ArticlePDF Available

The Impact of Free-Trial Acquisition on Customer Usage, Retention, and Lifetime Value

Article

The Impact of Free-Trial Acquisition on Customer Usage, Retention, and Lifetime Value

Abstract and Figures

Many service firms acquire customers by offering free-trial promotions. A crucial yet unexplored question is whether customers acquired by free trials differ systematically from regular customers in terms of their usage and retention behavior, and customer lifetime value. To address this issue, the authors conceptualize how a consumer’s retention decision is driven by marketing communication and usage. They next develop hypotheses how the effects of these drivers are moderated by the mode of acquisition, i.e., free-trial or regular. To test the hypotheses, the authors model a customer’s retention decision and usage behavior of both flat-rate (e.g., watching TV programs), and pay-per-use services (e.g., watching videos-on-demand). The model allows for unobserved heterogeneity, selection effects, and endogenous marketing instruments. On the basis of panel data from a digital TV service, the analyses demonstrate behavioral differences which make free-trial customers, on average, worth 55% less than regular customers. However, free-trial customers are more responsive to changes in marketing communication and usage rates, which offers opportunities to target marketing efforts and enhance customer equity.
Content may be subject to copyright.
Journal of Marketing Research
Vol. LII (April 2015), 217–234
217
© 2015, American Marketing Association
ISSN: 0022-2437 (print), 1547-7193 (electronic)
*Hannes Datta is Assistant Professor, Department of Marketing, Tilburg
University (e-mail: h.datta@tilburguniversity.nl). Bram Foubert is Assis-
tant Professor, Department of Marketing and Supply Chain Management,
Maastricht University (e-mail: b.foubert@maastrichtuniversity.nl). Harald
J. van Heerde is Research Professor of Marketing, Massey Business
School, Massey University, and Extramural Fellow at CentER, Tilburg
University (e-mail: heerde@massey.ac.nz). The authors gratefully
acknowledge Marnik Dekimpe, Aurélie Lemmens, and Sungho Park and
thank Johannes Boegershausen, Kelly Geyskens, Caroline Goukens, Anne
Klesse, and Arjen van Lin for helpful comments on a previous draft. The
first author acknowledges the financial support of Maastricht University’s
Graduate School of Business and Economics and the Netherlands Organi-
zation for Scientific Research (NWO Vici Grant 453-09-004). This work
was carried out on the Dutch national e-infrastructure with the support of
SURF Foundation. Peter Verhoef served as associate editor for this article.
HANNES DATTA, BRAM FOUBERT, and HARALD J. VAN HEERDE*
Many service firms acquire customers by offering free-trial promotions.
However, a crucial challenge is to retain the customers acquired with
these free trials. To address this challenge, firms need to understand
how free-trial customers differ from regular customers in terms of their
decisions to retain the service. This article conceptualizes how marketing
communication and usage behavior drive customers’ retention decisions
and develops hypotheses about the impact of free-trial acquisition on this
process. To test the hypotheses, the authors model a customer’s
retention and usage decisions, distinguishing usage of a flat-rate service
and usage of a pay-per-use service. The model allows for unobserved
heterogeneity and corrects for selection effects and endogeneity. Using
household panel data from a digital television service, the authors find
systematic behavioral differences that cause the average customer
lifetime value of free-trial customers to be 59% lower than that of regular
customers. However, free-trial customers are more responsive to
marketing communication and usage rates, which offers opportunities to
target marketing efforts and enhance retention rates, customer lifetime
value, and customer equity.
Keywords: free trials, customer retention, usage behavior, customer
lifetime value, acquisition mode
Online Supplement: http://dx.doi.org/10.1509/jmr.12.0160
The Challenge of Retaining Customers
Acquired with Free Trials
A popular way to acquire new customers, especially
among service providers, is to offer free-trial promotions.
Customers on a free trial are allowed to try the service for a
limited amount of time at no charge. Well-known examples
are the free trials offered by mobile telephone operators
(e.g., AT&T in the United States), video streaming websites
(e.g., Netflix), and digital television providers (e.g., Sky
News in New Zealand). Although these free trials may be
popular with consumers, a crucial challenge for firms is to
retain customers who have been acquired with a free trial.
To address this challenge, firms need to understand whether
customers attracted with free trials are systematically differ-
ent from other customers. In this article, we argue that free-
trial acquisition may affect the nature of a customer’s rela-
tionship with the service provider and, as a consequence,
influence usage and retention behavior, consumers’ respon-
siveness to marketing activities, and—ultimately—customer
lifetime value (CLV).
An emerging body of research has shown that the condi-
tions under which customers are acquired have implications
for subsequent consumer behavior (e.g., Reinartz, Thomas,
and Kumar 2005; Schweidel, Fader, and Bradlow 2008).
The first group of studies documents the role of the sales
channel through which customers are attracted (e.g., Steffes,
Murthi, and Rao 2011). For example, Verhoef and Donkers
(2005) find that acquisition through the Internet leads to
higher retention rates than acquisition through direct mail or
direct-response commercials. The second set of articles
addresses the impact of customer referral (Chan, Wu, and
Xie 2011; Schmitt, Skiera, and Van den Bulte 2011). Vil-
lanueva, Yoo, and Hanssens (2008), for example, show that
customers acquired through word-of-mouth referral have
longer lifetimes with the firm.
The third stream of research, in which we position our
own work, examines the effects of the price structure or pro-
motional conditions under which customers are acquired.
Iyengar et al. (2011) find that customers of a telecommuni-
cation company who were acquired under a two-part tariff
structure have lower usage and retention rates than cus-
tomers charged on a pay-per-use basis.1Lewis (2006)
shows that acquisition discounts lead to lower retention
rates, while Anderson and Simester’s (2004) results indicate
that promotionally acquired customers choose cheaper
products but buy more. Table 1 summarizes the relevant
research.
This article contributes to the literature in three ways.
First, whereas previous work has examined the impact of
promotional customer acquisition on subsequent behavior,
the effects of free-trial promotion have remained largely
unaddressed. A free trial involves a distinct type of sales
promotion that enables consumers to start using a service
without a financial obligation and to revise their initial
adoption decision if they are not satisfied. A free trial thus
allows consumers to engage in a low-commitment relation-
ship with the firm (Dwyer, Schurr, and Oh 1987). Relying
on buyer–seller relationship theory, we argue that this type
of sales promotion may lead to systematic differences in
behavior between free-trial and regular customers.
Although some studies have examined the effects of free
trials and sampling, most focus on aggregate sales (e.g.,
Heiman et al. 2001; Jain, Mahajan, and Muller 1995;
Pauwels and Weiss 2008) or immediate purchase effects
(Scott 1976). Gedenk and Neslin (1999), who do study indi-
vidual customer behavior, find that sampling in the mineral
water category reinforces choice probabilities after the pro-
motion. However, they do not examine retention, because this
is not relevant for fast-moving consumer goods. Bawa and
Shoemaker (2004) show that free samples attract new buy-
ers who may remain customers in subsequent periods. Yet it
is unclear whether the retention rates of customers attracted
with a sample differ from those of regular customers.
As a second contribution, we extend insights on the role
of usage behavior in the customer value generation process.
Specifically, usage intensity can be an important driver of
retention because it reminds customers about the personal
value of the service (e.g., Bolton and Lemon 1999; Prins,
Verhoef, and Franses 2009). This article adds to these
insights by examining how free-trial acquisition influences
the relationship between usage and retention. In particular,
if this relationship turns out to be particularly strong for cus-
tomers acquired through free trial, it is in the firm’s interest
to encourage usage among these customers. An important
consideration in this respect is that many services involve
two types of usage: flat-rate usage (e.g., a regular television
subscription) and pay-per-use consumption (e.g., video on
demand [VOD]). Although both types of usage drive reten-
tion, a pay-per-use service is also a direct source of revenues
(e.g., Danaher 2002; Iyengar et al. 2011). Therefore, we dis-
tinguish between flat-rate and pay-per-use service compo-
nents and examine the role of usage not only as an antecedent
of retention but also as a direct component of CLV.
Third, a crucial yet unexplored question is whether acqui-
sition mode affects customers’ responsiveness to the firm’s
marketing communication efforts. Therefore, we evaluate
the differences in marketing responsiveness between free-
trial and regular customers. Specifically, we consider cus-
tomers’ reactions to direct marketing and traditional adver-
tising because of the increased interest in marketing
communication as a way to actively manage customers’
tenure (e.g., Polo, Sese, and Verhoef 2011; Reinartz,
Thomas, and Kumar 2005). If free-trial and regular cus-
tomers respond differently to direct marketing and advertis-
ing, firms may decide to target retention efforts to the most
receptive group to reduce churn.
In summary, we investigate whether free-trial acquisition
influences retention behavior and CLV and explore how it
moderates the extent to which retention is driven by service
usage (flat-rate and pay-per-use) and marketing communi-
cation. We develop econometric models for customers
usage and retention decisions, accounting for unobserved
heterogeneity and endogenous marketing instruments.
Importantly, because we are interested in the impact of free-
trial acquisition on a customer’s behavior, we correct for
selection effects. In particular, free trials may attract con-
sumers with a priori lower valuations of the service (e.g.,
Lewis 2006). Whereas previous research on the role of
acquisition mode typically has ignored selection effects (see
Table 1), we consider two approaches to address these
effects: explicit modeling of the selection process (e.g.,
Thomas 2001) and matching (e.g., Gensler, Leeflang, and
Skiera 2012).
Using household panel data for more than 16,000 cus-
tomers of a large European digital television provider, we
find that free-trial customers have lower retention rates and
use the firm’s flat-rate service less intensively than regular
customers. As a result, their CLV is, on average, 59% lower
than that of regular customers. However, free-trial customers
are more responsive to marketing communication and more
likely to rely on their usage behavior when deciding
whether to retain the service. These findings offer managers
opportunities to better target their marketing efforts and
improve retention rates, CLV, and customer equity (CE).
CONCEPTUAL FRAMEWORK AND HYPOTHESES
Figure 1 presents the conceptual framework for this
