Standard Segmentation versus Automated Segmentation

Quick Hits

  • Standard segmentation requires a high level of knowledge and foresight to predict audience value and will often miss the mark.
  • Automated segmentation should uncover uncommon commonalities between attributes, specifically ones that normally would not be naturally discerned.
  • Automated segmentation should not operate in a vacuum and will often require validation. Because an algorithm picks it, doesn’t necessarily make it valuable.

There has been an underlying truth about traditional segmentation that very few people ever really want to talk about. To do it well is extremely difficult and the true process is overly complex. On the surface, we all feel that it’s easy to identify, build, activate, and provide insights on segments, in fact many of us do it daily. What we often fail to realize is that the vast majority of us are doing it inefficiently and with limited end value.

Standard segmentation in its simplest form is the combination of attributes through various logic statements with the intent of defining a group of IDs as an audience (I realize this is an oversimplification). The true art of audience segmentation comes in the ability to construct audiences that have value (analytic or monetary). But, the time it takes to evaluate, construct, and analyze each and every segment before the value proposition can be witnessed requires resources that many organizations can’t afford to allocate or aren’t available at all. What most applications have really created are systems of trial and error. Manually build a segment, compare it to something, test it in the market, extract value or return to sender. This cycle creates a long process in which value may never be derived. In a previous endeavor I worked on, typical clients would have on average over 2,000 segments in which less than 5% were actually used. This creates high amounts of waste in processing, people, and storage as well as adds a tremendous amount of noise to the system.

Automated segmentation, on the surface, eliminates much of the human element typically at the most tedious of stages. Rather than manually coupling attributes into an audience, automated segmentation applies algorithms to datasets as a way to find distinct outcomes. The same dataset can be dynamically evaluated in numerous ways without any human involvement thus dynamically producing a multitude of segments. Certain algorithms can find categorical or numerical commonality within the attributes of a dataset and return a series of clusters or cohorts based on the construction of the algorithm. This can have a dramatic operational impact since it reduces the once manual effort of constructing audiences and layers on additional value by finding uncommon commonalities that a person may never think of.

As a wise ghost once said, “If you build it, he will come.” The same can not be said for automated segmentation. Humans are still required to extract the full value out of derived audience clusters and there is still a need for skill in understanding where the value line exists. Automated algorithms will dynamically find countless clusters within a dataset and do so blindly. For the time being, these audiences need to be constantly validated so the algorithms can be updated and improved.

Every marketer needs some form of automated segmentation in their arsenal. Without it, they will spend increasingly more for less, as the need for audience derived value and insights continues to increase over the coming quarters. And, for the time being, they should supplement the technology with skills necessary to fulfill modern audience marketing strategies.