Segmentation and Lifetime Value Models Using SAS

• Segmentation and Lifetime Value Models Using SAS

• By: Edward C. Malthouse

• Publisher: SAS Institute

• Pub. Date: May 23, 2013

• Print ISBN-13: 978-1-61290-696-6

• Electronic ISBN-13: 978-1-61290-706-2

• Pages in Print Edition: 182

• Subscriber Rating: [1 Rating]

  
  

1
Strategic Foundations for
Segmentation and Lifetime Value
Models

Contents

1.1 Introduction

1.2 A process for increasing CLV

1.3 A taxonomy of CLV models

2
Segmentation Models

Contents

2.1 Introduction to segmentation models

2.2 K-means clustering

2.3 A process for building segmentations

2.4 The finite mixture model

2.5 Chapter summary

     Segmentation is one of the central concepts in marketing because in almost all situations customers have different wants, needs, preferences, and so on. Marketers call this heterogeneity. Whenever such heterogeneity exists, organizations that recognize and accommodate differences can achieve an advantage over competitors in a category. Conversely, organizations that do not accommodate differences across customers and offer only one version of their marketing mix create opportunities for competitors. If customers are heterogeneous then it follows that not all needs will be met with only one offering, and the needs of some customers will not be satisfied as well as they could be. A competitor can offer a better-targeted product and attract such customers.

     One approach for addressing heterogeneity is segmentation. Customers with similar wants and needs are grouped into segments so that an organization can better meet the different needs. This chapter discusses models for identifying segments. Before we begin this discussion, it is useful to distinguish between the marketing problems that the models address. A clear understanding of applications can inform many model-building decisions. The focus of the chapter is on the models rather than the marketing strategy, but analysts will have to make many subjective decisions and the more they understand about applications, the more effective they will be. There are two main applications:

• Market segmentation is covered in every marketing management book (for example, see Kotler and Keller, 2012). An entire market is first segmented into homogeneous groups. The organization usually targets one market segment and develops a product or service and brand for this segment. For example, automotive companies segment the market for cars and develop products or brands for different segments. There is a segment of families, and companies will produce, for example, minivans for this segment. There is also a segment of people seeking more luxury, and there are many cars targeted at this market segment.
• Customization and personalization to subsegments. Within a market segment there will still be heterogeneity. Customization is when a firm allows its customers to configure the marketing mix to meet these heterogeneous needs more closely. It is important to note that, with customization, it is the customers who configure to mix, not the company. The company enables the customization by offering up a menu of options, and the customer decides. For example, consider the Dell web site. Dell recognizes that customers have different needs depending on whom they are buying for: “home,” “small and medium businesses,” “public sector,” or “large enterprise.” The Dell landing page asks visitors to select one. Automobile companies allow consumers to configure aspects of a car. Cell phone customers can configure their plans. Note that the company does not target a single subsegment as it did with market segmentation. All subsegments are customers in the targeted market segment, and the organization must determine how to market to each. For further discussion, see Malthouse (2003a), Malthouse and Calder (2005), Calder and Malthouse (2004), and Malthouse and Elsner (2006) for discussion of the strategic issues.
  

3
The Simple Retention Model

Contents

3.1 The customer annuity model

3.2 The simple retention model

3.3 Estimating retention rates

3.4 Per-period cash flows m

3.5 Chapter summary

     Your company acquires customers, provides them with a product or service, and makes a certain amount of profit each month until they terminate the relationship forever. How much profit do you expect to make from customers during their lifetimes? How would increasing retention rates affect future profit?

     These are very common questions. Contractual service providers such as Internet service providers, health clubs, and media content providers (for example, Netflix, digital content subscribers, and so on), are in exactly this situation. For example, subscribers to Netflix pay a certain amount each month until they cancel. Companies that provide cellular phone service receive monthly payments from their customers until they cancel.

     One application is determining how much can be spent to acquire a customer. For example, it may cost a cellular phone company $400 to acquire a new customer and provide a handset; even if he only generates $50 in profit each month this would be a good investment as long as the cellular phone company can retain him sufficiently long to recoup the acquisition cost. Likewise, a company that is considering whether to invest marketing resources in retaining customers longer will need to know CLV.

     This chapter and the next show how to estimate the value of such customers in contractual situations. We begin with the case in which customers sign a contract for a certain number of periods and are not allowed to cancel. Next, we present the simple retention model, which allows customers to cancel, but assumes that the retention rate is constant over time and across customers, and that cash flows are independent of the cancelation time. The next chapter discusses when retention rates change over time, and when payment amounts depend on the time of cancelation.

