually sign-up for the advertising business when the business uses Yahoo! Run-of-Network advertisements. In addition, this study showed that attribution models that are not based on experimentation may be less effective because they give credit to users that clicked and responded to ads, while these users didn’t have to click on the ads to sign up.
Attribution that is click-based can also understate online advertising effectiveness because it ignores more than ninety nine percent of the users in this study who did not click the ads, but went to the advertiser’s website anyway and signed up.