6-Valuing Customers Gupta

8/12/2019 6-Valuing Customers Gupta

1/13

SUNIL GUPTA, DONALD R. LEHMANN, and JENNIFER AMES STUART*It is increasingly apparent tiiat the financial value of a firm depends onoff-balance-sheet intangible assets. In this article, the authors focus on

the most critical aspect of a firm: its customers. Specifically, they demon-strate how valuing customers makes it feasible to value firms, includinghigh-growth firmswith negative earnings. The authors define the value ofa customer as the expected sum of discounted future earnings. Theydemonstrate their valuation method by using publicly available data forfive firms. They find that a 1 % improvement in retention, margin, or acqui-sition cost improves firm value by 5% , 1 % , and . 1 , respectively. Theyalso find that a 1 % improvement in retention has almost five times greaterimpact on firm value than a 1 % change in discount rate or cost of capital.The results show that the linking of marketing concepts to shareholdervalue is both possible and insightful.

Valuing CustomersRecently, there have been many calls for marketingaccountability, measurement of marketing productivity, andbetter marketing metrics. Much of this stems from the dualrealities of crumbling functional boundaries, as evidencedby the growing roles of design in new product developmentas well as operations and information technology in cus-

tomer relationship managem ent, and the increasing pressureto relate marketing to stock market performance. This arti-cle relates the key focus of marketing effort, the customers,to the key measure of a firm's financial success, its marketvalue.Traditional accounting has focused on measuring tangibleassets, and the resulting data presented in annual reports, 10-Ks, and so on has formed the basis of firm valuation. How-ever, intangible assets (e.g., brand, customer, and employeeequity) are critical and often dominant determinants of value(Amir and Lev 1996; Srivastava, Shervani, and Fahey 1998),yet financial analysts at best tangentially cover these criticaldeterminants. Moreover, the dot-com bubble has been attrib-utedpost hoc to the use of too much marke ting (i.e., largeadvertising budgets and reliance on questionable marketingmetrics, such as eyeballs and click-throughs), which sug-gests that market-based measures m ay be in danger of beingrejected en masse.

*Sunil Gupta is Meyer Feldberg Professor of Business (e-mail: sg37@columb ia.edu), Donald R. Lehm ann is George E. Warren Professor of Busi-ness (e-mail: [email protected]), and Jennifer Ames Stuart is a doctoralcandidate, Columbia Business School, Columbia University. The authorsthank the three anonymous JM R reviewers for their helpful comments.They also thank Professor Noel Capon and the Center for Marketing ofFinancial Services at Columbia Business School for financial support.Finally, they thank the Teradata Center for Customer Relationship Man-agement at Duke University for its support and encouragement of thisresearch.

We merge traditional financial valuation methods basedon discounted earnings with the key marketing concept ofthe value of the customer to a firm. Specifically, we showhow a disciplined analysis of value based on customers andtheir expected future earnings (1) provides insights not pos-sible at the traditional more-aggregate level of analysis, (2)facilitates projections for new and growing businesses, and(3) provides an explanation for the dot-com bubble. Thebasis of this approach is customer lifetime value, which isthe discounted f'uture income stream derived from acquisi-tion, retention, and expansion projections and their associ-ated costs. In essence, we extend the concept of customerlifetime value and the works of several researchers (e.g.,Blattberg, Getz, and Thomas 2001; Niraj, Gupta, andNarasimhan 2001; Reinartz and Kumar 2000; Rust, Zei-thaml, and Lemon 2001) to the arena of financial valuation.

VALUING HIGH GROWTH BUSINESSESIn general, it is relatively easy to value stable and maturebusinesses. For these types of companies, the cash flowstream is relatively easy to predict. Therefore, financialmodels such as discounted cash flow (DCF) work reason-ably well. In contrast, the valuation of high-growth busi-nesses is complex, because these businesses have limitedhistory to draw on for future projections. They also typicallyinvest heavily in the early periods, which results in negativecash flows. Consequently, traditional financial methods arenot useful for evaluating these businesses; it is difficult touse a price-to-earnings (P/E) ratio for a company that has noearnings or negative earnings or to use the DCF approachwhen a firm has negative cash flow. This difficulty becam eevident during the height ofthe dot-com bubble, when manyinnovative valuation methods emerged.A popular measure that emerged in 1999-2000 is thenumber of customers, or eyeballs. This metric is based on


2/13

8 JOURNAL OF MARKETING RESEARCH, FEBRUARY 2004the assumption that growth companies need to acquire cus-tomers rapidly to gain first-mover advantage and to buildstrong network externalities, at times regardless of the costinvolved {The allStreet Journal 1999). Academ ic researchin accounting also has provided validation for this assump-tion. For example, Trueman, Wong, and Zhang (2000) com-bine fmancial information from company statements withnonfinancial information from Media Metrix for the periodSeptember 1998 to December 1999 for 63 Intemet firms. Aregression of market value on these components reveals thatthough bottom-line net income has no relationship to stockprice, unique visitors and page views add significantexplanatory power. In a related study, Demers and Lev(2001) use similar data for 1999-2000 for 84 Internet com-panies in order to examine the relationship between marketvalue and nonfinancial measures during and after the Inter-net bubble. They find that nonfinancial measures, such asreach (i.e., number of unique visitors) and stickiness (i.e.,site's ability to hold its customers), explain the share priceof Intemet companies, both before and after the bubbleburst.

