Predicting customer lifetime value

Patent metadata

Field	Value
Publication number	US-10290011-B2
Application number	US-201514832748-A
Country	US
Kind code	B2
Filing date	Aug 21, 2015
Priority date	Feb 23, 2015
Publication date	May 14, 2019
Grant date	May 14, 2019

Abstract

Method(s) and System(s) for predicting Customer Lifetime Value (CLV) based on segment level churn includes segmenting the customers into multiple segments based on weighted RFM scores associated with data within a dataset. The data is representative of purchasing behavior of customers over a predefined time period. The segmenting is performed such that customers with similar and close weighted RFM scores are placed in one segment. Further, the method includes computing a churn value for each of the customer segments based on the buying behavior of the customers within each segment. The churn value is associated with transaction characteristics associated with customers corresponding to the data in each segment. Expected lifetime period in years for the customers is then predicted from the calculated segment level chum values. Thereafter, CLV, that indicates profitability associated with customers, is predicted for each customer based on their expected lifetime value in years.

First claim

We claim:

1. A method for predicting Customer Lifetime Value (CLV), the method comprising:
segmenting, by a hardware processor, a dataset into a plurality of segments based on weighted Recency Frequency and Margin (RFM) scores, wherein the dataset includes data representative of purchasing behavior of customers over a predefined time period, wherein the weighted RFM scores are associated with the data of the dataset, and wherein the data corresponding to the customers with similar weighted RFM scores is placed in one segment and data corresponding to the customers with dissimilar weighted RFM score is placed in individual segments;
computing, by the hardware processor, a churn value for one or more segments from amongst the plurality of segments based on the buying behavior of customers within each of the plurality of segments, wherein the churn value is associated with transaction characteristics associated with the customers corresponding to the data in each segment; and
predicting, by the hardware processor, a CLV for a customer of each segment based on an expected lifetime period of the customer, wherein predicting the CLV comprises:
computing expected lifetime period for the customer of a segment based on the churn value, wherein the expected lifetime period corresponds to a time period for which the customer is expected to perform transactions with an organization;
performing a comparison of the expected lifetime period with a threshold and determine the one or more segments that include customers with the expected lifetime period greater than the threshold and the one or more segments that include customers with the expected lifetime period lesser than the threshold; and
estimating the CLV for each customer of each segment based on whether the customer has the expected lifetime period greater than the threshold,
wherein for the customer with the expected lifetime period lesser than the threshold, a first computation technique leveraging a calculated margin value, an interest rate for discounted cash flow is used to predict the CLV,
wherein for the customer with the expected lifetime period greater than the threshold, a second computation technique leveraging a calculated margin, a retention rate and an interest rate for discounted cash flow is used to predict the CLV,

wherein the CLV is indicative of association of the customers with the organization corresponding to each segment.

2. The method as claimed in claim 1 further comprising storing the CLV value for each customer of each segment in a Hbase database.

3. The method as claimed as claim 1 , wherein the weighted RFM scores are computed by performing data analysis on the dataset based on Recency Frequency and Margin (RFM) parameters associated with the data corresponding to the customers.

4. The method as claimed in claim 3 , wherein the data analysis comprises:
applying data cleansing on the dataset to eliminate at least one of incomplete data and corrupted data; and
analyzing the dataset based on Recency Frequency and Margin (RFM) to generate the weighted RFM scores.

5. The method as claimed in claim 1 , wherein the data set is segmented by utilizing distributive processing capability of MapReduce technique.

6. The method as claimed in claim 1 , wherein the churn value is computed based on an exponential moving average technique by assigning more weightage to a churn trend in a recent year than the remaining years of the predefined time period.

7. A Data Analysis System (DAS) communicatively coupled to a database and a retailer data system for predicting Customer Lifetime Value, the DAS comprising:
a hardware processor;
a data collection module coupled to the hardware processor, wherein the data collection module is configured to collect data related to transactions conducted by customers from the retailer data system through a network and collate the data to obtain a structured format for the collected data;
an analysis module coupled to the hardware processor, wherein the analysis module is configured to:
segment the dataset into a plurality of segments based on weighted Recency Frequency and Margin (RFM) scores, wherein the dataset includes data representative of purchasing behavior of customers over a predefined time period, wherein the weighted RFM scores are associated with the data of the dataset, and wherein the data corresponding to the customers with similar weighted scores is placed in one segment and data corresponding to the customers with dissimilar weighted RFM score is placed in individual segments; and
compute a churn value for one or more segments from amongst the plurality of segments based on the buying behavior of the customers within each segment, wherein the churn value is associated with transaction characteristics associated with the customers corresponding to the data in each segment; and

a prediction module coupled to the hardware processor, wherein the prediction module is configured to predict a CLV for each segment based on an expected lifetime in years of each customer, wherein the prediction module is further configured to:
compute an expected lifetime period for the customer of a segment based on the churn value, wherein the expected lifetime period corresponds to a time period for which the customer is expected to perform transactions with the organization;
perform a comparison of the expected lifetime period with a threshold and determine the one or more segments that include customers with the expected lifetime period greater than the threshold and the one or more segments that include customers with the expected lifetime period lesser than the threshold; and
estimate the CLV for each customer of each segment based on whether the customer has the expected lifetime period greater than the threshold,
wherein for the customer with the expected lifetime period lesser than the threshold, a first computation technique leveraging a calculated margin value, an interest rate for discounted cash flow is used to predict the CLV,
wherein for the customer with the expected lifetime period greater than the threshold, a second computation technique leveraging a calculated margin, a retention rate and an interest rate for discounted cash flow is used to predict the CLV,

wherein the CLV is indicative of association of the customers with the organization corresponding to each segment.

8. The DAS as claimed in claim 7 , further to store the CLV value for each of the customers in a Hbase database.

9. The DAS as claimed in claim 7 , wherein the analysis module is further configured to:
apply data cleansing on the dataset to eliminate at least one of incomplete data and corrupted data; and
analyze the dataset based on Recency Frequency and Margin (RFM) to generate the weighted RFM scores.

10. The DAS as claimed in claim 7 , wherein the analysis module is further configured to:
segment the dataset into the plurality of segments by utilizing MapReduce technique; and
compute the churn value based on an exponential moving average technique by assigning more weightage to a churn trend in a recent year than the remaining years of the predefined time period.

11. A non-transitory computer-readable medium comprising instructions for predicting Customer Lifetime Value (CLV) executable by a hardware processor resource to:
segment the customers into a plurality of segments based on weighted Recency Frequency and Margin (RFM) scores, wherein the dataset includes data representative of purchasing behavior of customers over a predefined time period, wherein the weighted RFM scores are associated with the data of the dataset, and wherein the data corresponding to the customers with similar weighted RFM sc

Assignees

• Tata Consultancy Services Ltd

Inventors

• MANOHARAN Vignesh
• Atchayalingam Lakshmikanthan Parthibarajan
• Palani Bommu Madhukumar

Classifications

• G06Q30/0202 Primary

  Market predictions or forecasting for commercial activities · CPC title