research. The core consists of a customer’s usage and reten-
tion decisions, which are influenced by the acquisition
mode (i.e., free-trial vs. regular acquisition).
218 JOURNAL OF MARKETING RESEARCH, APRIL 2015
1According to economic theory, the lower marginal consumption cost of
a two-part (as opposed to pay-per-use) tariff structure should lead to higher
usage rates. Iyengar et al. (2011) explain their counterintuitive finding by
pointing out that partitioned prices draw consumers’ attention and make
them more price sensitive.
The Challenge of Retaining Customers Acquired with Free Trials 219
Acquisition Mode
Inclusion of
Usage Behavior
Impact of
Acquisition Mode
on Response to
Retention Efforts
Correction for
Selection Relevant Findings with Regard to Impact on Customer Behavior
Channel Impact
Verhoef and Donkers (2005) Acquisition through Internet, direct-response
advertising, direct marketing, and more
Acquisition through direct mail or direct-response commercials
leads to lower retention rates than online acquisition.
Steffes, Murthi, and Rao (2008) Online and direct-marketing channels
(direct mail, direct selling, telesales)
Acquisition through direct mail yields the highest retention rates.
Steffes, Murthi, and Rao (2011) Online versus direct-marketing channels
(direct mail, direct selling, telesales)
— — The online channel leads to customers with higher transaction
amounts.
Referral Impact
Villanueva, Yoo, and
Hanssens (2008)
Word of mouth versus marketing-induced
acquisition
Customers acquired through word of mouth stay longer with the
firm and generate more future referrals than other customers.
Chan, Wu, and Xie (2011) Acquisition from Google advertising Customers referred from Google search advertising have greater
retention and transaction rates than other customers.
Schmitt, Skiera, and
Van den Bulte (2011)
Acquisition through referral program Customers attracted through a referral program have higher
retention rates than other customers.
Pricing/Sales Promotion Impact
Anderson and Simester (2004) Acquisition with price discount Promotionally acquired customers choose cheaper products but
buy more (often).
Lewis (2006) Acquisition with price discount Promotionally acquired customers have lower retention rates.
Iyengar et al. (2011) Acquisition under two-part versus pay-per-use
pricing
Not
necessarya
Customers acquired under two-part pricing have lower usage and
retention rates than customers charged on pay-per-use basis.
The current research Acquisition with free trial ✓✓✓Free-trial customers have higher churn rates and lower CLVs but
are more responsive to marketing communication and to own
usage behavior.
Table 1
COMPARISON OF THIS ARTICLE WITH EXISTING STUDIES ON THE RELATIONSHIP BETWEEN ACQUISITION MODE AND CUSTOMER BEHAVIOR
aCustomers are randomly assigned to one of two pricing structures.
Core Decision Process
The core decision process involves two types of periodic
(e.g., monthly) decisions. Every period, consumers decide (1)
how intensively to use the service and (2) whether to retain it.
Service usage. We distinguish two types of service usage
that are common for subscription services: (1) usage of a
flat-rate service, which is included in the subscription
charges, and (2) usage of a pay-per-use service, for which
consumers pay per unit of consumption. Although both
types of usage may foster retention, consumption of the
pay-per-use service also directly generates revenue.
Service retention. Every period, consumers decide
whether to retain the service. We differentiate two sets of
drivers for this decision. First, consumers rely on their
usage intensity for the flat-rate and pay-per-use component
to assess the utility of retaining the service (Bolton and
Lemon 1999). As a result, a high usage intensity will stimu-
late retention, whereas a low usage rate may lead to dis-
adoption (Lemon, White, and Winer 2002). Note that, com-
pared with the pay-per-use service, flat-rate usage may be
more consequential for customers’ evaluation of the service
subscription because it is included in the fixed periodical
fee (Bolton and Lemon 1999). Second, marketing commu-
nications also influence a consumer’s retention decision
(e.g., Blattberg, Malthouse, and Neslin 2009). Specifically,
direct marketing and advertising remind customers of the
benefits of the service or directly persuade them to retain it.
If consumers retain the service in the current period, they
go through the same usage and retention decision process in
the following period. As the dashed lines in Figure 1 indi-
cate, this repeated decision process drives CLV. In particu-
lar, the periodic retention decisions generate a stream of
fixed subscription fees (which cover flat-rate usage),
whereas usage of the pay-per-use service generates addi-
tional revenue.
Differences Between Free-Trial and Regular Customers
Central to our study is the expectation that the decision
process to use and retain the service differs between free-
trial and regular customers. Our hypotheses build on buyer–
seller relationship theory, which posits that customer behav-
ior depends on the nature of the relationship between
customer and firm (Dwyer, Schurr, and Oh 1987; Johnson
and Selnes 2004).
Baseline retention. Drawing on relationship theory, we
expect free-trial customers to churn sooner than regular cus-
tomers. Whereas the anticipated longevity of a regular con-
tract encourages customers to immediately commit to the
firm, subscription to a free trial resembles a discrete transac-
tion that merely increases a consumer’s awareness of the
firm and facilitates relationship exploration (Dwyer, Schurr,
and Oh 1987; Johnson and Selnes 2004). According to self-
perception theory, free-trial customers may make post hoc
inferences about the reasons for their behavior and thus
attribute their adoption decision to the availability of a free
trial rather than to a strong commitment to the company
(e.g., Dodson, Tybout, and Sternthal 1978; Gedenk and
Neslin 1999). In other words, a free trial decelerates the
relationship formation process (Palmatier et al. 2013).
Importantly, even after the free-trial period has expired, the
firm’s relationship with free-trial customers likely remains
more fragile than that with regular customers. In particular,
220 JOURNAL OF MARKETING RESEARCH, APRIL 2015
Figure 1
CONCEPTUAL FRAMEWORK
The Challenge of Retaining Customers Acquired with Free Trials 221
research by Gilbert and Ebert (2002) and Gilbert et al.
(1998) indicates that consumers who receive the opportu-
nity to first evaluate a product or service are more critical
than when they immediately commit to the firm, a tendency
that persists after the evaluation period. That is, the critical
reflections generated during exploration of a relationship
remain active even when the customer moves to a closer
relationship level. Thus, we hypothesize the following:
H1: Free-trial customers have a lower retention rate than regular
customers even after the free trial expires.
Impact of usage on retention. Because customers
attracted with a free trial arguably have a less developed
relationship with the firm than regular customers, they may
be more uncertain about the service benefits (Johnson and
Selnes 2004). A major factor that informs consumers about
the personal value of the service and thus helps resolve the
uncertainty is their own usage behavior (e.g., Bolton and
Lemon 1999). A customer may wonder, “Do I use the service
enough to stay subscribed?” We expect that, to overcome
their uncertainty, free-trial customers are more inclined than
regular customers to assess the service’s value on the basis
of their flat-rate and pay-per-use consumption. Regular cus-
tomers, who are more committed to the firm, are less likely
to base their retention decision on usage intensity. Thus, we
expect the following:
H2a: The impact of usage of a flat-rate service on retention is
greater for free-trial customers than for regular customers.
H2b: The impact of usage of a pay-per-use service on retention is
greater for free-trial customers than for regular customers.
Impact of marketing communication on retention. We
also argue that the firm’s marketing communication, in the
form of direct marketing and advertising, will be more
important to free-trial customers than to regular customers.
Marketing communication provides free-trial customers
with information that can compensate for their relatively
high uncertainty (Mitchell and Olson 1981). Regular cus-
tomers, in contrast, may be less susceptible to external
information because of confidence in their level of expertise
(Brucks 1985; Hoch and Deighton 1989). In line with these
principles, Johnson and Selnes (2004) postulate that it is
easier to boost commitment among customers in a lower-
level relationship with the firm than among already dedi-
cated customers. As a result, we expect that free-trial cus-
tomers are more responsive to the firm’s direct-marketing
and advertising efforts than regular customers:
H3a: The impact of direct marketing on retention is greater for
free-trial customers than for regular customers.
H3b: The impact of advertising on retention is greater for free-
trial customers than for regular customers.
Baseline usage. Free-trial acquisition may also affect cus-
tomers’ usage intensity. On the one hand, free-trial cus-
tomers are less committed to the firm and less convinced of
the service benefits, so they may have lower usage rates
than regular customers. Indeed, usage is one of the most
tangible reflections of engagement with the firm (Van
Doorn et al. 2010). This holds true in particular for usage of
the flat-rate service, which is the main object of the contrac-
tual relationship (Bolton and Lemon 1999). On the other
hand, exactly because free-trial customers’ relationship with
the firm is more exploratory (Dwyer, Schurr, and Oh 1987;
Gilbert et al. 1998), they may use the service more fre-
quently to become more certain about its benefits. These
opposing principles do not allow us to develop unidirec-
tional expectations regarding the impact of free-trial acqui-
sition on usage of the flat-rate and pay-per-use services.
DATA
Study Context
We test the hypotheses using a household panel data set
from a large European interactive TV (iTV) provider. The
iTV technology enables customers to interact with their
television, for example, by browsing an electronic program
guide or watching VOD. Furthermore, iTV offers enhanced
image quality over regular television. To use the iTV service,
customers need a broadband digital subscriber line Internet
connection from the same company and a set-top box that
decodes the digital signal. The focal company is the only
provider of digital television through a digital subscriber
line; at the end of the observation period, it had a market
share of 31% in the digital television market. Its main com-
petitor, which offers digital television through cable, had a
market share of 40%.2
Under regular conditions, the company’s customers for-
mally commit to a 12-month subscription period. They can
opt to cancel the service earlier, in which case they pay a
penalty (€50, plus €6 for every remaining month until the
end of the contractual period). After the first 12 months, the
contract is automatically renewed but can be terminated
each month without penalty. Customers are charged a one-
time setup fee for hardware and activation (on average,
€16.24) and pay for service usage according to a two-part
tariff structure (e.g., Ascarza, Lambrecht, and Vilcassim
2012; Iyengar et al. 2011). Specifically, the fixed monthly
subscription fee of €15.95 covers unlimited usage of the
basic iTV service (€9.95) and rent of the set-top box (€6).
In addition, customers can make use of a VOD service, for