Note that these studies are correlational, and assume thatthe market value represents the true intrinsic value of thefirm at any time, which is an efficient market argument.However, even if the markets are efficient in the long run,recent history suggests that significant deviations exist in theshort run. In other words, the value of the dependent variablein these studies is likely to change significantly over time,which may alter conclusions about the value of customers.Partly because of this, financial analysts are now quite skep-tical of nonfinancial metrics, especially number of cus-tomers. For example, an article criticized the Wall Streeticon Mary Meeker for relying too much on eyeballs andpage views and even putting them ahead of financial meas-ures {Fortune2001b).Our pproach

The mood on Wall Street suggests that customer-basedmetrics not only are irrelevant to firm valuation but also canbe misleading. We argue against this view. We suggest andshow that value based on customers can be a strong deter-minant of firm value. The premise of our customer-basedvaluation approach is simple: If the long-term value of acustomer can be estimated and the growth in number of cus-tomers can be forecast, it is easy to value a company's cur-rent and future customer base. To the extent that the cus-tomer base forms a large part of a company's overall value,it can provide a useful proxy for firm value. We demon strateour approach for one well-established firm for which tradi-tional fmancial methods work well. In addition, we use ourapproach to estimate the value of four Intemet firms forwhich traditional fmancial methods may not work well.We also show that it is not necessary to obtain detailed pro-prietary information (as is typically done in database market-ing and customer lifetime value research) to apply ourapproach. Except for retention rate, we use only publishedinformation from firms' annual reports and other financialstatements to estimate the value of their customer base. T here-fore,our approach can be valuable for extemal constituencies,such as investors, financial analysts, and acquirer companies,which may not have access to detailed internal data.The closest parallel approach to our own is that of Kim,Mahajan, and Srivastava (1995), who use a DCF method to

estimate the value of a business in the wireless communica-tions industry. Our work differs from their approach in sev-eral important w ays. First, our approach focuses on the com-pany level (rather than the industry level), and we apply ourmethod to multiple firms. Second, we do not need to makeany assumption about the time when growth ends. Weexplicitly model growth in customers and firm value. Third,we incorporate customer retention, which has a significantsubstantive and methodological impact. For example, indus-try reports show that the annual chum rate in the telecom-munications industry (which Kim, Mahajan, and Srivastavaexamine) is more than 20 . The industry estimates that thischum rate reduces firms' value by several billion dollars.Our analysis confirms that customer retention has a largeimpact on firm value. The inclusion of customer retentionrequires us to account for different customer cohorts thatchange the model conceptually and mathematically. Finally,the inclusion of customer retention and acquisition in themodel provides insights for managers about potential mar-keting levers that are available to them for improving cus-tomer and firm value.

In summary, the key contribution of our approach is theprovision of an estimate of the value of the current andfuture customer base of a firm, which in turn forms a proxyfor the value of high-growth firms for which traditionalfmancial methods do not work well. Our main contributionslie in three areas: (1) providing a better method for forecast-ing the future stream of income when it is not possible tosimply extrapolate the historical (negative) earnings of afirm, (2) providing insights about marketing levers (e.g.,retention) that can help managers improve firm value, and(3) suggesting that customers are indeed assets, and there-fore customer-related expenditures should be treated asinvestments rather than expenses.MODEL

Conceptually, the value of a firm's customer base is thesum ofth e lifetime value of its current and future cu stomers.We build a model for the lifetime value of a cohort of cus-tomers, aggregate the lifetime value across current andfuture co horts, and then construct m odels to forecast the keyinput to the model (e.g., the number of customers in futurecohorts).We begin with a simple scenario in which a customer gen -erates a margin m, for each period t, the discount rate is i,and the retention rate is 100 . In this case, the lifetimevalue of the customer is simply the present value of thefuture income stream:(1) LV =

This is identical to the DCF approach of valuing perpetu-ity (Brealey and M yers 1996). When we account for the cus-tomer retention rate r, the formulation is modified asfollows:'2 LV =

=

iWe recognize that retention rates may not be constant; however, wemake this simplified assumption for the ease of modeling and empiricalapplication. Our data for Ameritrade supports our assumption.


3/13

Valuing Customers 9Many researchers have debated the appropriate durationover which lifetime estimates should be based (Berger andNasr 1998). We build our model for an infinite time horizonfor several reaso ns. First, we do not need to specify arbitrar-ily the number of years that a customer will stay with thecompany. Second, the retention rate accounts for the factthat over time, the chances of a customer staying with thecompany decrease significantly. Third, the typical methodfor the conversion of retention rate to expected lifetime andthen calculation of present value over that finite time periodoverestimates lifetime value.2 Fourth, both retention and d is-count rates ensure that earnings from the distant future con-tribute significantly less to lifetime value. Finally, modelswith infinite horizons are significantly simpler to estimate.To estimate the lifetime value of the entire customer b aseof a firm, we recognize that the firm acquires new cu stomersin each time period. Each cohort of customers goes throughthe defection and profit pattern shown in Table 1, whichshows that the firm acquires ng customers at time 0 at anacquisition cost of CQ per customer. Over time, customersdefect such that the firm is left with nor customers at the end

of Period 1, nor2 customers at the end of Period 2, and so on.The profit from each customer may vary over time. Forexample, Reichheld (1996) suggests that profits from a cus-tomer increase over that customer's lifetime. In contrast,Reinartz and Kumar (2000) find that this pattern does nothold for noncontractual settings.Therefore, the lifetime value of C ohort 0 at Time 0 isgiven by3) LVn = 1 + i)>Cohort follows a pattern similar to that of C ohort 0, exceptthat it is shifted in time by one period. Therefore, the life-time value of C ohort at Time 1 is given by4) LV, =

2For example, consider a situation in which the annual margin from acustomer is 100, the retention rate is 80%, and the discount rate is 12%.Using Equation 2, we estimate the lifetime value of this customer to be 250. An alternative approach would suggest that the 80% retention rateimplies that this customer is expected to stay with the company for fiveyears. The present value of the 100 stream of income for five years is 360. an overestimate of approximately 44%.

It is easy to convert this value at Time 0 by discounting it forone period. In other words, the lifetime value of C ohort atTime 0 is5 ) LV, = n,c.

