Jonathan has helped optimize...

« Launch Interviews on Yieldbot | Main | The Market Value of Supply Side Data »

Dynamic Segmentation (La Nouvelle Vague)

I’ve always loved segmentation. One of the keys to the success of my first web business and the number one reason we survived the dotcom bubble was that we segmented our email list like crazy. Every supplement purchased qualified you for a segment. We knew the categories you were interested in and what products within those categories. We also knew what days worked best to send the emails to in each segment. Weight loss was sent on Mondays, bodybuilding on Fridays. This built an insanely successful email program despite the fact I knew nothing about online marketing back then. Relevantly segmenting, testing and targeted emails just made business sense -- and the web made it easy.

But Email is generally a static environment. The rest of the web is not. In the 12 years since that first segmentation strategy I have learned quite a bit about segmentation. One of the most important things I learned about is the dynamic nature of segments.

Segments are malleable. They can be shaped based on any number of factors and those shapes can change based on look-back or realtime. They can also have sub-segments -- literally segments of segments. As every Search marketer knows the success of segmentation is finding the optimal blend variables from which to form your segment. Most targeting success is predicated on how well you can do this. How well you can do this is based on the data you are using to support making your decisions. This is the gravitational force of segmentation optimization.

In my experience many SEMs fail because it is incredibly difficult to deal with all the segment variables in a dynamic data environment and a dynamic marketplace. It’s harder still when your success is only measured against your own metrics and not the metrics of others. I’ve never been a fan of benchmarks (most people use these to create “good enough” goals) but hey, at least it’s something to measure against.

That brings us to my current segmentation obsession, Display.

The emerging display ecosystem is being built on the back of data from any number of sources and like Search this data is being used to create segments. In fact, Display is opening up data and pricing in a way Search never did and never will. With that comes a giant market risk. Sometimes I think the whole “display new wave” is a crazy idea but like Jean-Paul Belmond in Goddard’s Breathless there is no turning back. The current environment was killing display and when you ain’t got nothing you got nothin’ to lose.

How does it feel? The entire remarketing / cookie market is based on segmentation as a proxy for audience buying. Display now uses words like targeting, addressable, personal, one-to-one and realtime. All these are based on one very simple idea - creating a segment to target more relevant advertising against. Still, collecting data is actually a very small part of ensuring success. Optimizing segments is where dreams are found and lost.

As hard as the opaque market of Search is, it is self-contained. In many respects that makes optimization easier than in Display. In my opinion (and there will be more than a few start-ups that disagree with me) truly optimizing segments (to the point of diminishing returns) across an ad exchange or a network is a Google level problem. Much easier is looking at segmentation at another self-contained environment  - the site. This is one key reason why our analytics focus at Yieldbot begins with the site data and our targeting focus ends at the site visitor.

Make no mistake, brand or performance, there will continue to be less and less media spend on irrelevance and a more and more spending on relevance. Segments are created for the simple purpose of delivering relevance. Every web business, be it advertising or publishing, needs to have more than data. The dynamic nature of the web requires systems that can both optimize and make actionable the data in ways as dynamic as the web itself. Welcome to the new wave.


TrackBack URL for this entry:

Listed below are links to weblogs that reference Dynamic Segmentation (La Nouvelle Vague):


Regarding dynamic segmentation, have a look at the description of Profile Clusters and Predictive Targeting, the methodologies employed to "discover" the right segmentation criteria currently used in online display, both for branding and performance campaigns, at

So what do you need to be doing to truly optimizing segments to the point of diminishing returns? Adjusting bid levels, spend levels frequency caps, creative messaging? Anything else?

And, do you believe that the dynamic segmentation optimization must be done in a unified fashion across the display experience and the site experience?

Thanks Paulo. I'll check that out.

Greg- the answer is really that every segment requires different strategies & executions. So all the things you mentioned and more are likely required to optimize and you need the dials & switches to execute on those things.

This is new stuff so there are and sholdn't be rules but if we have learned from Search I would say inventory is the overriding factor of influence. The more static you can make this element (think of the generally consistent performance nature of keyword inventory)the better chance you have to reach that point of diminishing returns.

Interesting thoughts. There exists, of course, the ability to feed dynamic content through banner ads. This doesn't solve the problem, of course, because you still need to develop optimization algorithms to do the work of picking and choosing which content to feed where. But I wonder if display has awhile to go before it succumbs to math. As you point out, SEM is much simpler than display. SEM is text, which creatively has a very limited set of possible elements (i.e., the letters of the alphabet). Whereas display is a combination of text, images, color, textures, geometry, motion, and so forth. If a segmentation algorithm is sophisticated enough to master a display campaign, why couldn't an algorithm create the "perfect movie" (movies being composed of mostly the same elements as display ads)? Perhaps you see what I'm getting at -- math-driven creative may some day come to dominate advertising (who knows what the future will bring), but it's a ways off yet, so perhaps we shouldn't expect display to behave like search (or email). I'm not saying that display can't be segmented AT ALL, I'm just saying that a purely mathematical solution could prove more elusive than everyone thinks. I think there are a lot more tangible inefficiences to wring out of the market first.

The comments to this entry are closed.