which they are charged on a pay-per-use basis. They can
select VODs from an electronic catalog containing movies,
live concerts, and soccer games. Rental of VOD for 24
hours costs approximately €3, with some limited variation
in price due to differences in genre and length.
The company’s acquisition strategy offers a unique set-
ting in which to study the impact of free-trial acquisition.
For a period of 10 months (months 10–19 after launch of the
service), the company offered free trials parallel to its regu-
lar subscriptions. Adoption of the free trial (as opposed to
the regular subscription) is largely driven by consumers’
awareness of the ongoing free-trial promotion, which was
mainly promoted through direct marketing. Customers
acquired with a free trial did not pay setup costs and were
not charged monthly subscription fees for the usage of the
flat-rate service during a three-month period. However,
VOD usage was not free of charge. Free-trial customers
could revise their adoption decision by returning the set-top
box to one of the company stores before the end of the trial
period without paying a penalty. If the product was not
2The remaining 29% is captured by smaller players operating through
satellite or cable.
returned by the end of the three-month trial period, the sub-
scription was converted into a paid one such that the next
nine months were considered part of a regular contract.
Data Set
From the initial sample of approximately 21,000 cus-
tomers who adopted iTV when both the free trial and regular
subscription were available, we retained a subset on the basis
of several criteria. Specifically, we eliminated customers
who had missing sociodemographic information, were
employees of the focal company, or did not speak the local
language (and thus could not understand the advertising and
direct-marketing messages). We thus retained 16,512 cus-
tomers, of which 12,612 (76%) were acquired with free tri-
als and the remaining 3,900 (24%) signed a regular contract.
We observed customers’ retention and usage behavior
until two years after launch of the service. Of the free-trial
customers, 6,079 (48%) churned before the end of the
observation period, whereas only 1,327 (34%) of the regular
customers did so. Furthermore, the data set includes two
types of usage: (1) flat-rate usage of the basic iTV service,
which is measured by a customer’s monthly number of
“channel zaps” (i.e., how often a customer switches chan-
nels),3and (2) usage of the VOD service, for which we use
the monthly number of VODs the customer has watched.
Compared with regular customers, free-trial customers
average usage intensity is 11% lower for the flat-rate service
(169 vs. 189 zaps per month) but 26% higher for the VOD
service (.73 vs. .58 VODs per month). However, these aver-
age retention and usage measures are merely indicative of
the actual differences because they do not account for selec-
tion effects or the impact of marketing activities.
The company uses two types of marketing communication:
direct marketing and mass advertising. We operationalize
direct marketing as the monthly number of direct-marketing
contacts with a given customer (through phone, e-mail, or
regular mail). On average, free-trial and regular customers
are contacted .34 and .16 times per month, respectively. In the
analyses, we account for these systematic differences in con-
tact frequency. In addition, the data set includes a measure
for the company’s spending on mass advertising (through
television, print media, radio, and the Internet). In particu-
lar, this variable quantifies the company’s advertising
expenditures for a given region in a given month relative to
the total advertising spending for the same region and
month by the company and its main competitor. This share-
of-voice advertising measure varies between 0 and 1 and
has an average of .74 for free-trial and .79 for regular cus-
tomers. Table 2 lists summary statistics of the variables in
the data set, and Web Appendix A reports the correlations
between the independent variables. We provide more details
on the control variables and sociodemographic variables
when we discuss the model.
MODEL
We specify a set of equations that incorporates the inter-
relationships between customers’ usage (flat-rate and pay-
per-use) and retention decisions. We account for unobserved
customer heterogeneity, the endogeneity of marketing com-
munication, and selection effects.
222 JOURNAL OF MARKETING RESEARCH, APRIL 2015
3We measure a customer ’s active use of the flat-rate service by using
monthly channel zaps. We also have a partially observed measure of a cus-
tomer ’s passive use. Specifically, for a period of just six months, we
observe the variable hours, capturing the monthly number of hours that the
set-top box was switched on. However, this number may not be an accurate
indication of the time that the customer actively watched television: it was
technically possible to switch off the television while leaving the set-top
box switched on, making this variable less than ideal. For the months in
which we have both zaps and hours, we find a significant positive correla-
tion of .66 (p< .01).
Table 2
DESCRIPTIVE STATISTICS FOR FREE-TRIAL AND REGULAR CUSTOMERS
Free-Trial Customers Regular Customers
M SD M SD
Dependent Variables
Retention probability .93 .25 .96 .19
Usage of flat-rate service (channel zaps) 169.35 176.90 189.28 183.80
Usage of pay-per-use service (number of VODs) .73 2.19 .58 1.91
Marketing Communication
Direct marketing (number of direct-marketing contacts) .34 .55 .16 .41
Advertising (share of voice) .74 .28 .79 .26
Customer-Specific Variables
Age (years) 46.22 12.64 44.89 12.60
Household size 2.98 1.48 2.88 1.53
Annual income (in €10,000) 2.44 .56 2.43 .59
Time to adoption (in months following the launch of the service) 13.14 2.45 12.38 2.20
Control Variables
Monthly subscription fee (in €) 9.66 7.80 13.22 4.85
Cancellation penalty (termination fees – future fees, in €) –59.80 68.08 –8.14 25.02
VOD credit (in €) 5.50 8.90 1.38 5.08
Temperature (in degrees Celsius) 12.82 4.97 12.83 4.99
Time since adoption (in months) 5.34 3.52 5.95 3.68
Direct-marketing contacts before acquisition .62 .61 .26 .48
Advertising (share of voice) before acquisition .73 .27 .76 .27
Total fees for regular 12-month contract at the time of acquisition (in €) 193.83 50.04 181.97 44.79
Number of customers 12,612 3,900
The Challenge of Retaining Customers Acquired with Free Trials 223
Retention
The probability of retention is modeled with a binomial
probit model. Each customer i decides at the end of every
month t after acquisition whether to retain the service (rit =
1) or disadopt (rit = 0). We express the utility vit of retaining
the service as follows:
(1) vit= a0i+ a1iTriali+ a2iUsageFRit+ a3iUsagePPUit
+ a4iDMit + a5iAdvit + a6iUsageFRit ¥Triali
+ a7iUsagePPUit ¥Triali+ a8iDMit ¥Triali
+ a9iAdvit ¥Triali+ a10, iInitialit + a11, iFeeit
sub
+ a12, iPenaltyit + a13, iTempit + a14, ilog(Timeit) + xit.
Thus, the utility of retaining the service at the end of month
t is influenced by the dummy Triali(1 if customer i was
acquired with a free trial; 0 otherwise). This enables us to test
whether free-trial acquisition increases a customer’s churn
rate (H1). Other drivers include the customer ’s usage of the
flat-rate (UsageFRit, measured in monthly channel zaps
divided by 100) and pay-per-use (UsagePPUit, measured in
number of VODs) services in month t. Retention also
depends on the company’s monthly direct-marketing efforts
(DMit, the number of direct-marketing contacts received by
customer i) and advertising intensity (Advit, the company’s
share of voice in customer i’s region).4Equation 1 also
includes the interactions between Trialiand the usage and
marketing communication variables to test H2a–b and H3a–b,
which posit that the impact of usage and marketing commu-
nication is stronger for free-trial than for regular customers.
Finally, the equation includes a set of control variables.
The model accounts for a general pattern of high defection
rates during the first four months of a customer’s tenure
through the dummy variable Initialit. The subscription fee
for customer i in month t, Feeit
sub, captures the influence of
price on a customer’s retention decision (Ascarza, Lam-
brecht, and Vilcassim 2012). Variation in Feeit
sub is due to
temporary price reductions and the zero-price in the begin-
ning of a free-trial customer’s tenure. Moreover, we include
the variable Penaltyit to account for the fact that customers
were able to cancel their 12-month subscription by paying
an early-termination fee. We assume that customers trade
off the termination fee against future subscription fees
within the current contractual period. Thus, Penaltyit equals
the termination fee for immediate disadoption minus the
sum of all future subscription fees that the customer would
have to pay during the remaining months of the contractual
period. The higher Penaltyit , the more likely it is that the
customer retains the service.
Importantly, the variables Feeit
sub and Penaltyit control for
the systematically higher defection rates of free-trial cus-
tomers compared with regular customers during the free-
trial period. In this period, free-trial customers face zero
subscription and cancelation fees compared with nonzero
subscription and cancelation fees for regular customers.
Thus, these control variables enable us to obtain a clean test
of H1through the trial dummy in Equation 1, which equals
1 for a free-trial customer even after the free-trial period is
over.
We include the monthly average temperature (Tempit) to
control for seasonality,5and the log of time since adoption
(Timeit) to accommodate fluctuations in the baseline reten-
tion probability (Prins, Verhoef, and Franses 2009). The
model coefficients a0 i, ..., a14, i are customer specific and
normally distributed. Finally, the probit error term xit is nor-
mally distributed, with a standard deviation set equal to 1
for identification purposes.
Usage of the Flat-Rate Service
We model flat-rate usage (UsageFRit) as a log-log regres-
sion to account for the skewed nature of this variable (Iyen-
gar et al. 2011):
(2) log(UsageFRit + 1) = b0i + b1iTriali+ b2ilog(UsageFRi, t – 1 + 1)
+ b3ilog(Feeitsub + 1) + b4ilog(Tempit) + b5ilog(Timeit) + qit.
The free-trial acquisition dummy Trialicaptures differences
in flat-rate usage between free-trial and regular customers.
The lagged dependent variable UsageFRi, t – 1 accounts for
persistence in usage behavior. We also include the control
variables subscription fee, average temperature, and time
since adoption.6Before taking the logarithm, we add 1 to all
variables for which zeros occur (e.g., Iyengar et al. 2011).
b0i, ..., b5i are normally distributed customer-specific coeffi-
cients, and qit is an error term following a normal distribu-
tion, N(0, s2).
Usage of the Pay-per-Use Service
We model usage of the pay-per-use component (i.e., a
consumer’s monthly number of VODs) with a zero-inflated
Poisson model, in which the zero inflation accommodates
the spike at zero in the VOD usage variable:
(3) UsagePPUit = UsagePPU*
it ~ Poisson(lit) with probability qi
UsagePPUit = 0 with probability 1 -qi,
where qiis the probability that customer i is a potential user
of the VOD service, modeled with a probit structure with a
customer-specific normally distributed intercept, and
UsagePPU*
it is the number of VODs watched by customer i
in period t, given that the customer is a potential VOD user.
The expected number of VODs, lit, is specified as an expo-
nential function to ensure a positive sign:
(4) lit = exp[g0i + g1iTrial
i+ g2iUsagePPU
i,t – 1 + g3iFeeit
sub
+ g4iTempit + g5ilog(Time
it) + g6iCreditit].