In general, the lifetime v alue for the kth cohort at Time 0 isgiven by(6) LV,, = m. ,;The value of the firm's customer base is then the sum of thelifetime value of all cohorts:(7) Value = rt-k

k = 0 t = k = QAlthough it is easier to conceptualize the model in dis-

crete terms, in reality customer acquisition and defection isa continuous p rocess. Schmittlein and Mahajan (198 2) showthat estimation of an inherently continuous process, such asa Bass (1969) diffusion mod el, with a discrete version pro-duces biases. Furthermore, we model key input (e.g., n^) asa continuous function. Therefore, we use a continuous ver-sion of customer value.If the annual discount rate is i and we continuously com-pound m times a year, the discount rate at the end of the yearis 1/[1 + (i/m)])m. As m approaches infinity, the discountrate becomes e-'' (Brealey and M yers 1996). Similarly, it iseasy to show that rV(l + i)' is equivalent to e-tC + ' - ^Wt.Therefore, the continuous version of Equation 7 is(8) Value = 'dtdk

k = O t = k

- JEquation 8 provides customer value before any tax consid-erations. C onsistent with financial models, we use the after-tax value as a proxy for firm value. Here, we use a corpo ratetax rate of 38% for all firms. Before building models of n^and so on, we turn to data in our empirical application to

Table 1NUMBER OF CUSTOMER S ND M RGINS FOR E CH COHORT

TimeCohort Cohort 1 Cohort 2

Customers Margin Customers Margin Customers Marginnonor mom . n,r mom , m,

ni3

Notes: We have assumed that each customer cohort follows the same pattern of margins (mo, m,, m2, ... m^ -Although it ispossible to make thispatternvary across cohorts, this increases the model com plexity significantly. In addition, literature lacks theoretical justification for a specific pattern. Finally mostdata sets are insufficient to validate a specific pattern empirically.


4/13

JOURNAL OF MARKETING RESEAR CH FEBRUARY 2004understand the nature of available information. The avail-able data, their empirical patterns, and theory guide us in ourselection of appropriate models for these input variables.

APPLICATIONData

We estimated our model using data from five companies:one traditional firm (Capital One) and four Internet compa-nies (Amazon.com, Ameritrade, eBay, and E*Trade). Weused a traditional firm to show that similar to standard finan-cial models, our approach is capable of providing good esti-mates of firm value. We used four Intemet companies toshow the usefulness of our approach when standard finan-cial models may not apply well because of low or negativecash flows. We based our choice of companies on the avail-ability of public data.We used quarterly data from annual reports, 10-K and 10-Q statem ents, and other company repo rts for the period from1996-1997 to March 2002. The data for each quarterinclude number of customers, margin, and marketing costs.Using these data, we estimated the acquisition cost andquarterly margin per customer. A summary of the data isprovided in Table 2.Number of customers Figure 1shows the growth in num -ber of customers for each of the five firms. The data show aremarkable consistency with classical diffusion theory. Anatural candidate for estimation of the number of customersin future p eriods is the Bass (1969) diffusion model. Thecontinuous Bass m odel (p. 218) is based on the solution to anonlinear differential equation, and the resulting sales ornumber of customers' equation is quite complex. The dis-crete analog model is simpler but still poses challenges inour context, because sales or number of new customers arefunctions of either cumulative sales or customers. Thisrecursive relationship makes the integration (or summation)more complex.Therefore, we modeled customers using an S-shapedfunction that is similar to the Bass (1969) diffusion modelbut is mathematically more convenient in our context.Specifically, we suggest that the cumulative number of cus-tomers Nj at any time t is given by

(10) dN, _ ay exp(-P - yt)exp(-|3 -

(9) 1H exp(-P - 'The number of customers reaches a as time approachesinfinity. The parameter y captures the slope of the curve. Thenumber of new customers acquired at any time is

This model, also called the technological substitutionmodel, has been used by several researchers to model inno-vations and to project the number of customers (e.g.. Fisherand Pry 1 971; Kim, Mahajan, and Srivastava 1995). Bass,Jain, and Krishnan (2000) suggest that estimates from thismodel are comparable to those from the Bass model.Margin It is relatively straightforward to obtain the quar-terly revenues for a firm from financial statements. How-ever, the assessment of costs poses challenges because firmsdo not report direct costs in a consistent manner. For exam-ple, although fulfillment cost (i.e., shipping and han dling) isa large portion of Am azon .com 's operating exp ense, the firmdoes not include it in calculating its margin. We includedthese costs in our estimate of the margin. Similarly, a majorexpense for the credit card company Capital One is thesalary of its emplo yees. We included salaries as a direct costto arrive at margins for two reasons. First, Capital One(2001, p. 22) explicitly states in its annual report that its salaries and associate benefits expense increased 36% as adirect result of the cost of operations to manage the growthin the compa ny's accounts. Second, to separate fixed andvariable cost, we ran a regression between employeeexpenses and the number of customers (Anthony, Hawkins,and Merchant 1998), which produced an R2 of .974, withalmost all the cost allocated as variable. In other words, as adirect marketing company, an increase in the number of cus-tomers for Capital One is directly associated with anincrease in employee expenses. We followed a similarprocess for the other firms.

After we determined total margin for a quarter, we esti-mated quarterly margin per customer by dividing the totalmargin by the number of current customers in that quarter.Unlike the num ber of customers, there is no systematic trendin margins. We confirmed this by mnning a regression. Thelack of a systematic pattern echoes the debate amongresearchers in this area. For example, Reichheld (1996)finds that the longer a customer stays with a company, themore the customer buys. Reichheld also suggests that thecompany has the potential to cross-sell its products to itscustomer base. In addition to increased revenue, Reichheldfinds that the longer a customer stays with a company, thelower is the cost of doing business with the customer. How-ever, Reinartz and Kumar (2000) challenge these findingsand show that duration of stay is not necessarily related toincreased margin.

Table 2DESCRIPTIVE DATA

CompanyAmazon.comAmeritradeCapital OneeBayE*Trade

Data PeriodFrom

March 1997September 1997December 1996December 1996December 1997

ToMarch 2002March 2002March 2002March 2002March 2002

Number ofCustomers33.800,000

1 87746,600,00046,100,0004,117,370

QuarterlyMargin$ 3.8750.3913.714.3143.02

AcquisitionCost$ 7.70203.4475.4911.26391.00

RetentionRate70 %95 %85 %80 %95 %

Notes: Num ber of customers is as of the end of Jvlarch 2002. Quarterly margin is per customer based on the average of the previous four quarters. Acqui-sition cost is per customer based on the average of the p revious four quarters.


