How we beat AdWords with super-premium display
A new case study show how large, prominent (seemingly expensive ) display units can be extremely efficient for direct response advertising
In the past few years we at Bonnier Tidskrifter have made significant efforts to enhance the display product that we offer on our sites. One of the issues that we have tried to tackle is the online advertising market’s dependence on clicks as the singular and dominant metric for success. Numerous studies performed by us and others have convincingly displayed the problem in using clicks as a proxy for effect, a problem that is increasing with the exponential growth of non-human traffic and click-fraud globally. A very small proportion of the Internet population ever clicks on banner ads and those who do generally convert worse than traffic from other sources when they land on the advertiser’s site. This has led a lot of advertisers to conclude that display works poorly as a direct response channel; instead directing their budgets to other direct response channels such as paid search where a click is generally a good proxy for effect.
We argue that the effect of a display ad is correlated to a much higher degree to a view than to a click and to track the effect of a display campaign correctly, one has to account for the amount of viewable impressions to which a user has been exposed, and look at how each of those views affect the users subsequent behavior. This case study does precisely that. By looking at the behavior of every user who sees an ad at least once, and subtracting the natural overlap on the publisher and advertiser site, we are able to determine how much traffic and how many conversions that can be attributed to every singular impression. The results are striking and indicate that premium display inventory has been grossly underestimated by the market as an efficient way of driving traffic and conversions to your site.
The campaign
AdLibris, Sweden’s largest ecommerce bookstore, ran a campaign for four weeks in December 2013 on our sites, primarily mama.nu and amelia.se of 1,2M viewable impressions (using the IAB definition of 50% of the ad being in an active screen for at least 1 second). The campaign ran across devices (mobile, tablet, desktop) with the desktop format being 980x480 pixels. Instead of measuring clicks, we measured view through, tracked the behavior of those who had viewed the ad and compared them to all other visitors to AdLibris in the campaign period. By having the same scripts installed over the three sites (mama, amelia and AdLibris) over a longer time we could further differentiate behavior between loyal users of AdLibris and new users to the site, and understand how campaign fundamentals like frequency affected the two groups.
Results: Driving traffic
Of the browsers exposed to the campaign, 26% visit the advertiser’s site in the campaign period, while only about 4% of the exposed traffic actually click on the ad. Just looking at the traffic generated by clicks in this campaign would thus have underestimated the volume by more than 6,5 times.
This obviously has a dramatic effect on the perceived price that the advertiser pays for traffic, where in this case the actual cost per visitor is more than 50% lower than the “straight” CPC. This is core to the case: a premium CPM campaign in the $20+ range, generated actual traffic to AdLibris at less than $1 per visitor, effectively being able to compete on cost with SEM. This is why it is crucial for advertisers to start to look at viewthrough attribution, rather than blindly staring at clicks.
Another interesting takeaway is the difference in effect between loyal users to the advertisers site, and completely new users. Looking at how much more likely a visitor who has been exposed to the campaign is to visit AdLibris, compared to someone who has not been exposed to the campaign, we can see that very few impressions are needed to generate traffic for loyal users, whereas new users need a higher frequency to visit the site. This would imply that retargeting campaigns (at least in certain categories) should be frequency capped at a lower rate than non-retargeting campaigns, and that attracting new users can indeed imply that an unusually high frequency (10+ impressions) is needed.
Results: Increased conversion
In contrast to many other direct response channels, display also has the advantage of affecting users by making them more likely to convert when they arrive to the advertiser site. In this case, users who had been exposed to the campaign were almost 30% more likely to convert than users who had not been exposed in the same period. Loyal users, who had been “reminded” by an impression, were almost 70% more likely to convert than the average indicating that “preaching to the choir” can indeed be an efficient strategy with display.
Interestingly, users who have been exposed to a display ad are more likely to convert independent on how they finally arrive to the advertiser’s site. It is also interesting to note that the traffic that arrives to the advertiser’s site through a direct click converts worse than through any other channel (exposed, or unexposed) further indicating the low quality of data that will come through evaluating a display purchase through straight CPA. The “real” effect of the display ad is thus way more significant in driving SEM/SEO and making that traffic convert better, than in generating clicks that convert.
In conclusion, the actual CPA (or eCPA as we like to call it) differs from a traditional CPA in two ways:
a) The amount of users attributed to a display campaign is much higher (in this case 6.5x higher) than the traffic that comes through clicks
b) The users who are exposed to a campaign convert much better (in this case 28% better) than users who have not been exposed
The joint effect of these two yield an actual cost of customer acquisition that may well be lower than more traditional direct response channels, such as paid search.
Results: Demographic effects
Since we (Bonnier Tidskrifter) have a large amount of demographic data on our users, we can also examine how different groups react to the campaign. Is the group that is exposed to the campaign the same group that later visits the advertisers site and finally converts? A lot of interesting information around creative execution and placement/media buys can be derived from this sort of analyses, but to save space I’ll let you draw your own conclusions from the following images.
Takeaways:
This is only one campaign, and a bunch of more tests need to be run to make firmer generalized conclusions, but there are three main findings from this campaign that I want to highlight:
1. Premium display can be an incredibly efficient direct response medium. It can be a cheap way of acquiring traffic and there is a bonus effect in that the traffic that is exposed will convert better.
2. The most important proxy in finding good inventory for direct response is looking at viewability. For an impression to have an effect it must be seen. Quality is key. There should be a flight towards it. Frankly, it surprises me that the market is not more actively looking for high quality impressions.
3. Understanding frequency for a certain category and subset of users can be really important. Why waste impressions on users where it has only marginal effect? Why skimp on impressions where the 10th impression may be decisive?
It would be really interesting to hear your thoughts on this, especially if you are a traditional direct response advertisers. Do let us know!
(This was also posted at www.btdm.se)
