CLV PredictionBy

Bita Houshmand

Advisor:Dr. HosseinAlizadeh

AUG 2014

2Head of contents

Introduction CLV Definition CLV Prediction Methods & Techniques Variables incorporated into CLV models Mathematical Models RFM & Related Models MK_SVR Framework Using Longitudinal Data Performance Metrics CLV Applications Conclusion

3Introduction

CRM is not optional in today’s Business environment

At the core of CRM is Development long-term Relationships Maintenance long-term relationships

Differentiating Profitable and Non-profitable customers In fact, CLV fundamentally measures the financial return of the customer

and the firm relationship (Kumar, Peterson & Leon, 2010)

Using CLV, Firms are capable of

With Customers

Investment on profitable customers

Customer Segmentation

Customized Promotion

New Product Offer

4CLV Definition

Author Definition(Kumar et al., 2004) The sum of cumulated cash flows—discounted using the weighted average cost

of capital—of a customer over his or her entire lifetime with the firm P.61(Jain & Singh, 2002) The net profit or loss to the firm from a customer over the entire life of

transactions of that customer with the firm (Bauer et al., 2003) The profit streams of a customer across the entire customer life cycle. P.333

(Berger & Nasr Bechwati, 2001) The excess of a customer's revenues over time over the company costs of

attracting, selling, and servicing that customer P.49

(Sargeant, 2001) The total net contribution that a customer generates during his/her lifetime on a house-list P.28

(Hoekstra & Huizingh, 1999) LTV is the total value of direct contributions and indirect contributions to overhead and profit of an individual customer during the entire customer life cycle that is from the start of the relationship until its projected ending. P.266

First Group: Emphasize on Profit

5CLV Definition

Author Definition

(Payne & Holt, 2001)The net present value of the future profit flow over a customer’s lifetime P.167

(Venkatesan et al., 2007) The net present value of long-term cash flows from a customer

(Gloy et al., 1997) The net present value of cash flows a customer is expected to generate for a firm over the length of The customer's relationship with the firm P.336

(Peppers & Rogers, 2005) The net present value of the future stream of cash flows a company expects to generate from the customer

(Stahl et al., 2003)The net present value of future cash flows generated by the firm’s assets, discounted at an appropriate interest rate and adjusted for inflation and risk P.267

Second Group: Emphasize on Net Present Value

6CLV Definition

Author Definition(Pfeifer et al., 2005) The present value of the future cash flows attributed to the customer

relationship P.17(Hidalgo et al., 2007) The present value of all future earnings a firm may generate from a customer

P.695(Chen & Fan, 2013) The Present value of future profits obtained from a customer P.123

(Benoit & Poel, 2009) Present value of the future cash flows associated with a customer P.10475

(Dwyer, 1997) The customer’s present value of the expected benefits less the burdens

Third Group: Emphasize on Present Value

7Variables incorporated into CLV Models

Independent Variables (Uncontrollable) Customer Demography Purchase Behavior

Controlled Variables Marketing Intervention Customer-Firm Interaction

(Khan, 2009; Gupta et al., 2009; Kumar et al., 2011; Khajevand et al., 2011)

Number of household members

Nationality

Social Class

Customer Churn

RFM Variables

Customized Promotion

Customer Status (Active, Inactive, Potential)

Examples

8CLV Prediction Methods & Techniques

One-Step Methods (Relationship-Level Models) Linear Regression Quantile Regression Markov Chain Mathematical

Two-Step Methods (Service-Level Models) Pareto/NBD Hierarchy Bays RFM MK-SVR

9Prediction Models Comparison

Providing more insight into the factors that drive customer value

Advantage- Amount of modeling required and the often poorer predictions- Modeling rare event

Disadvantage

- Less Amount of Modeling - CLV estimation is much more Straightforward-Prediction accuracy is higher

Advantage

Due to aggregation, insight into the factors that drive consumer profitability is limited

Disadvantage

Two-Step Models One-Step Models

(Benoit & Poel, 2009; Vankatesan et al., 2010; Donkers, Verhoef & Jong; 2012)

10Prediction Models Comparison

Pareto/NBD

Powerful Predictor in non-contractual

business

Needs very long history of

transaction

MK-SVR

Good in non-normal distribution

(imbalance data)

Consider dynamic prediction

Quantile Regressio

n

High predictive performance

Provide view of relationship between

covariate and dependent variable

Provide prediction interval

Combine best of two approaches

Linear Regressio

n

Challenge in model non-normal data

Markov Chain

Flexible

Supporting Probabability

Having well developed theory

RFM

Predict customer short term behavior

Popular in segmentation

Combine two teqnique

Hierarchy Bays

Conditional Inference on

observed data

Deals better with parameter uncertainty

11Mathematical Models

(1) sales take place once a year (2) both yearly spending to retain customers and the customer retention rate remain constant over time 3) revenues achieved per customer per year remain the same.