5/13

Valuing CustomersFigure1

NUMBER OF CUSTOMERS

Amazon.com Ameritrade4

35

3

25

2

15

10 000 000

5 000 000.

0 Mil

2.000,000

1,800,000

1,600,0001,400,000

1 21,000,000

800 000600 000400 000200 000

0

11ill 11111111

5

45

435

325

2

151

5

Capitai One

LI 1 1

50 000 00045 000 00040 000 00035 000 00030 000 00025 000 00020 000 00015 000 00010 000 0005,000,000

0

eBay

-

i l l

E Trade4 500 0004 000 0003,500,000

3,000,000

2,500,000

2,000,000

1,500,000

1,000,000

500 0000

1 1111111


6/13

12 JOURNAL OF MARKETING RESEAR CH FEBRUARY 2004In addition to the debate about the pattem of margins overtime within a cohort, the issue is further complicated in ourcase because our aggregate data combine margins acrossseveral cohorts, each of them at a different life cycle stage.As a company expands its customer base, it tends to drawmore marginal customers who do not spend as much withthe company as its original customers did. Consequently,average revenue per customer may decline over time. This isespecially true if a company's customer base expands rap-idly, thereby changing its customer mix. For example,CDNow's revenue per customer fell from $23.15 to $21.16in 1998. In the first quarter of 1999, it acquired its com peti-tor N2K, which further contributed to the decline in its rev-enue per customer from $18.15 in the first quarter of 1999to $14.42 in the second quarter of 1999 (Kaufman 1999).Given conflicting evidence in research and the lack of anysystematic pattem in our data, we used the average of thelast four quarters as the margin for future periods.3 We per-formed sensitivity analysis to determine how customer andfirm values change with changes in margins.Acquisition cost Although acquisition cost is easy to

define, it is difficult to estimate precisely in an empirical set-ting. Companies use different accounting and managementpractices to define the costs that should be included in thismeasure. Consequently, some marketing studies (e.g.,Reinartz and K umar 2000) do not include acquisifion cost inthe analysis.We operationalized acquisition cost by dividing the totalmarketing cost by the number of newly acquired customersfor each time period. Although some of the marketing costis incurred for retention purposes as well, we did not haveinformation to separate the two costs. However, this simpli-fication is not likely to have significant impact on our resultsfor several reasons. First, the firms in our data set were in thegrowth stage of their life cycle, and thus customer acquisi-tion was a dominant factor. Second, several studies showthat, in general, customer acquisition costs are significantlyhigher than customer retention costs (Reichheld 1996). Forexample, Thomas (2001) estimates acquisition cost per cus-tomer to be $26.94 and retention cost per customer to be$2.15. Finally, our estimates of acquisition costs are quitesimilar to estimates published in various industry reports.As with profit margins, there is no systematic trend inacquisition costs. We confirmed this by running a regres-sion. There are two opposing forces that affect acquisitioncosts. As competition intensifies and a company acquiresmarginal customers (i.e., customers to whom the firm'sproducts and services are less convincing), its acquisifioncost increases. This is most evident in the telecommunica-tions industry, in which the acquisifion cost per subscriberdramatically increased from $4,200 when AT&T boughtTCI and MediaOne to $12,400 when Vodafone acquiredMannesmann. However, as a company increases its cus-tomer base and reputation in the market, word of mouth andbranding pow er make it easier to attract new customers. It isdifficult to know how these two forces counterbalance eachother. Because our data show no significant patterns in theacquisition costs over time, we used the previous four quar-

^Four-quarter average, also known as trailing 12-month average, is alsoa common practice among financial analysts.

ters' average as the cost for future customer acquisifions.4We also assessed the sensitivity of our results to changes inacquisition costs.Retention Customer retention is one of the most criticalvariables that affect customers' lifetime profit, yet mostcompanies do not make it publicly available. Therefore, weestimated retention rates from several sources.For Ameritrade, we obtained detailed account informa-tion from Salomon Smith Bamey that shows Ameritrade'saccount retenfion rates to be 95.0% for fiscal year 1999,96.2% for 2000, 95.7% for 2001, and 94% (annualized) forthe quarter ending March 2002. These figures show twothings. First, 95% is a good estimate for Ameritrade's aver-age retention rate. Second, over time, this retention rate hasnot changed significantly. We were un able to obtain any spe-cific informafion about E*Trade. Given its similarity toAmeritrade, we used a 95% retention for E*Trade.For Capital One, we obtained retention-rate estimatesfrom an industry expert, who suggested that the retentionrate for North American credit card companies is in therange of 85 % to 88% . According to the expert, there aremany factors that contribute to retention (e.g., credit quality,pricing, customer serv ice). He further suggested that CapitalOne scores slightly worse than many other companies (e.g.,MBNA) on some of these factors, and its retention rate wasin the range of 84% to 86%. Therefore, we used the averageof 85% as our best estimate for Capital One's customerretenfion rate.Amazon.com has changed the way it reports its number ofcustomers in financial statements. Previously, Amazonreported cumulative customers (both active and inactive),but now it reports only active customers (retroactively fromfourth quarter of 1999). Using data on active and cumu lativenumber of customers for 2000-2001, we estimated Ama-zon's retention rate to be in the range of 65.3% to 74.6%,with an average of approximately 70%. This estimate is sim-ilar to Amazon's self-stated retention rates and slightly lessthan the 78% retention rate suggested by some consultants(Seybold 2000).EBay does not provide estimates of its retention rate. Inthe absence of any data, we used an 80% retention rate foreBay, which is the average observed among U.S. firms(Reichheld 1996). For all companies, we also conductedsensitivity analysis.Discount rate Standard financial methods (e.g., capitalasset pricing model) can be used to estimate discount rates.Damodaran (2001) estimates the cost of capital for Amazonto be 12.56%. In general, finance texts suggest a range of8% to 16% for the annual discount rate. Therefore, we usedthe average of 12% for our analysis. We also show the sen-sitivity of our results to different rates of discount.Estimation

For each company, we have historical data on the actualnumber of customers. These numbers are a net effect of allcustomers who ever tried the services of the company minusthe defectors. For example, if a company has 100,000 cus-tomers in Period 0 and 130,000 customers in Period 1 andits retention rate is 80%, it acquired 50,000 customers fromPeriod 0 to Period 1. Therefore, the cumulative number of A firm has already incurred acquisition costs for its existing customers.Therefore, this cost is sunk and is not considered in valuation.


