Case Study: L101
-Fiber bits
Business Objective:
“Fiber Bits” is an internet service provider company. They
are in the market from last 10 years. They lost almost 42% of their customers
in last 3 years. Some new customers joined during the same time period. But, the
company is concerned about the high attrition rate in the customer base. They
want to get an idea on what are the main factors that lead to customer
attrition. To reduce the attrition rate,
they have introduced vouchers and other benefits program. The objective is to
identify the customers who are most likely to quit in next 2 years and try to retain
them by offering free vouchers and benefits.
Problem Statement and Scope of Model
The company has collected around 10000 customer historical
data from last three years. We need to build a model that identifies the
customers who are most likely to leave. We need to quantify the chance of
attrition for each of the customer. The model will be used on the active
customers. The free vouchers and benefits will be given to customers with
higher probability to attrite in next three years. For example below are two
customers, who is most likely to leave. Which customer should we try to retain
by sending free vouchers
Customer
|
Cust1
|
Cust2
|
income
|
2586
|
1581
|
months_on_network
|
75
|
35
|
Num_complaints
|
4
|
3
|
number_plan_changes
|
1
|
2
|
relocated
|
1
|
0
|
monthly_bill
|
121
|
133
|
technical_issues_per_month
|
4
|
1
|
Speed_test_result
|
85%
|
95%
|
Data:
The data
consists of nearly 10,000 customers. Below are the list of variables and their
descriptions.
Variable name
|
Description
|
active_cust
|
The Dependent variable
Active-1 (Customer
Attrition=No)
Not Active – 0 (Customer Attrition=Yes)
|
income
|
Estimated monthly income
|
months_on_network
|
Months on network (Moths from the service start day))
|
Num_complaints
|
Total complaints till now
|
number_plan_changes
|
Number of times the service plan is changed
|
relocated
|
1- Relocated
0 – Not relocated
|
monthly_bill
|
Average monthly bill
|
technical_issues_per_month
|
Average monthly bill
|
Speed_test_result
|
Percent of (Actual speed/Promised speed)
|
Download data from
below link
Analysis Steps
·
Data validation
·
Data cleaning
·
Identification of analysis technique
·
Building predictive model
·
Removing multicolliniarity
·
Final model
·
Calculating probabilities for cust-1 &
cust-2
·
Final observations and inferences
·
Documentation of the approach, codes and results
Discussion forum
·
Facebook:
·
Blog :
·
Slide share
References:
·
Case study id: Case Study: L101 -Fiber bits