4We also estimate a retention model with lagged effects of DM and Adv,
leaving model fit virtually unaffected (DHit probability = .0002, DTop-
decile lift = –.0418, DGini = .0016). We thus opt for the more parsimonious
Model 1.
5To account for differences in the length of each month and the occur-
rence of holidays, we also estimate retention and usage models with
monthly dummy variables. However, the results are very similar and model
fit h ardly changes (DHit probability = .0004, DTop-decile lift = –.0748,
DGini = .0009, DrFla t rate = .0242, DrPay per use = .0031). In addition, the
model becomes cumbersome to estimate with 66 seasonality parameters
(11 heterogeneous month coefficients in three equations). We therefore
remain with the more parsimonious in which Tempit captures seasonality.
6We also estimate models in which we allow direct marketing and adver-
tising to affect service usage, but this does not lead to a notable improve-
ment in model performance (DrFlat rate = –.0020, DrPay per use = .0137).
Similar to the model for flat-rate usage, the expected
number of videos watched is a function of acquisition mode
(Triali), a customer’s prior usage (UsagePPUi,t – 1), and the
variables Feeit
sub, Tempit, and log(Timeit). Credititis the
VOD credit for customer i in month t (measured in euros).
This VOD credit is granted by the company for a maximum
period of four months to stimulate service usage. g0i, ..., g6i
are normally distributed customer-specific coefficients.
Customer Heterogeneity
We include customer heterogeneity by modeling all
response parameters (intercepts and slope coefficients) as
normally distributed across customers. To incorporate inter-
dependence between the different model components, we
allow for correlations between the intercepts.7The expected
values of the retention and usage intercepts a0i, b0i, and g0i
are functions of the concomitant customer characteristics
age, household size, income (e.g., Rust and Verhoef 2005),
and time to adoption (Prins, Verhoef, and Franses 2009;
Schweidel, Fader, and Bradlow 2008):
where Ageiis the age of customer i (in years, shortly after
service launch), Hhsizeiis the size of customer i’s house-
hold (in number of people), Incomeiis the average income
in the census block to which customer i belongs (in
€10,000), and Adopttimeiis the time-to-adoption of cus-
tomer i (measured in months following the launch of the
iTV service).
Correction for Endogeneity of Marketing Instruments
Endogeneity due to temporal correlation. The first type
of endogeneity we address involves temporal correlation of
DMit and Advit with the error term of the retention equation.
For example, the company may counteract expected
increases in the churn rate by boosting its marketing efforts.
Following Park and Gupta (2012) and Schweidel and Knox
(2013), we use Gaussian copulas to model the correlation
between marketing and the error term. Whereas classical
methods to correct for endogeneity rely on instrumental
variables to partial out the exogenous variation in the
endogenous regressors, copulas do not require instrumental
variables (Park and Gupta 2012; Schweidel and Knox
2013). In line with Park and Gupta (2012, p. 573), we add
the following regressors to Equation 1:
()
()
()
α
β
γ
=
α+α +α +α +α
β+β +β +β +β
γ+γ +γ +γ +γ
(5)
E
E
E
Age Hhsize Income Adopttime
Age Hhsize Incom e Adopttime
Age Hhsize In come Adopttime
,
0i
0i
0i
0,0 0,1 i 0,2 i 0,3 i 0,4 i
0, 0 0, 1 i 0, 2 i 0 ,3 i 0, 4 i
0,0 0,1 i 0, 2 i 0,3 i 0,4 i
where F1is the inverse of the normal cumulative distribu-
tion function, and HDM(•) and HAdv(•) represent the empiri-
cal cumulative distribution functions of direct marketing
and advertising, respectively.8For identification purposes,
the endogenous regressors must be nonnormally distributed
(Park and Gupta 2012), which a Shapiro–Wilk test shows to
be the case (direct marketing: W = .8255, p< .001; advertis-
ing: W = .7948, p< .001).
Endogeneity due to cross-sectional correlation. Second,
endogeneity may arise from cross-sectional correlation of
the marketing activities with the random intercept. Specifi-
cally, the company may target its direct-marketing efforts on
the basis of consumer characteristics (unobserved to the
researcher) that correlate with customers’ churn rates. To
address this type of endogeneity, we follow Mundlak (1978)
and include the average number of direct-marketing con-
tacts per customer, DMi, as a covariate in Equation 1 (e.g.,
Risselada, Verhoef, and Bijmolt 2014). Because the focal
firm uses advertising as a mass-communication device,
Advit is not subject to this type of endogeneity. In the esti-
mation, we assess the added value of the Mundlak correc-
tion for cross-sectional endogeneity, in addition to the cor-
rection for temporal endogeneity.9
Correction for Selection
Selection model. We use two alternative approaches to
correct for selection effects: a selection model and match-
ing. The selection model approach estimates the retention
and usage models jointly with an additional model for con-
sumers’ selection into the free-trial or regular customer
group. By allowing for correlation between the error of the
selection equation and the random intercepts of the usage
and retention models, we account for selection effects due
to unobserved variables (Thomas 2001). Because the selec-
tion is a single event (sign up as a free-trial or regular cus-
tomer), we need to allow for correlations with the random
intercepts in retention and usage rather than with the time-
varying error terms. To model whether a customer was
acquired with a free trial (Triali= 1) or not (Triali= 0), we
use a binary probit structure and write the underlying utility
of free-trial acquisition as follows:
(7) wi= w0+ w1Agei+ w2Hhsizei+ w3Incomei+ w4Adopttimei
+ w5DMi*+ w6Advi*+ w7Feei*+ zi.
[]
[]
()
()
(6) DM H DM and
Adv H Adv ,
it 1DM it
it 1Adv it
224 JOURNAL OF MARKETING RESEARCH, APRIL 2015
8Following Park and Gupta (2012, footnote 3 and expression 10), we use
empirical instead of estimated densities to generate DM
~
it and Adv
~
it. To
keep estimations tractable, we treat DM as a continuous variable and use
regular standard errors instead of bootstrapped standard errors (shown to
be virtually the same; see Park an d Gupta 2012). We test the retention
model on a simulated data set with endogenous regressors and recover the
true parameters well.
9In the models in which we correct for both temporal and cross-sectional
correlation, we compute DM
~
it using the mean-centered direct-marketing
variable: mean-centering ensures that we only retain the part of the original
direct-marketing variable that may be correlated with the error term xit in
Equation 1 (Mundlak 1978).
7To determine whether we should structurally account for any correlation
between the slope coefficients, we inspected the correlations between the
consumer-specific posterior slopes (Train 2009) and found the mean absolute
correlation to be very small (.0084). Furthermore, the low correlation
between the residuals of the usage models (.0970) suggests that, after con-
trolling for cross-sectional correlation, there is not much interdependency
between the error terms left.
The Challenge of Retaining Customers Acquired with Free Trials 225
The drivers include age, household size, income, time to
adoption, and three marketing variables. DMi
*represents the
average number of direct-marketing contacts received by
customer i in the three months before signing up. Because
the free trial was often promoted in direct-marketing con-
tacts (e.g., in outbound telephone calls), DM i
*likely has a
positive impact on consumers’ awareness of the trial offer.
Advi
*is the average share of voice for customer i in the three
months before signing up. Because advertising usually pro-
moted the regular offer, we expect Advi
*to decrease the
probability that a customer was acquired through a free trial.
Feei
*refers to the total fees for a regular 12-month subscrip-
tion at the time of customer i’s sign-up. Higher fees for the
regular subscription may lead consumers to search longer
for a special deal or push harder when in touch with a cus-
tomer service agent such that Feei
*should have a positive
effect on the probability of free-trial acquisition. Finally, zi
is a standard-normal error term.
Matching procedure. As an alternative to jointly estimat-
ing the selection, retention, and usage models, we apply a
matching procedure that pairs free-trial customers with the
most comparable regular customers. We estimate the mod-
els on this matched data set such that differences in reten-
tion and usage behavior can be attributed to the acquisition
mode rather than to differences in sample composition
(Gensler, Leeflang, and Skiera 2012, 2013). To match cus-
tomers, we first estimate Equation 7 to compute their
propensities of adopting the service on a free trial. Follow-
ing Gensler, Leeflang, and Skiera (2013), we apply a hybrid
procedure that combines customers on the basis of not only
their propensity scores but also drivers of the selection
process (i.e., the variables DMi
*, Advi
*, and Feei
*). Next, we
compute the Mahalanobis distances among consumers on
the basis of their propensity scores and these three
variables.10 Finally, we use the one-nearest neighbor algo-
rithm to identify 12,610 pairs of free-trial customers and
their most comparable regular customers (Gensler, Leeflang,
and Skiera 2012).
EMPIRICAL RESULTS
We use simulated maximum likelihood with Halton draws
to calibrate the usage and retention models. Before dis-
cussing parameter estimates, we check model performance
and the robustness of the results across alternative
approaches of addressing endogeneity and selection. We
then evaluate model fit for the selected approach in detail
and discuss the estimation results.
Robustness Checks and Model Selection
We compare two ways to correct for marketing endo-
geneity (copula and Mundlak terms versus a more parsimo-
nious approach with just the copula terms) and two ways to
address selection effects (the selection model vs. matching).
We discuss model performance and the robustness of the
results across all four combinations. To assess model per-
formance, we evaluate the retention model’s in- and out-of-
sample fit (cross-sectional and longitudinal) on three fit
measures: hit probability (Gilbride, Allenby, and Brazell
2006), top-decile lift, and Gini coefficient (Lemmens and
Croux 2006). Although in- and out-of-sample fit measures
are not suited to compare models with and without endo-
geneity correction, they can be used to validate different
approaches that all correct for endogeneity (Ebbes, Papies,
and Van Heerde 2011). Table 3 shows that, using matched
samples and including both copula and Mundlak terms,
model M1 outperforms the other models for seven of nine
fit criteria. Therefore, the remainder of our discussion
focuses on model M1. Note, however, that the four models
yield comparable outcomes for the hypothesis tests, under-
scoring the robustness of the results.
Estimation Results
Table 4 reports in- and out-of-sample fit measures for the
selected model M1 in more detail. In addition to the measures
for the retention model, we use the correlation between
actual and predicted values as a fit measure for usage of the
flat-rate and pay-per-use services. Overall, the different
measures suggest a good in- and out-of-sample model fit for
both free-trial and regular customers. The fit of the auxiliary
propensity model (Equation 7) is adequate (hit probability:
66.71%), and the parameter estimates, reported in Web
Appendix B, are face-valid.
Table 5 presents the parameter estimates for the focal
models. For the heterogeneous parameters, we focus on the
population means. We first discuss the relationships of the