7/13

Valuing Customers customers who ever tried the company's services is 100,000in Period 0 and 150,000 in Period 1. In our valuafion mod el,nt is the number of customers acquired during time t, notthe number of net (i.e., acquired minus defected) newcustomers. Therefore, we model number of customers whoever tried a firm's services (i.e., Nj). When the parameters ofthe model have been estimated, it is easy to obtain nj byusing Equation 10. We estimated the model for forecastingnumber of customers using nonlinear least squares, as Srini-vasan and Mason (1986) suggest. We then used parametersof this model as well as estimates of acquisition cost, reten-fion rate, margin, and discoun t rate as input to the valuationmodel in Equation 8. We then evaluated the model usingMathematica.

Note that the procedure we described previously assumesthat all customers (both active and inactive) potentiallyaffect the future grow th of num ber of custom ers. It is possi-ble to modify this assumption and to construct alternative,and potentially complex, models of diffusion. For example,an alternative model is to assume that whereas active andinactive customers define the remaining market potential,only currently active customers spread positive word ofmouth to affect future customer growth. This model is sim-ilar to a diffusion model that incorporates replacem ent pur-chases (Kamakura and Balasubramanian 1987). We esfi-mated this model for Amazon.com and found that its resultsare similar to those obtained from our model. For example,this model projected that the total market potential for Am a-zon was 71.8 million, and our model esfimated total marketpotential to be 67 million. Further research should inyesfi-gate alternafive models of custom er growth, such as a m odelthat assumes negative word of mouth from defectors andpositive word of mouth from currently active customers.ResultsWe report results for the number of customers and thendiscuss results for the value of a firm's customer base as ofMarch 3 1, 2002.Number of customers. Table 3 provides parameter esti-mates and fit statistics for each of the five companies. Wereport mean absolute deviation (MAD) and mean squarederrors (MSE) as measures of fit, because traditional meas-ures such as R2 are not appropriate for nonlinear regressionmodeling (Bates and Watts 1988; Srinivasan and Mason1986).Our model fits the data quite well, as is indicated bylow MAD and MSE.

All the parameters are significant. Parameter a providesan estimate of the maximum number of customers who areexpected to try a company's product and services. Table 3results show that the maximum num ber of triers isexpected to be 67 million for Amazon .com, 2 .48 million forAm eritrade, 171.2 million for Capital One, 81.95 million foreBay, and 4.72 million for E*Trade. The maximum numberof actual customers will be less than this number as a resultof defecfion.From Equation 10, it is easy to show that the peak for cus-tomer acquisition occurs at -p/y . Table 3 results suggest thatthis peak occurs approximately 10 to 21 quarters from thestart of our data period (1997). In other words, for the com-panies in our data set, customer acquisition has alreadyreached a peak.5 After this time, companies will continue toacquire customers but at a slower rate. For example, Ama-zon added four million new customers in December 2000but only three million customers in the subsequent twoquarters.Value of the customer base. The number of current cus-tomers and a forecast of customers to be acquired in thefuture enabled u s to esfimate the value of a firm's customerbase (current and future). We used average acquisition costs,margins, and retenfion rates from Table 2 and parameterestimates from Table 3 as input to Equation 8. Table 4 pres-ents estimates of customer value and market value for thesefirms as of March 31, 2002 (the end of our data period).Because stock prices change every day, firm value varies(sometimes dramafically) in a quarter. Therefore, in Table 4,we include the high and low market value for the January-March 2002 q uarter. We also include P/E ratios for the com-panies because (1) they are comm only used in financial val-uation methods, and (2) we wanted to emphasize that it isdifficult to rely on this metric for fast-growing companies.For example, two of the companies (Amazon.com andE*Trade) have negative eaming s, so P/E ratio is not defined.Furthermore, two other companies (Ameritrade and eBay)have only modest eamings, and thus their P/E ratios areextremely high and significantly outside the market averagerange of 20-30 .

5To estimate an S-shaped curve, we need an inflection point in the data.The inflection point is the time of peak customer acquisition. For data setsin which the inflection point is not observed, there are two possible solu-tions: (1) provision of an extemal estimate of a parameter such as marketsize or (2) use of a Bayesian method to provide priors for the parameters.

Table 3PARAMETER ESTIMATES FOR NUMBER OF CUSTOMERS IN MILLIONS)

Parametersay

Amazon.com67.045(3,615)-^.114 (,139).265 (.015)

Time to Peak of Customer Acquisition-p/y 15.64Calendar date December 2000Fit StatisticsMADMSE .556.594

Ameritrade2,482 (.121)-3.345 (.114).263 (.016)

12.72September 2000.041.004

Capital One171.200(15.864)-3.052 (.079).149 (.003)

20.48December 2001.393,346

eBay81.945(3,995)-6.009 (.145).317 (.013)

18.96June 2001.590.763

E Trade4.719 (,064)-3.441 (.086).365 (,012)

9.43March 2000.049.004

Notes: -p/y gives an estimate of the number of quarters from the start of the data for a company when customer acquisition is expected to reach its peak.


