Customer segments allow business users to understand the behaviour of the customers and make data driven strategies to engage the customer. Having a blanket marketing strategy can be helpful in the early stages but once the loyalty of customers is established they demand specific attention to their individual needs. Segmentation helps bridge that gap of ‘wanting more’, as it allows business to take a closer look at –
and these simple things can give a peek into their individual preferences and needs.
These metrics can help the business engage the customers in a more targeted fashion that is tailored to the customer’s needs and preferences, hence winning the trust of the customer.
By slicing and dicing the data, we can create hundreds of segments but Segmentation as an exercise has to be actionable. We look at the priorities of business and then map those requirements on to data using science and technology. This results in segments that are explainable to business and since they understand the segment definition it will help them make better decisions and in a more scientific manner rather than “trial and error” method.
We looked at the customer data for a global beverages chain and by using simple mathematical and statistical approach we brought about 40 Behaviour indicators ranging from customer’s food & drink preference to travel pattern and shopping preferences [as in Weekdays shopper on popular street or Mall] to Flying patterns.
All the above segments helped the business to understand their customers and target each segment with more personalized campaigns and communication.
A majority of practitioners think that Segmentation is a clustering problem and straight forward data crunching with all kinds of Clustering technique. However, we believe that Segmentation done to meet business objectives have more impact and add better value.
Segmentation is not just a statistical problem but a business use case to solve. We include both Clustering and business judgement and intuition in the form of features and then go about using a Clustering technique. This renders the resulting clusters more understandable for Business Executives, as they are the people who will make decisions based on the results.
A key aspect of making behaviour driven segmentation is to make sure that there is a mechanism to track the segments’ stability and movement. We review segments at periodic intervals to check the health and stability of segments. It is important to calibrate or refresh the segments once the metrics start showing the signs of degrading stability or high movements in the segment calibration.
In order to demonstrate a segmentation, we took a credit dataset from Kaggle. We created 3 clusters on the data. Here is a chart of centroids for each of the variable –
Then we took one cluster(cluster_id = 1) as reference and other 2 as label to see what are the drivers for customers in each group. We created a Decision Tree to understand the behavior of the customers in each group. We did the same for another cluster(cluster_id = 2). Below is the decision tree –
Then we took one cluster(cluster_id = 1) as reference and other 2 as label to see what are the drivers for customers in each group. We created a Decision Tree to understand the behavior of the customers in each group. We did the same for another cluster(cluster_id = 2). Below is the decision tree –
Cluster ID = 1
Clearly this is a group of Customers who make purchases as we are seeing that variable that are coming significant here are purchases, purchase transactions, purchase frequencies
Cluster ID = 2
This cluster is coming strongly as a Cash Advance cluster. Here the most powerful variables are Cash Advance amount, frequency of cash advance.
We did a PCA on the 18 variable set to get reduced feature set and key variables for segmentation –
From PC1 and PC2, it is visible that these 2 components are in 2 direction –
So we took these two components and plot our clusters to see the profiles of customers and we got 3 segment profiles –
Head of Consumer Business at Actify Data Labs
Related Posts
––
English
