You are on page 1of 3

Case: Technology Client – Sells laptops desktops tablets servers etc.

Objective: selecting 1000 out of 10000 customers for targeting


This case study is of Unsupervised learning, segmentation techniques can be used to solve this case
• Kmeans clustering technique
• Hierarchical clustering technique
The given case can be explained based on Kmeans. Since there are no data available from the case. I have assumed few of the variables as my
starting point. Having said that, it is for B2B so my end user would be a business entity and not direct customer. In this case RFM analysis can
be used to select the variables for clustering.
Variables which I assumed are: (all numeric)
1. Customer (B2B business entity)
2. Tenure as customer (taking it as >= 12 months, each customer has made purchase atleast once during their life cycle)
3. Recency of last transaction (based on RFM analysis)
4. No.of times transactions made in 24 months (based on RFM analysis)
5. Total revenue earned in the last 24 months (based on RFM analysis)
6. Total cost incurred in the last 24 months.
Since I don’t know what is my K, I assumed it to be k=3. (K can be found using Elbow method)
I can use Python for Kmeans
From sklearn.cluster import KMeans
Kmeans=KMeans(n_cluster=3, n_init=TBD, max_iter=TBD, random_State=TBD)
Case: Technology Client – Sells laptops desktops tablets servers etc.
Output: Tenure as Recency of last No. of times tranaction Total revenue earned Total cost incurred
Customer Customer transaction made in Last 24 months in last 24 months in last 24 months Cluster
1 12 0 30 1000 200 1
2 18 1 25 900 100 1
3 13 1 22 800 100 1
4 24 3 18 1000 200 1
5 35 4 12 750 100 2
6 40 5 11 600 70 2
7 31 7 9 500 80 3
8 26 9 5 200 60 3
9 29 2 21 900 100 1
10 33 2 23 650 50 2

Note: Above data is just for example, there is no relation between clusters and the data. It just for explanation.
cluster 1 has more high chance of taking subscription
cluster 2 has second most chance of taking subscription
cluster 3 has least chance of taking subscription
So what I want to point out here is
It can be possible that my 1000 customers might have come from cluster 1 only then in this case I will not look for cluster 2 and 3 customers
as I got my most profitable customers within cluster1.
if not then It might possible that I get 300 customers from cluster1 and 700 from cluster 2 so and so forth….
Case: Technology Client – Sells laptops desktops tablets servers etc.
How it will benefit to the Business?
Assuming that I have get 30% customers in cluster 1 then I can select top 1000 customers whom I want to target for
Marketing offer. This can be done by calculating profits and then giving them a ranking from highest to lowest.
Example:
Total
revenue Total cost
earned incurred
in last 24 in last 24
Cluster1 months months Profit Rank
1 1000 200 800 1
2 900 100 800 2
3 800 100 700 3
4 700 200 500 5
5 750 100 650 4

Note: Ranking based on profits (without weights). Above data is just for explanation.
Conclusion:
Once I select my most profitable customers, further I can drill down to calculate
Future expected revenue for next 12 months
Future expected cost for next 12 months

You might also like