8/13


9/13

Valuing CustomersFigure 2

MARKET VALUE AND CUSTOMER VALUE OVER TIME

15

Amazon.com Ameritrade6,00

5.00

= 3,00

2 00

1 00

0

5,i;5.36

04

2,22

_

112B

June '01 Sept. '01 Dec. 01Time

March02

3.00

1.00

June01 Sept. '01 Dec. 01Time

March02

June 01

12.00-. 10.00-o

ffl6.00

2.00 -0 -

14,64

J ?[ 1-B

CapitalOne14.08

11.7011,01

H: |

7-^10,901I

^ ^ k o o

eBay

Sept. '01 Dec. '01 March 02T ime

20.00

15 00

.9 10.00

18.72

12.59

1|i 80

18,5515,85

June01 Sept. '01 Dec. 01Time

March02

E Trade4 00

3 00

2 00

1 00

3 25

^ 8HE

2 83

58

3 35

June01 Sept. '01 Dec. 01

Time

March02

I Market value Customer value


10/13

16 JOURNAL OF iVlARKETING RESEARCH, FEBRUARY 200companies' market valtie, customer value estimates forAmeritrade, E*Trade, and Capital One are within reasonablerange of their market values. We emphasize that, in general,market value shows significant fluctuations within a quarter,often without any new information about a company's oper-ations. For example, at the end of the third quarter of 2001,market value for Capital One w as $9.7 billion; however, dur-ing that quarter, the company's market value fluctuatedbetween $7.7 billion and $14.2 billion.''To confirm further the relationship betw een c ustomer andmarket value, we ran a simple regression with market valueof a company as the dependent variable and customer valueas the independent variable. Having used data for four quar-ters for each of the five companies, the regression producedan R2 of only .139. However, when we ran the regressionwithout Amazon and eBay (two companies whose marketvalues are significantly different from our estimate of theircustomer va lue), the R2 was .927. Furthermore, the interceptin the regression was not significantly different from zero,and the parameter estimate of customer value (1.026) wasnot significantly different from one.

MANAGING C USTOMER VALUEOur analysis shows that customer value provides a goodproxy for firm value. Because the estimation of customervalue requires more detailed input than traditional valuationmethods, its benefit is not only in terms of firm valuation. Agood metric for customer value is the starting point for bet-ter management of customers as assets. In this section, wefocus on two aspects: (1) how changes in acquisition costs,margins, and retention rates affect customer value of a firmand (2) the relative importance of customer retention, whichis a key component of the marketing function, and the dis-count rate or cost of capital, which is a traditional focus of

the finance function.Impact of Acquisition Cost, Margin, and Retention Ra te

Table 5 shows how customer value changes with changesin acquisition cost, margin, and retention rate. Our resultsshow a consistent pattem: Improved customer retention hasthe largest impact on customer value, followed by improvedmargins, and reduced acquisition cost has the smallestimpact.''Sometimes market value fluctuations are significant across quarters aswell (e.g.. during the height of the dot-com fever). For example, in March2000, market value for Amazon.com was $23.45 billion. A year later, inMarch 2001. its market value dropped to $3.67 billion, and as of March

2002, it had climbed back to $5.3 billion. Our estimates of Amazon's cus-tomer value for the entire two-year period are consistently less than 1billion.

A 1% improvement in acquisition cost improves customer value by .02% to .32%. The greatest impact oreduced acquisition cost is for Capital One, which is consistent with Capital One's having passed its customer acquisition peak only recently (see Table 3): It is still acquiring large number of customers. Therefore, any improvement iacquisition cost has a significant impact on the company'overall value. Tn contrast, Ameritrade and E*Trade passedtheir acquisition peaks several quarters previously andtherefore have the least impact of improving acquisitioncost.A 1% improvement in margins, such as from crossselling, improves customer value by approximately 1%This result is consistent across all firms. A 1% improvem enin customer retention improves customer value by 2.45% to6.75%. The higher the current retention rate of a company(e.g., Ameritrade 9 5% versus Amazon.com 70% ), the higheis the impact of improved retention.In summary, we find that retention elasticity is 3-7times margin elasticity and 10-100 times acquisition elasticity. These results are consistent with previous studiethat emphasize the importance of retention (e.g., Reichheld 1996). Notably, after the dot-com bubble burst. WalStreet and many Internet firms began to focus on andreduce acquisition costs. Demers and Lev (2001) explainthis by showing that before the market's correction foInternet stocks, the market treated expenditures on bothmarketing and product development as assets rather thancurrent expenses. Demers and Lev further found that in theyear 2000, after the shakeout, product developmenexpenses but not marketing expenditures continued to becapitalized as assets. Consistent with our study and contrary to current market perception, Demers and Lev showthat Web-traffic metrics (e.g., traffic, loyalty) continue tobe value relevant.We note two caveats for interpreting the results of Tabl5. First, we did not include the cost of improving retentioor margin. Therefore, even though improvement in retentionhas the largest impact on customer value, we cannot suggesthat a firm should always improve its customer retentionUsing a game theoretic model, Shaffer and Zhang (2002show that it is not advisable for firms to eliminate eithechum or customer defection completely. If a firm has 100%customer loyalty, it may be underpricing or leaving moneyon the table. Second, our analysis ignores interactionamong acquisition, retention, and margins. It is quite likelythat certain acquisition programs (e.g., price promotionsattract customers with low retention rates, and studies (e.g.Thomas 2001) have provided methods to link customeacquisition and retention.

Table 5IMPACT OF 1 IMPROVED RETENTION, ACQUISITION COST, MARGINS, AND DISCOUNT RATE ON CUSTOMER VALUE

Amazon.comAmeritradeCapital OneeBayE*Trade

Customer alue( Billion)Base Case

,821.6211,001.892,69

Percentage Increase in CustomerRetention

2.45%6.755.123.426.67

Acquisition Cost.07%.03.32.08.02

alue(with a 1% Improvement)Margin1.07%1.031.321,081,02

Discount Rat.46%1.171.11.63

1,14


11/13

Valuing Customers Impact of Retention ersus Discount Rate

Discount rate, or cost of capital, is a critical variable in theevaluation of the net present value of any cash flow streamand firm valuation, and thus the finance community spendsconsiderable effort in measuring and manag ing a firm's costof capital (see, e.g., Brealey and Myers 1996). In contrast,the marketing and business community has just begun tomeasure and manage customer retention, of which theimportance to firm valuation is even less evident. To com-pare the relative importance of customer retention and dis-count rate, the right-hand column of Table 5 shows howchanges in discount rates, in contrast to changes in market-ing levers, affect customer value for the firms. The resultsshow that a 1% improvement in customer retentionenhances customer value (and, in tum, firm value) byapproximately 2.45% to 6.75%, whereas a similar decreasein the discount rate increases customer value and thus firmvalue by only .5 % to 1.2%. In other w ords, the retentionelasticity is almost five times the discount rate elasticity.