4
The General Retention Model

Contents

4.1 The general retention model

4.2 Introduction to survival analysis

4.3 Product-moment estimates of retention rates

4.4 The discrete-time survival model

4.5 Application: trigger events

4.6 The beta-geometric model

4.7 Chapter summary

     The simple retention model (SRM) discussed in the previous chapter assumes that the retention rate is constant over time, as illustrated in the left side of Figure 4.1. But retention rates are not always constant. For example, companies in many different industries offer products or services at a lower rate for the first few periods. Credit cards commonly offer a low rate of interest for balance transfers during the first few months and then increase the rate. Cable, Internet service, and telephone companies commonly offer one monthly fee for the first few months and then increase it. In such cases, the retention rate often begins high and then drops after the rate increase, as shown in the right side of Figure 4.1. The last section shows how to account for unobserved heterogeneity in the retention rates with the beta-geometric model.

     The first section of this chapter develops the General Retention Model (GRM), which extends the SRM by allowing retention rates and cash flows to vary over time, and cash flows to depend on the time of cancelation. It also shows how to find the expected value of CLV in such situations. The next three sections show how to estimate retention rates with a class of statistical models commonly called survival analysis. section 4.5 discusses an important strategy for increasing CLV, which is detecting trigger events.

5
The Migration Model

Contents

5.1 Migration models: spreadsheet approach

5.2 Migration model: matrix approach

5.3 Estimating transition probabilities

5.4 Chapter summary

     This chapter discusses one model for estimating CLV when an organization does not have contractual relationships with its customers. We assume that such organizations acquire customers who might or might not generate profit during discrete, equal-length periods of time such as a month or a year. In contrast, the retention model from the previous two chapters assumes that inactivity indicates the end of the relationship. The migration model covered in this chapter assumes that inactivity does not necessarily signal the end of the relationship.

     This migration model is appropriate for organizations in many industries, including travel, most traditional and online retailers (clothing, supermarkets, electronics, appliances), automotive, financial services, and nonprofit organizations. For example, a customer who does not fly with an airline during some month might or might not fly again in the next month. Inactivity does not indicate that the customer is gone for good.

     We would like to forecast the future value of customers and answer related questions such as:

• What is the CLV of a customer? (The answer to this question informs a marketing decision such as whether to acquire or retain the customer.)
• How many active customers will we have after n periods in the future and how much profit will we make?
• If we invest more in customer retention (and could therefore increase retention rates), how would the size and profitability of our customer base change?
• Assuming some level of retention marketing, how many customers do we need to acquire to achieve some strategic goal such as maintaining current levels of profitability or increasing the number of active customers by 20% over the next two years?
• Organizations having contractual relationships with their customers often re-acquire customers who have previously canceled. For example, a cell-phone customer could cancel, switch to a competitor, and return later (after learning that the competitor is even worse). The retention models in the previous two chapters do not account for customers who reactivate. How can such firms correctly account for such reactivated customers?
  

     We will develop models to answer these questions by assuming that a customer is in some state during each time period. An example state might be defined by having bought in the previous period. Customers generate cash flows depending on their state, and migrate between states over successive periods with certain transition probabilities. This chapter first considers different ways of defining states and shows how to forecast future customer value using two equivalent approaches. The first approach, called the spreadsheet approach, is intuitive and easy to understand, but it is also a bit cumbersome. The matrix approach is a little more abstract and difficult to understand at first, but it is less cumbersome, and we will be able to derive closed-form expressions that will make evaluating perpetuities easy. Next, we survey applications and discuss how to estimate transition probabilities from data.

6
Data-Mining Approaches to
Lifetime Value

Contents

6.1 The data-mining approach to predicting future behaviors

6.2 Regression models for highly skewed data

6.3 Evaluating data-mining models

6.4 Accounting for the long-term effects of a marketing contact

6.5 Chapter summary

     In Chapters 3–5 we discussed probabilistic models for CLV. Each chapter began with a set of assumptions. For example, customers join and generate some fixed cash flow until they cancel and never return; the chance that a customer cancels is constant over time and customers; the event that a customer is retained in one period is independent of the event in other time periods. Based on such assumptions, we could derive an expected CLV. The point is that these models begin with a set of assumptions that characterize the relationship between the customer and the organization.

     This chapter explores an alternative data-mining approach that begins with data rather than assumptions about the customer relationship. The value of a customer in some future period is modeled directly as a function of what is known prior to the future period. We seek a function that, above all else, fits the data well, and the quality of the fit on an independent holdout sample will be the top priority instead of the assumptions and mathematical model characterizing the relationship. The data-mining approach is also used to predict response to a single contact.

     Both approaches have strengths and weaknesses and there is a long record of discussion about the topic.1 We take a pragmatic position on this debate and recommend using the model that does the best job of answering the questions in a particular situation.

     Perhaps the most important aspect of this chapter is that it provides a way to estimate both the short-term and long-term value of a marketing contact point. While this approach could be used with probabilistic as well as data-mining models, it is more closely associated with the data-mining models.

Multivariate 发表于 2015-2-18 04:50
1
Strategic Foundations for
Segmentation and Lifetime Value

可以下载吗?

哪里能下载？sos