What's your problem? Segmentation

One of the first questions an organization should answer before committing to data science is, "What's our problem?" Those problems can be anywhere: finance, marketing, IT, operations, business development, or product development. This is a first-order prioritization problem.

The next set of posts summarizes some of the most common problems that data scientists address, why data science would be involved, opportunities for integrating data science, and how the results are commonly used.

"Who are our clients or customers?"

Philip Kotler, one of the all-time great marketers, defined marketing as "an administrative and social process through which individuals and groups obtain what they need and desire by the generation, offering and exchange of valuable products with their equals." Within marketing, Kotler defined the target market as consisting "of a set of buyers who have common needs and/or characteristics to those that the company or organization decides to serve."

It can be surprising to learn when organizations have been gathering data for years but don't understand who their clients or customers are or their behavior that leads to acquiring new customers, keeping existing customers, or growing customer relationships. Kotler's definition of the target market could also be called segmentation. Data scientists can refer to this as clustering. Data scientists use algorithms--of which there are many--against the available data to define those classifications, or groups. Clustering results in groupings that highlight the similarities of the individuals within the group while differentiating between the groups. Clustering is foundational in marketing. Clustering is an input into other marketing, such as:

• What types of specialists are in this group?
• Which channels should we target for this group?
• What's the purchasing behavior of clients or customers in these groups?
• Do any of these group descriptions inform the language we'd use to drive engagement or purchase?
• After a probability model has been developed for these individuals what's the cluster average probability of doing what we want?
• Is there a geographical similarity or difference in or between these groups we could take advantage of?
• Do these groups indicate future targets for our products or services?

The benefits of clustering include removing much of the overt bias of human experience and/or heuristics (rules of thumb)--"Let's start with this, and then split by this, and then split again…"--and developing richer, more valuable data-driven segmentation built by letting the data tell the story. While this could seem antagonistic to marketing or business experience, it's not meant to be. The development of data-driven segmentation, like many aspects of data science, falls under the analytical 80/20 rule, paraphrased from the renowned baseball statistician Bill James: "80% of analysis confirms what we already believe while the other 20% surprises us." Using data to confirm our hunches or heuristics is incredibly powerful. And while a small percentage of time we can be surprised positively or negatively by those results, challenging our world view of how things work is how the best change occurs. 

Traditional marketing can rely on qualitative data to develop segmentation, sometimes summarized in group descriptions as personas. Much of this information can be found through 3rd party aggregation services and through surveys and market research. This type of data can be very expensive to acquire and fraught with what I call "mis-generalization representation". Would you, as a business leader, believe that 8 neurologists in Chicago represent the beliefs of all neurologists in the United States?

The combination of qualitative and quantitative data to develop more efficient, effective, and detailed profiles is an integration opportunity for data science and marketing or market research. The summary, qualitative data can be rich input for the development of hypotheses to be investigated in the quantitative data of transactions and demographics, for example. Many times the qualitative, summary data of market research focuses on the why. The why can also be crudely summarized as "how clients or customers feel." The quantitative data could be summarized as the behavior, or the "will" data. The "will" data is often the input into predictive models that lead to the probabilities that an individual will do what we think, expect, or want, such as "Will this client/customer buy our next product?"

Jumping into data science without first determining what the problem or problems are is putting the cart before the horse. While data science can help throughout the organization to make smarter, data-driven decisions, issue prioritization should be the first step. Data science can bring a new lens to segmentation and driving business success through understanding. If you're interested in creating deeper, richer segmentation for your business, reach out to me at sam.johnson@bluejacketsol.com. If you've already made data science a successful tool for segmentation I'd love to hear about your experiences, too.