Another way to examine the effects is to assess the valueof customers for the typical range of retention and discountrates. The finance literature suggests a typical range of dis-count rates of 8% to 16% (Brealey and Myers 1996). On thebasis of industry information (e.g., Reichheld 1996) and theretention rates for the five compa nies in our em pirical analy-sis,we used a range of 70% to 90% for retention rate. Usingthese ranges, we reestimated customer value for the compa-nies in our data set.Table 6 reports our results. Several notable things emergefrom this table. First, consistent with the Table 5 results,retention rate has a larger impact on customer value thandoes discount rate. For example, an improvement in cus-tomer retention from 70% to 90% increases customer valueforA ma zon .com by 1.38 billion - .75 billion = .63 bil-lion (for a 16% discount) to 1.07 billion (for an 8% dis-count). In contrast, an improvement in discount rate from16% to 8% increases Amazon's customer value by .15mil-lion (for 70% retention) to .59 billion (for 90% retention).Second, there is a strong interaction between discount rateand retention rate. Specifically, the impact of retention oncustomer value is significantly higher at lower discountrates. This suggests that companies in mature and low-riskbusinesses should pay even more attention to customerretention. Third, the value of customers and, by im plication,the value of a firm for the high retention-low discount sce-nario is 2.5 to 3.5 times its value under the low retention-high discount scenario. Although we do not consider the rel-ative cost of improvement to the retention rate versus thediscount rate, this analysis suggests the importance of mar-keting levers, rather than financial instruments, in improv ingcustomer and firm value.

CONCLUSIONCustomer lifetime value is receiving increasing attentionin marketing, especially in database marketing. In this arti-cle, we attempt to show that the concept not only is impor-tant for tactical decisions but also can provide a useful met-ric to assess the overall value of a firm. The underlyingpremise of our model is that customers are important intan-

gible assets of a firm, and their value should be measuredand managed as is any other asset. Our article builds onrecent work in marketing in the area of customer lifetimevalue by extending it to the arena of financial valuation. Wealso build on recent work in accounting in which theapproach has been to regress current market value of a firmagainst tangible and intangible assets. Implicitly, thisapproach assumes that the market is correctly valuing firm s.The recent history of dot-com companies casts doubt on thisassumption. In contrast, we estimate the value of a firm'scurrent and future customer base from basic principles,which makes our analysis more stable than the typicalaccounting approach, which depends on the vagaries of thefmancial marketplace.We use data from one traditional and four Intemet firmsin our empirical application. Our analysis reveals severalnotable results. First, we find that our estimates of customervalue are reasonably close to the market valuation at thetime of our study for three of the five firms. In contrast, tra-ditional valuation methods do not work well for the valua-tion of many of the firms because most of them have nega-tive eamings. The results show that customer-based metricsare still value relevant. Second, consistent with previousstudies in marketing, we find that retention has a significantimpact on customer value. Specifically, we find that reten-tion elasticity is in the range of 3 to 7 (i.e., a 1% improve-ment in retention increases customer value by 3%-7%). Incontrast, we find m argin elasticity to be1 and acquisitioncost elasticity to be only .02% to .3%. Notably, the marketappears to have treated marketing (and customer acquisi-tion) expenditures as investment before the Intemet crashbut now treats them as expenses. Our results indicate thatreducing acquisition costs may not be the most effective wayfor firms to improve value. Furthermore, to the extent thatcustomers are assets, the m arket may be incorrect in treatingcustomer acquisition costs as current expenses rather thaninvestments. Third, we find that the retention rate has a sig-nificantly larger impact on customer and firm value thandoes the discount rate or cost of capital. Financial analystsand company managers spend considerable time and effort

in measuring and managing discount rate because theyunderstand its impact on firm value. However, our resultsshow that it is perhaps more important not only for market-ing managers but also for senior m anagers and fmancial ana-Table 6

CUSTOM ER VALUE AT TYPICAL RETEN TION AND DISCOUNT RATES ( BILLIONS)

Rate8%

12%16%

Amazon.com:Retention Rate70 80.90 1.25.82 1.10.75 ,98

901,971,621,38

Ameritrade:Retention Rate70 80.71 ,97.65 ,85,59 .76

901,521,251,06

Capital One:Retention Rate707.336.215,35

8010.958.887,39

9018.9414,1411,04

eBay:Retention Rate701,561.391.29

802.181.891,70

903.472.822,41

E*Trade:Retention Rate701.07.98.90

801.521.341.20

902,462,031,73


12/13

18 JOURNAL OF MARKETING RESEARCH FEBRUARY 200

lysts to pay close attention to a firm's customer retentionrate.

We acknowledge several limitations of our study. First,we had several quarters of data that enabled us to provide agood estimate of the number of future customers, which isan important input to our valuation model. The accuracy ofthe model would be hampered significantly in the earlystages of a firm, when there is only limited information. Thisis similar to forecasting demand for an innovation with onlya few data points. Advances in diffusion modeling suggestthat in these cases, it may be desirable to use a Bayesianapproach in which previous studies can provide informativepriors (Lenk and Rao 1990; Sultan, Farley, and Lehmann1990).Such an approach would be a useful extension in ourcase as well. A second limitation of our study is the assump-tion of a constant retention rate. This assumption impliesthat as a firm reaches maturity and its customer acquisitionslows,it will eventually lose all its customers as a result ofa constant defection rate. This aspect is likely to have asmall impact on our valuation, because this effect occursonly in the distant long run and the future events have min-imal impact on value as a result of discounting. Nonetheless,further research should examine this issue in greater detail.For example, two possible ways to alleviate the impact ofthis assumption is to have either dynamic retention rates orgrowth in market size. We also ignore links among acquisi-tion costs, retention rates, margins, and number of cus-tomers. In reality, we would expect correlation among thesefactors, and a model that captures these links would bevaluable.