conceptual framework and then address the effects of
the control variables. Hypothesis tests are one-sided, and
the remaining tests are two-sided.
Core decision process. The results confirm the expecta-
tions for the core retention decision process. Retention is
positively affected by flat-rate usage and usage of the pay-
per-use service (a2= .728, p< .001 and a3= .039, p< .001,
respectively). Usage drives retention because, to the con-
sumer, it is an indication of the personal value of the service
(e.g., Lemon, White, and Winer 2002). The impact of both
direct marketing (a4= 1.565, p< .001) and advertising (a5=
.310, p< .001) on retention is positive, which is consistent
with previous findings that marketing communication cre-
ates interest in the service and increases retention (e.g.,
Polo, Sese, and Verhoef 2011). The significant coefficients
of the copula and Mundlak terms suggest that correcting for
marketing endogeneity is indeed required (a15 = –.596, p<
.001; a16 = –.017, p< .001; a17 = –2.111, p< .001).
Differences between free-trial and regular customers. In
support of H1, acquisition through free trials has a direct
negative impact on retention (a1= –.278, p< .001), even
after controlling for higher churn during the free-trial
period.11 This finding is in line with the expectation that
free-trial customers are less confident about retaining the
service because they are in a tentative relationship with the
firm and are likely to attribute their subscription to extrinsic
10From the 12,612 free-trial customers in the data set, we eliminated two
customers because their propensity scores did not lie within the region of
common support, a region defined as the overlap between the propensity
score distributions of the two customer grou ps (Gensler, Leeflang, and
Skiera 2012).
11Because free-trial customers may be particularly likely to churn in
their first three months, we reestimate the model only with customers who
“survived” the first three months. The results are robust to this alternative
specification (a1= –.040, z = –3.390, p< .001).
incentives (i.e., the trial), resulting in lower commitment
(Dwyer, Schurr, and Oh 1987).
The positive interaction between the free-trial dummy and
usage variables indicates that the usage effects are stronger
for free-trial than for regular customers, which confirms H2a
and H2b (flat-rate service: a6= .040, p< .001; pay-per-use
service: a7= .013, p< .001). This finding is in line with the
notion that free-trial customers, because of their lower com-
mitment, are more uncertain about the service’s benefits and
therefore rely more on their usage behavior when making
retention decisions (Bolton and Lemon 1999).
In support of H3a and H3b, free-trial customers are more
responsive to marketing communication instruments than
regular customers (direct marketing: a8= .133, p< .001;
advertising: a9= .154, p< .001). Similar to usage, advertis-
ing and direct marketing provide free-trial customers with
cues that make them more secure about the service’s value
and assist them in their retention decision (Mitchell and
Olson 1981).12
Free-trial customers use the flat-rate service less inten-
sively than regular customers (b1= –.269, p< .001). This
226 JOURNAL OF MARKETING RESEARCH, APRIL 2015
Table 3
SELECTION OF MODEL FOR RETENTION DECISION
M1 (Selected Model) M2 M3 M4
Model Specification
Correction for selection Matching Matching Joint estimation Joint estimation
Correction for marketing endogeneity Copula + Mundlak Copula only Copula + Mundlak Copula only
Hypothesis Testing
H1: Trial effect on retention < 0 -.278*** -.106*** -.255*** -.066***
H2a: UsageFR ¥Trial effect > 0 .040*** .107*** .008 .056***
H2b: UsagePPU ¥Trial effect > 0 .013** .015** .008* .013**
H3a: DM ¥Trial effect > 0 .133*** .057*** .096*** .051***
H3b: Adv ¥Trial effect > 0 .154*** .150*** .150*** .134***
Model Fita
Hit probability
In-sample .917 .914 .908 .904
Out-of-sample (cross-sectional) .922 .913 .904 .904
Out-of-sample (longitudinal) .953 .918 .948 .952
Top-decile lift
In-sample 5.853 5.308 5.776 5.153
Out-of-sample (cross-sectional) 7.111 5.130 5.800 5.306
Out-of-sample (longitudinal) 4.244 3.055 4.453 5.219
Gini coefficient
In-sample .755 .713 .748 .707
Out-of-sample (cross-sectional) .826 .711 .752 .706
Out-of-sample (longitudinal) .578 .402 .568 .664
*p< .10.
**p< .05.
***p< .01.
aWe highlight the best model fit in boldface. To compute out-of-sample fit measures, we reestimate the model on a random cross-section of 70% of the cus-
tomers or on the first 70% of the longitudinal observations and compute model fit for the holdout sample.
Notes: In the table, we use one-sided tests of significance. UsageFR = customer usage of the flat-rate service; UsagePPU = customer usage of the pay-per-
use service; DM = direct marketing; Adv = advertising.
Table 4
MODEL FIT FOR SELECTED MODEL (M1)
Flat-Rate Usage Pay per Use
Retention Correlation Between Correlation Between
Hit Probability Top-Decile Lift Gini Coefficient Actual and Predicted Actual and Predicted
In-Sample Fit
Free-trial customers .899 5.696 .750 .811 .750
Regular customers .932 5.634 .750 .815 .740
Out-of-Sample Fit (Cross-Sectional)a
Free-trial customers .906 6.946 .821 .809 .724
Regular customers .936 7.052 .819 .815 .727
Out-of-Sample Fit (Longitudinal)b
Free-trial customers .948 4.526 .599 .751 .673
Regular customers .958 3.785 .545 .778 .593
aWe reestimate the model on a random draw of 70% of the customers and compute model fit for the holdout sample.
bWe reestimate the model on the first 70% of the longitudinal observations and predict usage and retention for the holdout sample.
12We test whether the marketing response estimates are affected by the
inclusion of customers who churned during the first months after acquisi-
tion. Excluding these customers yields comparable results (a8= .046, z =
2.30, p< .05; a9= .08, z = 5.11, p< .001; both one-tailed).
The Challenge of Retaining Customers Acquired with Free Trials 227
finding is in line with the notion that free-trial customers are
less committed to the service’s benefits. Notably, however,
they use the pay-per-use service more intensively than regu-
lar customers (g1= .113, p< .001). Because free-trial cus-
tomers make less use of the flat-rate service, they may dedi-
cate more time to exploring the paid add-on service (Schary
1971). In addition, they may perceive the free trial as a
windfall gain, enticing them to spend more on the pay-per-
use service (e.g., Heilman, Nakamoto, and Rao 2002).
Figure 2 shows plots of the average predicted and
observed values for retention (Panel A), flat-rate usage
(Panel B), and usage of the pay-per-use service (Panel C).
The plots show that the model fits the retention and usage
data well for both free-trial and regular customers. In Panel
Table 5
ESTIMATION RESULTS
Population Mean SD
Estimate SE Estimate SE
Retention
Intercept 3.034*** .006 .025 .018
Age -.002*** .000
Household size -.037*** .002
Income .059*** .002
Time to adoption -.002*** .000
Trial H1: effect < 0 -.278*** .008 .016 .021
Flat-rate usage .728*** .006 .333*** .006
× Trial H2a: effect > 0 .040*** .009 .159*** .015
Pay-per-use usage .039*** .004 .015*** .005
× Trial H2b: effect > 0 .013** .005 .010 .007
Direct marketing 1.565*** .012 .190*** .034
× Trial H3a: effect > 0 .133*** .014 .086** .043
Advertising .310*** .008 .000* .027
× Trial H3b: effect > 0 .154*** .011 .080** .033
Copula correction term for DM -.596*** .005
Copula correction term for Adv -.017*** .002
Mundlak correction term for DM -2.111*** .014
Control Variables
Initial period -1.377*** .007 .040** .016
Monthly subscription fee -.008*** .000 .001 .001
Cancellation penalty .003*** .000 .000 .000
Temperature .012*** .000 .000 .001
log(Time since adoption) -.572*** .004 .022 .014
Flat-Rate Usage (Log Zaps)
Intercept 5.609*** .006 .881*** .005
Age .000*** .000
Household size .036*** .002
Income -.034*** .002
Time to adoption -.048*** .000
Trial -.269*** .008 .081*** .014
Control Variables
log(Monthly subscription fee) -.033*** .002 .004 .004
log(Temperature) -.417*** .002 .016*** .004
log(Time since adoption) -.319*** .003 .340*** .004
Lagged log(Flat-rate usage) .242*** .001 .005** .002
log(s) -.007*** .002
Pay per Use (VODs)
Intercept (incidence) –.140*** .006 1.410*** .004
Intercept (amount) -.243*** .007 1.245*** .018
Age -.014*** .000
Household size .090*** .002
Income .042*** .003
Time to adoption -.018*** .001
Trial .113*** .010 .016 .015
Control Variables
Monthly subscription fee -.006*** .000 .001 .000
VOD credit .005*** .001 .008*** .001
Temperature -.005*** .000 .015*** .001
log(Time since adoption) .053*** .003 .156*** .004
Lagged pay per use .023*** .000 .008*** .000
*p< .10.
**p< .05.
***p< .01.
Notes: Log-likelihood = –310,958; N = 196,251; Bayesian information criterion = 622,892; Akaike information criterion = 622,077. In the table, we report
two-sided tests of significance.
A, the predicted retention rates after the trial look very simi-
lar for free-trial and regular customers. However, the model
reveals that these predicted retention rates are obtained very
differently for these two customer groups. For free-trial cus-
tomers, the baseline retention rate is lower because the main
effect of Trialion retention utility is –.278. However, this
lower baseline rate is compensated by free-trial customers’
stronger marketing responsiveness (see Table 5) and by the
higher levels of direct marketing they received (see Table 2).
Control variables. The control variables have significant
and face-valid effects. Television usage drops in warmer
months for obvious reasons, and the VOD credit increases
usage of the VOD service. In addition, we find positive
carryover effects for both usage components (e.g., Bolton
and Lemon 1999).13 Higher subscription fees reduce a cus-
tomer’s probability to retain and use the service (e.g., higher
fees reduce customers’ budgets and, as a result, decrease
their paid VOD consumption). The concomitant consumer
characteristics also play a significant role. For example, all
else equal, larger families are likely to use the service more
intensively (both the flat-rate service and pay-per-use com-
ponent). Finally, the table in Web Appendix C indicates that
there are significant correlations between the random inter-
cepts of the retention and usage equations.
IMPLICATIONS FOR CLV AND CE
To examine how the estimated effects influence CLV, we
compare the CLV of free-trial and regular customers and
compute the elasticities of CLV with respect to changes in
marketing communication efforts and customers usage
intensities. Throughout the calculations, we use customer-
specific posterior parameter distributions (Train 2009).
For maximum comparability between free-trial and regu-
lar customers, we use the same global means for the market-
ing variables. Furthermore, to avoid comparing the behavior
of (relatively tentative) free-trial customers in their first three
months with (relatively confident) regular customers, we
only analyze those customers who survived the first three
months, retaining 8,624 free-trial customers and 3,145 regu-
lar customers. In addition, in the simulations, we shock mar-
keting or usage in the first month after the trial period (i.e., in
month 4 of a customer’s tenure). Next, we explain how we
derive CLV from customers’ retention and usage behavior.
Calculating Net CLV
We simulate customers’ usage and retention behavior
over a three-year time horizon. Previous research on other
high-tech products and services has used the same simula-
228 JOURNAL OF MARKETING RESEARCH, APRIL 2015
Figure 2
MODEL FIT
B: Usage of Flat-Rate ServiceA: Retention C: Usage of Pay-Per-Use Service
1.05
1.00
.95
.90
.85
.80
.75
Retention Rate
Time Since Adoption (Months)
1 3 5 7 9 11 13
 
