Product Marketing meets Data Science: Customer Segmentation with Cluster Analysis
Customer segmentation is a core task for Product Marketers. Applying cluster analysis on a customer database enables PMs to extract the precise temperament of the company’s Customer Personas.
In this post, I write a case study as if I was a Product Marketer for a B2B business with retail stores as clients. Conducting cluster analysis, I found 4 segments where different Marketing tactics should be employed.
The Business Problem
Let's say we work for this B2B company that sells an SKU tracking platform to retailers enabling them to have a clear view of what they sell. Our developers have made a new superstar feature that goes one step further and allows retailers to give loyalty cards to their clients so that they know the personalized aspect of their client’s purchase behavior. Our job is to create a go-to-market plan in order to upsell this feature to our existing customers. In this plan, we will need to build a value proposition, a USP prioritization, and the messaging, all of which directly relate to our customer segments.
Why solve it with Data Analysis?
There are different customer intelligence gathering methodologies. Customer interviews are strong in acquiring qualitative insight, customer Advisory Boards give trusted feedback continuously and in the long-term, and customer surveys allow for targeted questions to be answered. Yet, many times we use these techniques when in many cases there is a superior alternative: the use of customer databases. Recording tens of variables per customer and storing this info into a database can be a huge blessing. If there is a database, we can perform all kinds of multivariate analyses and harvest decision-supporting insights. What is more, by applying data analysis to segment our customers we win in time because we skip hours of customer interviews and accuracy because we end up with a defined differentiation between our segments.
The Cluster Analysis
To exemplify the case of our B2B business where we work as PMs, I gathered a hypothetical database of 72 retail stores from Tilburg University’s “Intro to Research in Marketing” class. The database has variables per retail store such as its size, its revenue level, and demographic information for its clients. Trying to segment our retailers, I perform the cluster analysis methodology.
The first step in the process is to conduct a hierarchical agglomerative clustering (don’t let the terminology intimidate you, check my Github for the code). This analysis provides us with a certain graph named Scree Plot which hints at the number of clusters that we could use in the following step. Finally, we perform a k-means cluster analysis (where k = the number of clusters as indicated by the Scree Plot) to get our retailers categorized.
The Customer Segments
After running the analysis we get 4 types of retailers with specific attributes.
Cluster 1, the Specialty Store: Mid-sized mid-revenue retail stores that sell to high-income elders with big households in semi-urban areas.
Cluster 2, the Student Discount Store: Mid-sized mid-revenue retail stores that sell to low-income educated young people who stay in small houses in semi-urban areas.
Cluster 3, the Kiosk at the Corner: Small-sized low-revenue retail stores that sell to medium-income non-educated middle-aged people in city centers.
Cluster 4, the Supermarket of the Village: Big-sized high-revenue retail stores that sell to medium/high-income middle-aged people who stay in mid-sized houses in rural areas.
The Product Marketing follow-up:
Now that we have the profile of the customers we can filter our go-to-market plan with the knowledge we just got. We can craft value propositions per segment, come up with those USPs that stimulate each segment most, and adjust our messaging accordingly.
For example, for Cluster 1, the Specialty Stores, we can assume that since they serve a niche market, let's say the book industry, creating long-term loyal customers can be a competitive advantage that can claim hard-fought market share. In a landscape reigned by Amazon and other major eCom Businesses that offer lower prices, loyalty cards in mid-sized Specialty Stores can be seen as an effort to add brand value with the inclusion of valuable perks that can drive engagement. Therefore, brand value and competitive edge can be used as the primary USP.
With regards to Cluster 2, the Student Discount Stores, we can brief our Account Managers so that when they pitch our new feature they can raise the card of the non-loyal & price-elastic nature of millenials and articulate how a loyalty program of further discounts, which we enable, could raise the visits of their clients and drive their business goals.
Cluster 3, the Kiosk at the Corner, could potentially be a customer segment that we don’t want to serve and might think it isn’t worth it to pursue upsells. Yet, knowing they have low revenue levels, we can attempt to use a low-price USP. Looking at the three different revenue levels of our customers we can design a good-better-best price scheme, one pricing package for every revenue-tier, and decide to charge per number of loyalty cards used, not a monthly subscription or another high-starting-barrier price scheme. With the low-cost option in mind, the Kiosk at the Corner might decide to actualize the upsell after all.
As for Cluster 4, the Supermarket in the Village, its management could be persuaded by the fact that our feature allows them to move towards digital transformation. Nuanced insight and slight changes in big organizations have bigger results in absolute terms for profitability. Hence, calculating the customer lifetime value and diagnosing customer retention can satisfy a major information need of our client. When framing the message for them, it might be effective to use the USP of technology diffusion in the store’s daily activities.
All in all, every cluster analysis that produces customer segmentation can offer a variety of insights dependent on the case. Combined with the Product Marketer’s expertise and knowledge of the product and the market, the go-to-market plan can be enriched with those few details that make a big difference.
