Why and how do you segment Customers ?
Customer segmentation is the process of dividing customers into groups based on common characteristics. By creating segments, businesses can tailor their marketing efforts, product offerings, and customer service experiences to better meet the specific needs of each segment.
It's a pretty generic definition, but the key phrase here is "common characteristics." Putting people in age groups, income brackets, and spending brackets are all forms of segmentation, but what makes your analysis useful is understanding why you are segmenting in the first place.
I like think of it in three broad categories;
Case 1: Strategy to segmentation
Case 2 : Segmentation to Strategy
Case 3: Segments as variables for machine learning
Strategy to segmentation
Say the Head of Marketing walks into your office, and your conversation goes like this?
HOM: Data human
You: Yes
My team has come up with a brilliant customer reactivation strategy, and we want to target Millennials who are upper middle class and have done over $100,000 in transaction volume but aren’t currently inactive.
You: Sure thing boss, I will get you the list then we can start looking at a campaign monitoring strategy
You are essentially an SQL, Excel, or whatever jockey your data is. Its idea first,code later
You get a group of customers who meet their conditions and pass it to them in Excel or maybe upload it to their CRM.
The end
You can’t exactly say if their strategy was data-driven or just built on ✨Vibes✨ and experience.
Segmentation to Strategy
Let’s say management has the head of customer experience in a chokehold. The board has been looking at your churn and retention dashboard, and they see the organization’s churn rate growing every month, and they are not happy.
The Head of CX comes into your office to ask for help and It goes like this
HCX : Data human how do think we can increase customer retention
You: "I think we need to figure out who might leave and how much effort you will be willing to put towards reactivating customers." We also need to be sure there isn’t low-hanging fruit—maybe the churn is concentrated in a certain region or we are losing a certain age group.
HCX :This seems like a lot, how are we going to do it
you: I am going to need your team’s help me to decide the size of segment bins and the strategies for each segment
HCX: Sure thing, we are happy to help.
Starting with just transaction data at first, you decide the recency of purchase, the frequency of purchase, and the transaction volume for each customer, and you create bins for each category.
The CX team makes the decision on the size of the bins in your segments and what to do with different segments, then they act on it.
Segmentation based on R-F-M analysis often serves as the foundational strategy, while other demographic and geographical segments are used as additional factors for more specific product offers.
Pros :
a. The CX team is carried along at every stage of the project, it a great way to build trust and confidence in analytics in your organisation
b. It can serve as a great foundation for your more complex projects, the business people are going to provide so many possible feature pointers for your machine learning models
Cons:
a. Too many buckets can make it annoying very fast
b. Opinions can change constantly and people may not agree
Segments as variables for Machine learning .
My dear data human, remember when you told the Head of CX ,
“we need to figure out who who might leave and and how much effort you will be willing to put towards reactivating customers”
A few paragraphs ago you wanted to present the data then let the domain experts make decisions, now you want to do a churn prediction model and a customer lifetime value prediction model and all sorts of things.
This leads to something I said in my introductory article, There no label’s and features in the real world you decide what should go into your model.
Transaction data won’t magically become features, you decide what features have be created and often time they results in categorical variables which are basically segments or numerical variable that are precursors to segments. Demographic and geographic data are often a bit too detailed and often have to be segmented.
What did we learn?
Why and how we segment is dependent on the goal we need to reach
Are you trying to get data to implement a strategy?
Are you trying to create a strategy from your data?
Do you have a specific business metric you need to predict?
If your answer to any of these questions is yes, now you know what to do.
What’s Up next?
Do you think I missed out on other reasons? How is it different in your experience?
I will expand on case 2 and case 3 in upcoming articles that will have snippets from and links to notebooks and Pbix files as needed.