5.5
5.0
4.5
4.0
3.5
3.0
Usage of Flat-Rate Service (Avg. Log Zaps)
Time Since Adoption (Months)
1 3 5 7 9 11 13
 
1.2
1.0
.8
.6
.4
.2
0
Usage of Pay-Per-Use Service (Avg. VODs)
Time Since Adoption (Months)
1 3 5 7 9 11 13
 
13The carryover coefficients for the two usage components are not
directly comparable because we use a log-log regression for flat-rate usage
and a zero-inflated Poisson model for the pay-per-use service. Nonetheless,
the relative magnitude of the coefficients suggests that the flat-rate service
is characterized by higher state dependence than the pay-per-use service.
This finding is consistent with the idea that consumers are accustomed to
watching the same television shows or series (flat-rate usage) but irregu-
larly consume unrelated content from the VOD service (pay per use).
Free-trial customers (predicted) Regular customers (predicted)
Free-trial customers (observed) Regular customers (observed)
The Challenge of Retaining Customers Acquired with Free Trials 229
tion horizon, arguing that most of a customer’s value is typi-
cally captured during these first three years (Kumar et al.
2008; Rust, Kumar, and Venkatesan 2011). We compute net
CLV as the total revenue stemming from a customer’s con-
secutive retention and usage decisions minus the costs to
acquire and retain that customer. For a given month, revenue
consists of the fixed subscription fee and the pay-per-use
fees for watching VODs (corrected for content costs) if the
customer retains the iTV service, or the early-termination
fee if the customer disadopts. On the cost side, we distin-
guish direct-marketing and advertising expenditures to
acquire and retain the customer. Other costs are not directly
related to the number of customers or their usage intensity
(e.g., the network infrastructure is owned by the company)
and can thus be ignored in the computation of net CLV. To
compute the present value of future cash flows, we use an
annual discount rate of 8.5%, which is common in this
industry (e.g., Cusick et al. 2014).14 Web Appendix D pro-
vides the equations for the net CLV calculations.
Survival times. As a first step toward calculating net CLV,
we investigate customers’ expected survival times during
the three-year simulation period. Figure 3 shows that free-
trial customers churn much earlier than regular customers.
The survival times are also characterized by considerable
consumer heterogeneity, especially for free-trial customers.
Thus, we can expect substantial differences in net CLV
between and within the two customer groups.
Net CLV for free-trial and regular customers. Figure 4
presents the net CLV results. Panel A presents average reve-
nues, average costs, and the resulting CLV for both groups.
The major part of revenues comes from subscription fees.
The pay-per-use service accounts for 13% (€36.82) of the
total revenues from free-trial customers and 10% (€41.25)
of the total revenues from regular customers. Regular cus-
tomers, who cannot cancel the service for free during a trial
period, pay a higher expected cancellation fee than free-trial
customers (respectively, €16.19 vs. €7.11). The costs for
acquisition (€170.29 for free-trial and €169.89 for regular
customers) and retention (€10.79 for free-trial and €14.01
for regular customers) are comparable. After taking into
account all revenues and costs, the resulting net CLV is, on
average, 59% lower for free-trial customers than for regular
customers (respectively, €101.25 vs. €244.62).
Figure 4, Panel B, shows that there is also considerable
heterogeneity in net CLV within the free-trial and regular
customer groups. Although free-trial customers, on average,
have a lower net CLV, several of them generate a relatively
high value, which becomes clear from the substantial over-
lap between the two distributions.
Marketing communication and usage elasticities. The
parameter estimates suggest that free-trial customers are
significantly more responsive to changes in marketing com-
munication and usage levels than regular customers. To
quantify whether these differences are also managerially
relevant, we calculate elasticities of net CLV with respect to
marketing communication and usage. We find these elas-
ticities to be substantially higher for free-trial customers
than for regular customers. As we show in Figure 5, Panels
A and B, the average advertising and direct-marketing elas-
ticities of free-trial customers are .26 and .41, respectively.
For regular customers, however, we find average elasticities
of only .02 and .08, respectively.15
In a similar vein, we assess the extent to which net CLV
changes in response to an increase in usage (see Figure 5,
Panels C and D). For example, the iTV company could
enhance usage by offering access to more channels (flat-rate
service) or extending their VOD catalog (pay-per-use service).
We find that the average flat-rate usage elasticity equals .75
for free-trial customers but only .17 for regular customers.
Likewise, the average pay-per-use elasticity is .04 for
free-trial customers yet only .02 for regular customers.
Should a Company Offer Free Trials?
A major finding of this research is that the expected value
of customers attracted with free trials is less than the
expected value of regular customers (E(CLVfree trial) =
€101.25 vs. E(CLVregular) = €244.62). Consequently, man-
agers may wonder whether offering free trials is worth the
effort. For a free trial to be beneficial, it should attract
enough lower-value customers to at least keep the aggre-
gated CLV, or CE, constant. But how many customers does
the free trial need to attract?
14The relative results remain the same when we vary the simulation hori-
zon or discount rate.
15We compute customer-specific arc elasticity on the basis of a 1%
increase in the focal variable in the first period after the expiry of the free
trial and apply a 95% Winsorization on estimated net CLV elasticities to
ensure that the summary statistics are not driven by outliers (Luo, Raithel,
and Wiles 2013). Differences in elasticities between free-trial and regular
customers are higher than what perhaps could be expected on the basis of
the estimated coefficients (e.g., .310 for regular customers and .310 + .154
for free-trial customers, in the case of advertising). However, due to differ-
ences in the net CLV base levels between the two customer groups, the
same absolute lift in net CLV represents a much higher percentage lift for a
free-trial customer than for a regular customer.
Figure 3
DISTRIBUTION OF EXPECTED SURVIVAL TIMES FOR
FREE-TRIAL AND REGULAR CUSTOMERS
.08
.06
.04
.02
0
Density
Expected Survival Time (in Months)
5 10 15 20 25 30 35
  