(1) Sales occur more frequently than once a year (2) both yearly spending to retain customers and the customer retention rate remain constant over time 3) revenues achieved per customer per year remain the same.

(1) Sales occur more frequently than once a year (2) both yearly spending to retain customers and the customer retention rate remain constant over time 3) revenues achieved per customer per year remain the same.

0

0.51

1

* / 1

* / 1

nii

i

nii

i

CLV GC r d

M r d

/'

0

( 0.5)/1

1

* ( ) / 1

* ( ') / 1

pni pi

i

pni pi

i

CLV GC r d

M r d

/'

0

0.5( 1)/

1

* ( ) / 1

* ( ') / 1

n qiqi

i

nii q

i

CLV GC r d

M r d

(Berger & Nasr, 1998; Benoit & Poel, 2009; Bauer, Hammerschmitt & Braheler; 2010; Chen & Fan, 2013)

12Mathematical Models- Basic Formula

,

, 0 1T i t

i t

profitCLV

d

, , , ,1

* * argJ

i t ij t ij t ij tj

profit Serv Usage M in

In multi-service industries, Profit is defined as:

CLV for Customer i at time t

Pre-determined Discount rate

Finite time horizon

number of different services sold

dummy indicating whether customer I purchases service j at time t

amount of that service purchased

average profit margin for service j

(Berger & Nasr, 1998; Benoit & Poel, 2009; )

13RFM Model

RFM

Recency: the period since the last purchase; higher probability of a repeat purchaseFrequency: number of purchases made within a certain period; higher frequency indicates greater loyaltyMonetary: the money spent during a certain period; a higher value indicates that the company should focus more on that customer

WRFM

-Weighted RFM- instead of RFM. They dedicated weights to Recency, Frequency, and Monetary. depends on characteristics of the industry, different weights should be assigned to RFM parameters.

RFMTC

Recency, Frequency, Monetary value, Time since first purchase, and Churn probability using Bernoulli sequence in probability theory.

ci ci ci ci ci ci ciClV NR WR NF WF NM WM

(Abdolvand et al, 2009; Khajvand, Zolfaghar, Alizadeh, 2010; Chen & Cheng, 2013)

linking with other approaches such as

Pareto/NBD and Markov Chain Model. Pfeifer and Carraway (2000) are the

first researchers who introduced the idea of

combining two techniques to

overcome limitations of each method.

14MK-SVR Framework

Controlled Variables

Independent Variables

String Kernels

Gaussian Kernels

Ensemble Prediction

Promotion Optimization

Dynamic Prediction

UpdateLongitudinal

Data

MK-SVR

(Chen & Cheng, 2013)

Major Contributions

1. Longitudinal data are first introduced into the CLV prediction model to facilitate the dynamic prediction and improve the prediction performance.

2. An improved MK-SVR approach is proposed to simultaneously model the controlled and independent variables using multiple kernels.

3. A controlled variable about multiple promotion activities is incorporated into the dynamic CLV predictionmodel, and the dynamic customized promotion policy is determined by maximizing the CLV.

15Prediction Performance Metrics

Quantile Regression

Linear Regression

Quantile Regression

SVR

• Measures of the deviation between the actual and predicted values

MAE (Mean Absolute Error)

• Appropriate for prediction interval comparison from one data set not multiple

RMSE(Root Mean Squared

Error)

• percentage of customers whose predicted CLV falls into the same category as their true CLV

Hit-Rate

• provides an indication of the correctness of the direction

DS(Directional Symmetry)

(Donkers & Verhoef, 2007; Benoit & Poel, 2009; Abdolvand, Albadvi & Koosha, 2009; Chen & Fan, 2013)

16CLV Applications

Marketing Strategy

Customer Accusation

Customer retention

Customer Service

Mergers & accusations

CLV affects on

Valuation Optimization

Product Offering

Performance measurement

of Business Activities

Targeting & Selecting customers

Marketing resource allocation

Forecasting Pricing

Strategic Decisions

SegmentationCL

V Ap

plica

tion

prop

osed

in li

tera

ture

(Berger & Nasr, 1998; Donkers & Verhoef, 2007; Benoit & Poel, 2009; Abdolvand et al., 2009; Kumar et al., 2010; Chen & Fan, 2013)

