Blogs

Morphing Banner Advertising - by Urban, Liberali, MacDonald, Bordley, Hauser

The authors doubled the CTR on banner ads using Morphing over the control group. The sample size was 116,168 unique CNET consumers with 451,524 banners. They used information about consumers behavior to infer one of four cognitive segments. They used that information to modify banner ads to match those cognitive segments.

They describe four segments based on two axes. There's an axis of impulsive versus deliberative which captures how hard somebody thinks. The second axis is analytic versus holistic. Analytic thinkers tear things apart into constituent parts, holistic thinkers do not. These segments are operationalized based on responses to questions like "I find that to adopt a careful, analytic approach to making decisions takes too long," "I rely on my first impressions," and  "I read the text carefully." The two axes are summarized into four segments. These cognitive segments are deliberative-holistic, deliberative-analytic, impulsive-analytic, and impulsive-holistic.

In a second test, they describe a number of segments in use at General Motors. There is a Collection segment that includes customers that are more than a year away from buying and are collecting information. There is a Comparison segment including customers that are less than a year away from buying, and a Commitment segment including customers who plan to purchase in the next three months, have collected all the information, and have visited a dealer.

They used clickstream data to match cognitive segments with web navigation styles using an in vitro (in laboratory) method of survey plus observing navigation. Designers "morphed" the banner ads based on the cognitive segments. The website learned which banner ads should be shown to different segments of people using a method driven by Gittins indices and those clickstreams. They then put these morphed banners in vivo (in the field) and observed the lift. The authors find that in the CNET study alone, morphing doubled banner CTR.

The concept of behavioral targeting isn't new. The idea of recording high level clickstream data and using it in ad networks isn't new. Ad re-targeting isn't new.

Extracting signals from a recency-frequency tables from the clickstream record is not particularly novel.

Cognitive segmentation is a concept that has been kicking around the digital intelligence industry for quite some time. It's not new, but could be executed more often.

The concept layering a theory-driven segmentation, based on cognitive style, using it to inform creative, mapping it against an owned media clicksteam, with an iterative learning algorithm to paid media, and publicly sharing the performance results, is new and noteworthy.

Could your organization mount such an attempt?

Perhaps.

The primary blocker isn't so much the technology - it's the way marketing organizations work.

Marketing institutions are generally not designed for deep empathy with consumers. They are designed around a sequence of repeatable activities.

A theory of the customer is required. GM has a theory. The authors of this paper also had a theory about thinking styles. It's not the case that there is a unified theory of a customer in many marketing departments. The existence of a theory, (not just an opinion or a hypothesis), is predictive of a empathy somewhere in the marketing structure.

Executing a survey to observe differences in customers, then exposing those customers in a website and watching how they engage, combines two functions that typically reside in two different departments. A research department traditionally fields, or commissions, market research. A usability department typically executes usability studies. These two activities, rightly or wrongly, are executed by very different people for very different purposes.

Then there's creative. The authors note that if they were to pay full cost for 75 banner creative, they would have had to pay $250,000. This is quite cost prohibitive. It's a lot of material to create, and, it's not as though banner creative can be varied using a machine-driven approach. There is also an issue with the goodness of fit between the segments selected and the creative that's matched. It can be easy to get wrong.

Then there's development. The final issue is how to implement clickstream analytics with the ad serving software, and programming the Gittens algorithm. These systems, in general, do not talk with each other. There's a very large red line between performance statistics and personalization. As a result, this is not regular, normal, development. Such solutions, if they were ever to be offered by adtech providers, would likely end up being black box solutions. The motivation for that is twofold: weak patent protection and easier account management. You save time by not having to explain Gittens every single time. In general, this is a very difficult lift for a majority of marketing technology organizations.

Very few digital marketing departments are organized in such a way to enable this workflow. Very few digital agencies are capable of such a lift. This workflow breaks the way that digital experiences are put together.

As a result, such an institution would need to be architected around the consumer, not the message. The physical technology of morphing isn't so hard. The social technology of morphing is.