In summary, our article provides a starting point for cus-tomer valuation and its relationship to the value of firms. Weemphasize that we do not suggest replacing traditionalfinancial models; indeed, our approach uses the well-established finance approach of DCF. However, by usingDCF at a customer level, we are able to provide a usefulmethod for forecasting the stream of future eamings, whichis a key input to any valuation model. We hope that our worksparks more interest in this area and brings the fields of mar-keting and finance closer together.

REFERENCESAmir, Eli and Baruch Lev (1996), Value-Relevance of Nonfinan-cial Information: The Wireless Communication Industry, Journal ofAccounting and Economics 22 (August-December), 3 -30.Anthony, Robert N., David F. Hawkins,andKenneth A. M erchant(1998), Accounting: Text atid Cases 10th ed. New York:Irwin/McGraw-Hill.Bass, FratikM.(1969), A New Product Growth ModelforCon-sumer Durables, Management Science 15 (5), 215-27., Dipak Jain,andTrichy Krishnan (2000), ModelingtheMarketing-Mix Influence in New-Product Diffusion, inNewProduct Diffusion Models Vijay Mahajan, Eitan Muller,andYoram Wind, eds. Boston: Kluwer Academic Publishers.Bates, Douglas M. and DonaldG.Watts (1988),Nonlinear Regres-

sionAnalysis and Its Applications.NewYork:John W iley & Sons.Berger, Paul D. and Nada I. Nasr (1998), Customer LifetimeValue: Marketing Models and Applications, Journal of Interac-tive Marketing 12 (Winter), 17-3 0.Blattberg RobertC, Gary Getz,andJacquelynS.Thomas (2001),Customer Equity: Building and Managing Relationships as aluableAssets.Boston: Harvard Business School Press.Brealey, Richard A. and Stewart C. Myers (1996), PrinciplesofCorporate Einance 5thed. New York: M fcGraw-Hill.

Capital One (2001), Annual Report, (accessed February 2003[available at http://www.capitalone.com/about/invest/annualreports/2001/].Damodaran, Aswath (2001), The Dark Sideof Valuation: ValuinOld Tech New Tech and New Economy Cotnpanies.Upper Sadle River, NJ: Financial Times/Prentice Hall.Demers, Elizabeth andBaruchLev(2001), ARude AwakeningInternet Shakeout in 2000, Review ofAccounting Studies(2-3), 331-59.Fisher, J.C. and R.H. Pry (1971), A Simple Substitution ModelfoTechnology Change, Technological Forecasting and SociaChange 3 (1) , 7 5 -88 .Eortmie(2001a), The 2001 Fortune All Stars, (June 11), 170-88(2001b), Where Mary Meeker Went Wrong, (May 1468-82.Kamakura, Wagner A, and Siva K. Balasubramanian (198 7), LongTerm Forecasting with Innovation Diffusion Models: The Impacof Replacement Purchases, Journal of Eorecasting 6(1),1-19Kaufman, Leslie (1999), Cutting ThroughFog ofGrowthforNRetailers, NewYork Ttmes (September 1),Cl.Kim, Namwoon, Vijay Mahajan, and Rajendra K. Srivastav(1995), Determiningthe Going Market Valueof aBusinessian Emerging Infomiation Technology Industry: The CaseofthCellular Communications Industry, Technological Eorecastinand Social Change 49(3), 257-7 9.Lenk, Peter J, and Ambar G. Rao (1990), New Mod els fromOldForecasting Product Adoption by Hierarchical Bayes Procedure, Marketing Science 9(W inter), 42-53 .Niraj, Rakesh, Mahendra Gupta, and Chakravarthi Narasimha(2001), Custo mer Profitability in aSupply Chain, JournaloMarketing 65(July),1-16.Reichheld, Frederick E (1996),The Loyalty Effect: TheHiddeEorce Behind Grow th Profits and Lasting Value.Boston:Havard Business School Press.Reinartz, W ernerJ.and V. Kumar (2000), On theProfitabilityoLong-Life Customersin aNoncontractual Setting: An EmpiricaInvestigationandIm plicationsforMarketing, Journal of Marketing 64(October), 17-3 5.Rust, Roland T, ValarieA.Zeithaml, and Katherine N. Lemo(2001),Driving Custotner Equity: How Customer Lifetime alIs Reshaping C orporate Strategy.New York: The Free Press.Schmittlein, David C. and Vijay Mahajan (1982), Maxim um Likelihood Estimation for an Innovation Diffusion Model of NewProduct Acceptance, Marketing Science 1 (Winter), 57-78 .Seybold, Patricia (2000), Don 't Count OutAmazon, Busines2.0 (October 10),99.Shaffer, Greg and Z. John Zhang (2002 ), Com petitive One -to-OnPromotions, Management Science 48(September), 1143-61.Srinivasan,V. andCharlotte H.Mason (1986), Nonlinear LeaSquares Estimationof New Product Diffusion Models, Marketing Science 5 (2),169-78.Srivastava, Rajendra, Tasadduq A. Shervani, and Liam Eahe(1998), Market-Based Assets and Shareholder Value:AFrameworkforA nalysis, Journal of Marketing 62(January), 2 -18 .Sultan, Eareena, JohnU.Farley,andDonaldR. Lehmann (1990 A Meta-Analysis ofApplicationsofDiffusion Mod els, Journal of Marketing Research 27(February), 70-77.Thom as, Jacquelyn (2001), A MethodologyforLinking Custom eAcquisition to Customer Retention, Journal of MarketinResearch 38(May), 262-68 .Trueman, Brett, M.H. Franco Wong,andXiao-Jun Zhang (2000 The Eyeballs Have It: Searching for the Value in IntemStocks, Review ofAccounting Studies 38 (Supplement), 1 37-62The Wall Street Journal (1999), Buying theBuyers:TheGoThese Days Seemsto Be toAttract Custome rs, Whatever TheCost You, (Novem ber 22),BL


13/13

6-Valuing Customers Gupta

Documents

Transcript of 6-Valuing Customers Gupta