Marc Meyer

Today at OMMA Global San Francisco, I attended a panel on applying so-called Behavioral information to advertising campaigns.   The sense of behavioral being discussed was a broad interpretation of the word, encompassing the use of many forms of data to target, be it behavior, intent, search keywords, even contextual information and social connectivity information.  A recurring theme was how does this get measured and how does this perform.

In the context of the panel session I’ll be leading tomorrow entitled “The Last Click Standing” about how to allocate resources to and measure effectiveness of different stages along the engagement path of an advertising campaign, it occurs to me that we may be falling prey to the a certain kind of limited thinking. 

There’s an old joke about someone looking for keys at night under a streetlamp.  When asked exactly where he dropped his keys, he responds he dropped them off somewhere in the dark.  When asked why he’s looking under the light, he responds that it is because that’s where he can see.

Similarly, we all recognize and I think budgets show that the big gains for big advertisers come from the Don Draper, Mad Men-type, intuitive advances in campaign creation for which we don’t have easy on-line tools for measurement.  A lot of branding campaigns fall into that category.  Even a lot of on-line campaigns (the apple ads on the NYT come to mind), are like that.  I may not watch them all the way through, I may not click, but I will buy the next Apple gadget that comes out.

Because we don’t have tools for that, we optimize the things we do have numbers for and we jigger things so we get numbers out of advertising vehicles that didn’t use to produce them by including “calls to action” whether they’re appropriate or not.

What statistical tools exist to help us look at the real bottom-line effects of making changes in marketing?  I can imagine uses of machine learning to create regressions by introducing small variations in the campaign parameters.  These would be useful both for on-line as well as off-line initiatives.  Yahoo is doing something like that with their Computational Advertising initiative.

Machine directed A/B testing is established for keyword optimization, creative placement and selection.   Can the same techniques be used more widely to improve marketing allocations on- and off-line without requiring that all elements be directly measurable?