Dynamic Online Advertising, Could It Work?
I caught an interesting article in The New York Times today that spurred a healthy debate in our department. Two West-coast companies, Adisn and Tumri, are extending their ad serving model beyond targeting to include a dynamic presentation of the ad unit.
In short, they are both trying to serve a custom design ad unit to a consumer based on data as well as the social Web in the case of Adisn.
It is a very interesting solution, and out of our internal discussion a few points rose to the top.
- While both companies are trying to further target ad units based on research, trying to make more appealing ads based on color or font will likely sacrifice some brand guideline.
- Neither company talks about their sample size for coming to logical conclusions, but is it possible to test enough variations of an ad and auto-optimize quickly enough to make the technology cost-effective?
- And, although both companies are collecting data on how the ad performs, neither company seems to be taking in account the context in which the ad unit is displayed.
The last point was really important for me. In my experience as the Online Design Manager (basically creative director) at The Milwaukee Journal Sentinel I would see some ads perform really well in context to the site design because a bright color contrasted well against the site. Or a particular font stood out because it was different from the content set in Arial.
If they are not tracking that information I really think they are missing an important factor in their equation. Is the ad above the fold? Below? What about frequency? Maybe none of this could be added to the article, but I also saw little about that information collection on their sites.
It seems like a great idea in theory, especially trying to continue to expand content targeting based on the social Web. But without understanding the ad unit in context of the placement it seems a little less ground breaking.
6 Comments
Mijo
December 4, 2008To address your last point: Context plays a huge part of ADISN’s dynamic ad unit. The Natural Language Processing component of the system detects the context of the page the ad is served on. This data directly influences which components are shown for a particular dynamic ad unit. So not only does the ad unit optimize around particular component combinations: taglines, color, component order,etc but also it displays components (let’s use different products from the same brand as an example) relevant to a particular demographic. That is, it will target the component ( or set of components ) to the demographic most likely to be associated with the context of a particular page.
djenders
December 4, 2008In short, what canvas does the creative sit on? It makes a world of different if a black ad is on a site with a black background too, where it blends in.
So when I say context, I mean the context in which the ad unit has been placed. Was the ad unit itself on the New York Times site, with a white background and serif fonts. Was it more successful because it had an orange background with sans-serif fonts that stood out within that site? I don’t think the algorithm is tracking enough, it really needs to track more than the actual page content and the dynamic ad design – it needs to extend one layer further to track the design of the actual site it is on.
Mijo
December 4, 2008As far as placement within the page, what better way is there to figure out which font and color work best than to let the ads performance dictate those attributes over a statistically significant sample set?
djenders
December 4, 2008Well I do think sample size needs to be addressed. I’m curious what the threshold is for that with Adisn. Other than that, to my previous point, I want to know what site it was on and the design elements that make up that site. Make sense?
Mijo
December 4, 2008Makes sense, I misread the posting in my first response. Aren’t design elements irrelevant when its all about what the “majority” of users prefer to see? Which font works with what other font are subjective, so why not try a range and see which works best according to the people who actually click? I agree that optimizing around each particular domain is crucial but that’s not a hard problem…
djenders
December 6, 2008I think it’s a very valid point when you say it is what the “majority of users prefer”. If you can take that to a site or demographic level, then it would be pretty damn effective.