 
8.73)='625.73, =0Regular customers (
9.85)='619.66, =0trial customers (Free
Free-trial customers (M = 19.66, SD = 9.85)
Regular customers (M = 25.73, SD = 8.73)
  
 
  
 
o
Notes: To calculate survival times, we consider a maximum time horizon
of three years (36 months).
To determine the necessary number of additional cus-
tomers, we can use the results of the CLV calculations.
According to the estimates, the 8,624 free-trial customers
who survive the free-trial period generate a total expected
CE of €873,180 (8,624 ¥E(CLVfree trial)). To generate the
same equity without the free trial, the firm would have had
to attract only 3,570 regular customers because 3,570 ¥
E(CLVregular) = 8,624 ¥E(CLVfree tria l) = €873,180. Thus,
offering consumers a free trial rather than a regular contract
should lead to k= 8,624/3,570 = E(CLVregular)/E(CLVfree
trial) = 2.42 times more customers to keep CE constant. If
firm managers believe a free trial is able to attract more new
customers than this threshold, we recommend offering the
free trial rather than the regular contract because the
expected expansion of the customer base will compensate
for the customers’ lower CLVs; otherwise, the company
should not offer the trial.
Importantly, kis not a fixed number but depends, to some
extent, on the company’s marketing efforts. Specifically, the
company can lower the threshold by increasing its market-
ing efforts after acquisition, making it easier for the free
trial to break even in terms of CE. Indeed, the results indi-
cate that higher marketing efforts lead to a smaller differ-
ence between E(CLVfree trial) and E(CLVregular) such that the
threshold k= E(CLVregula r)/E(CLVfree trial) decreases. We
illustrate this principle in Figure 6. For example, for direct-
marketing (vs. advertising) efforts 20% above the observed
average, kdrops from 2.42 to 2.26 (vs. 2.36). Thus, with
increased direct-marketing (advertising) expenditures, it
suffices if the trial is only 2.26 (2.36) times more successful
at attracting customers than the regular offer. These insights
help companies trade off acquisition and retention efforts
(e.g., Reinartz, Thomas, and Kumar 2005). For example, a
company may decide to boost retention expenditures when
a trial’s acquisition targets are not met. Conversely, a well-
designed free-trial promotion may lead to a higher-than-
expected increase in customers, which may justify lower
retention efforts afterward.
Improving Retention and CE After Acquisition
When a company has acquired customers with free trials
and regular subscriptions, a key managerial question is how to
improve consumers’ retention behavior and CE. Because free-
trial customers are worth less than regular customers, some
managers may decide to invest less in free-trial customers.
However, their higher responsiveness to marketing communi-
cation actually suggests that this is not the best strategy. There-
fore, we recommend that managers pay specific attention to
free-trial customers. To illustrate this recommendation, we
simulate the impact on retention and CE of alternative tar-
geting strategies after customer acquisition (see Table 6).
Suppose a manager has an additional monthly direct-
marketing budget that enables her to increase direct-marketing
efforts by .50 per customer for 20% of the customers.
Given the company’s total customer base of 160,000 cus-
tomers at the end of the observation period, this corresponds
to a substantial marketing investment of approximately
€576,000 (160,000 ¥20% ¥€.50 ¥36 months). The ques-
tion then becomes how to select the 20% target customers.
Using the customers in the simulation data set, we consider
different scenarios. In scenario 1, the manager simply
selects customers from the company’s database at random,
yielding an increase in retention (158.0%) and net CE
(115.9%) for the 20% targeted customers. The impact across
the entire customer base is also substantial: 31.6% for reten-
tion, 23.0% for net CE.
Alternatively, bearing in mind our finding that acquisition
mode influences customers’ response to marketing commu-
nication, a manager could follow scenario 2 and target a
random subset of free-trial customers. Such a strategy
230 JOURNAL OF MARKETING RESEARCH, APRIL 2015
Figure 4
NET CLV FOR FREE-TRIAL AND REGULAR CUSTOMERS
A: Average Net CLV B: Distribution of Net CLV
400
300
200
100
0
–100
–200
Free-Trial Customers Regular Customers
.006
.004
.002
0
Density
Net CLV in €
–200 –100 0 100 200 300 400 500
  
 €156.95)='6€244.62, =0Regular customers (
€183.12)='6€101.25, =0trial customers (Free
Free-trial customers (M = €101.25, SD = €183.12)
Regular customers (M = €244.62, SD = €156.95)
  