17Conclusion

Current limitations of CLV are applicability and utilizing its potential in improving business strategies. limitations of predicting customer behavior, accuracy, the lack of empirical implementations, implementation challenges, the lack of integration between customer data, and marketing efforts CLV models ignore the potential risks

Future Research Calculating referral Value of customer Take Risk into account in CLV Extend Gupta et al (2009) to analyze sensitivity of each CLV variables Determine the multiple personalized marketing interventions to maximize the CLV

over time. The optimization of the best composition of the promotions

18References [1]Abdolvand, N., Albadvi, A., & Koosha, H. (2009). Customer lifetime value: Litriture scoping map and agenda for future research. International Journal of Marketing Perspective, 41-59.

[2]Anderson, K. (2001). Customer Relationship Management. New York: McGraw-Hill.

[3]Balanko-Dickson, G. (2007). Traps for Writting an Effective Business Plan. USA: McGraw-Hill Professional.

[4]Bauer, H. H., Hammerschmidt, M., & Braehler, M. (2010). The customer lifetime value concept and its contribution to corporate valuation. Yearbook of Marketing and Consumer Research, 47-67.

[5]Bayon, T. S., Gutsche, J., & Bauer, H. (2010). Customer Equity marketing: touching the intangible. European Management Journal, 213-222.

[6]Benoit, D. F., & Poel, D. V. (2009). Benefits of quantile regression for the analysis of customer lifetime value in a contractual setting: An application in finantial services. Expert Systems with Applications, 10475-10484.

[7]Berger, P. D., & Nasr, N. L. (1998). Customer Lifetime Value: Marketig Models and Applications. Journal of Interactive Marketing, 17-30.

[8]Berger, P. D., Weinberg, B., & Hanna, R. C. (2003). Customer lifetime value determination an strategic implications for a cruise-ship company. Journal of Database Marketing & Customer Strategy Management, 40-52.

[9]Borle, S., Singh, S. S., & Jain, D. C. (2008). Customer lifetime value measurement. Management Science, 100-112.

[10]Buttle, F., & Turnbull, J. (2004). Customer RelationShip Management Concepts and tools. Elsevier.

[11]Chen, Y. -Z., Zhao, M. -H., Zhao, S. -L., & Wang, Y. -J. (2006). A customer intelligence system based on improving LTV model and data mining. Fifth International Conference on Machine Learning and Cybernetics.

[12]Chen, Z. Y., Li, J. P., & Wei, L. W. (2011). Multiple Kernel Support Vector machine based multiple tasks oriented data mining system gene expression data analysis. Expert System with Applications, 12151-12159.

[13]Chen, Z.-Y., & Fan, Z.-P. (2013). Dynamic customer lifetime value using طولی data: An improved multiple Kernel SVR. Knowledge-Based Systems, 123-134.

[14]Cheng, C. H., & Chen, Y. S. (2009). Classifying the segmentation of customer value via RFM model and RS theory . Expert Systems with Applications, 4176-4184.

[15]Crowder, M., Hand, D. J., & Krzanowski, W. (2011). On optimal intervention for customer lifetime value. European Journal of Operational Research, 1550-1559.

[16]Donkers, B., Verhoef, P. C., & Jong, M. G. (2012). Modeling ClV: A test of Competing model in the insurance industry. Quant Market Econ, 163-190.

[17]Etizon, O., Fisher, A., & Wasserkrug, S. (2010). A modeling approach in customer lifetime evaluation in E-commerce domains, with an application and case study for online auction. Information Systems Frontiers, 421-434.

[18]Fader, P. S., Hardie, B. G., Lee, & Ka Lok. (2009). RFM and CLV: using iso-value curves for customer base analysis . Journal of Marketing Research, 415-430.

[19]Fang, Shyng, Lee, & Tzeng. (2012). Exploring the preferences of customers between companies and agents based on TCA. Knowledge Based System, 137-151.

[20]Glady, N., Baesens, B., & Croux, C. (2009). Modeling Churn using customer lifetime value. European Journal of Operational Research, 402-411.

[21]Gloy, B. A., Akridge, J. T., & Preckel, P. V. (1997). Customer lifetime value: An application in the rural petroleom market. Agribusiness, 335-347.

[22]Gonen, M., & Alpaydin, E. (2011). Multiple Learning Algorithms. Journal of Machine Learning Research, 2211-2268.

[23]Gupta, & Lehman. (2003). Customer as Assets. Journal of Interactive Marketing, 9-24.

19

Thanks For Your Attention