 
  
 
o
The Challenge of Retaining Customers Acquired with Free Trials 231
would improve retention and net CE of the selected cus-
tomers even more (186.7% for retention and 136.0% for net
CE). These returns stand in sharp contrast to scenario 3, in
which the manager invests only in a random selection of
regular customers, leading to an increase of 95.6% for
retention and 75.6% for net CE. Finally, a manager could
opt for scenario 4, in which individual-level response coef-
ficients are used to select customers with the highest
responsiveness to direct marketing. This optimal strategy
would improve retention and net CLV, respectively, by
181.4% and 132.9% for targeted customers and 36.3% and
26.6% for all customers. Note that targeting free-trial cus-
tomers (scenario 2) leads to results that come close to the
results for the selection based on responsiveness to direct
marketing (scenario 4).
DISCUSSION
Recent research has suggested that the way in which cus-
tomers are acquired may have an enduring impact on their
behavior, even long after adoption (e.g., Villanueva, Yoo,
Figure 5
RESPONSIVENESS TO MARKETING COMMUNICATION
A: Responsiveness to Advertising B: Responsiveness to Direct Marketing
50
40
30
20
10
0
Density
Net CLV Elasticity
0 .05 .10 .15 .20 .25 .30
  
 .0662)='6.0233, =0Regular customers (
.4054)='6.2646, =0trial customers (Free
Free-trial customers (M = .2646, SD = .4054)
Regular customers (M = .0233, SD = .0662)
  
 
  

o
25
20
15
10
5
0
Density
Net CLV Elasticity
0 .1 .2 .3 .4 .5
  
 .2494)='6.0780, =0Regular customers (
.7096)='6.4113, =0trial customers (Free
Free-trial customers (M = .4113, SD = .7096)
Regular customers (M = .0780, SD = .2494)
  
 
  

o
Notes: We compute arc elasticities for a 1% increase in, respectively, advertising, direct marketing, and usage levels in month 4 after acquiring the customer
(i.e., after the trial period).
C: Responsiveness to Shocks in Flat-Rate Usage D: Responsiveness to Shocks in Pay-per-Use Usage
15
10
5
0
Density
Net CLV Elasticity
0 .2 .4 .6 .8
  
 .5811)='6.1697, =0Regular customers (
1.2845)='6.7453, =0trial customers (Free
Free-trial customers (M = .7453, SD = 1.2845)
Regular customers (M = .1697, SD = .5811)
  
 
  
 
o
150
100
50
0
Density
Net CLV Elasticity
0 .2 .4 .6 .8
  
 .0527)='6.0169, =0Regular customers (
.0862)='6.0427, =0trial customers (Free
Free-trial customers (M = .0427, SD = .0862)
Regular customers (M = .0169, SD = .0527)
  
 
  
 
o
Figure 6
REQUIRED IMPACT OF FREE TRIALS ON CUSTOMER BASE
EXPANSION FOR DIFFERENT LEVELS OF MARKETING
2.45
2.40
2.35
2.30
2.25
Required Customer
Base Expansion (k)
Marketing Efforts for Customer Retention
(Percentage Increase from Average)
+0% +5% +10% +15% +20%
Direct marketing
Advertising
Direct marketing
Advertising
Notes: krepresents the threshold value to keep CE constant and can be
interpreted as how much more successful the free trial should be at attract-
ing new customers compared with the regular offer.
and Hanssens 2008). Whereas some studies have found dif-
ferences between customers attracted with sales promotions
and regular customers (Anderson and Simester 2004; Lewis
2006), the challenge of retaining customers acquired
through free-trial promotions is not well understood. Cru-
cial questions have remained unanswered: How do free-trial
customers differ from regular customers in their retention
rates, marketing responsiveness, and CLVs? Should free-
trial customers be managed differently?
To answer these questions, we model a customer’s deci-
sions to use and retain the service—decisions that influence
CLV—and examine how free-trial acquisition affects this
decision process. We explicitly account for selection effects
(i.e., customers attracted with free trials may differ intrinsi-
cally from other customers) and the endogeneity of market-
ing communication. We estimate the model on a unique
panel data set, covering monthly usage (of flat-rate and
pay-per-use services) and retention decisions for 16,512
customers of an interactive digital television service
provider. We then use the parameter estimates to simulate
CLV and quantify its sensitivity to changes in marketing
communication and usage rates.
Throughout the analyses, we find strong evidence for sys-
tematic differences in behavior between free-trial and regu-
lar customers. In line with buyer–seller relationship theory,
which suggests that free trials may decelerate the relation-
ship formation process, free-trial promotions are associated
with higher defection rates, even beyond the free-trial
period. Similarly, flat-rate usage—an important driver of
customers’ retention decisions—turns out to be lower
among free-trial subscribers. Notably, however, free-trial
customers have a higher pay-per-use consumption rate.
Apparently, they readily reallocate the time they gain (due
to their lower flat-rate consumption) and the money they
save (due to the free trial) to the pay-per-use service. These
results add to the growing understanding of the role of
usage in the value generation process (e.g., Bolton and
Lemon 1999; Iyengar et al. 2011).
Managerial Implications
Our study has several key managerial implications.
Because of their higher churn rate, free-trial customers are
worth considerably less than regular customers. Specifi-
cally, we find the net CLV of free-trial customers to be 59%
lower than that of regular customers. Managers and busi-
ness analysts may thus need to temper profit expectations if
the customer base includes a substantial share of free-trial
subscribers. At the same time, we find free-trial customers
to be more “malleablethan regular customers. Because
they have a less developed relationship with the firm, free-
trial customers are more uncertain about the service bene-
fits. As a result, they rely more on marketing communica-
tion and their own usage behavior when deciding whether to
retain the service. We find that, compared with regular cus-
tomers, the net CLV of free-trial customers is much more
responsive to direct marketing, advertising, flat-rate usage,
and pay-per-use usage.
Therefore, companies should target direct marketing and
advertising more to free-trial than to regular customers. In
these marketing communications, firms are advised to remind
free-trial customers about their usage rates, especially when
they are high, to make these customers even more likely to
retain the service. For example, mobile phone providers
could use direct marketing to communicate the number of
minutes the customer has used the flat-rate service in the
previous period, cementing the relevance of the service.
Companies could also enhance actual usage of the service.
For example, in the context of digital television, firms could
enhance usage opportunities by offering recorded shows or
providing apps to watch television on mobile devices.
Other service providers can estimate our retention and
usage models to calculate the expected CLV for every cus-
tomer, whether acquired by a free trial or not. They can then
focus their marketing efforts on those customers with the
highest expected return on marketing investment. If the esti-
mation of the models is not feasible, our results suggest that
managers can rely on usage intensity as an indicator of cus-
tomer value and a criterion for marketing allocation deci-
sions. This enables managers to distinguish (likely) high-
and low-value customers, especially within the free-trial
group, because this group is most responsive to usage levels.
Our analysis also offers insights on the factors that play a
role when firms consider the launch of a free-trial cam-
paign. We show how the company can obtain an indication
of the free trial’s required expansion effect on the customer
base to compensate for free-trial customers’ lower CLV and
thus make the free-trial campaign worthwhile. Specifically,
relative to a situation with only the regular offer, the number
of adopters should increase by a factor that can be calcu-
lated as the expected CLV of regular customers divided by
232 JOURNAL OF MARKETING RESEARCH, APRIL 2015
Table 6
IMPROVING RETENTION RATES AND CE
Impact for Targeted Impact for Entire
Customers (20%) Customer Base (100%)
Impact on Impact on Impact on Impact on
Retentiona Net CEb Retentiona Net CEb
Increase Direct Marketing Efforts by €.50 for 20% of Customers
Scenario 1: No targeting (random) 158.0% 115.9% 31.6% 23.0%
Scenario 2: Target free-trial customers (random) 186.7% 136.0% 37.3% 27.2%
Scenario 3: Target regular customers (random) 95.6% 75.6% 19.1% 15.1%
Scenario 4: Target customers on the basis of highest 181.4% 132.9% 36.3% 26.6%
individual-level response coefficients
aNumber of retained customers in month 36 after the start of the free-trial program.
bNet CE for retained customers in month 36 after the start of the free-trial program.
The Challenge of Retaining Customers Acquired with Free Trials 233
the expected CLV of free-trial customers. If the company
expects the free trial to attract more customers than this fac-
tor prescribes, the trial offer may be a viable strategy.
Importantly, the company can lower this break-even factor
through targeted retention efforts after acquisition.
Further Research
Our work is the first to investigate the effects of free-trial
acquisition on customer behavior and CLV and therefore
leaves several opportunities for further research. First,
because this study is based on observational data, we need
to control for selection effects—a challenge intrinsic to
many studies examining the behavior of distinct customer
groups (e.g., Gensler, Leeflang, and Skiera 2012). Alterna-
tively, researchers may use controlled field experiments.
One setting particularly suited for field experiments is
online services, in which individual consumers can be ran-
domly assigned to different offers, and the likelihood of
consumers becoming aware of alternative offers is limited.
At the same time, such an approach would still not guaran-
tee full comparability of free-trial and regular customers
because the researcher cannot force consumers to accept an
offer.
Second, controlled experiments would also allow for the
uncovering of the underlying psychological processes that
are likely to play a role. One notable observation worthy of
further attention is that free-trial customers are more hetero-
geneous than regular customers—for example, in terms of
CLV and responsiveness to marketing and usage. We specu-
late that the segment of free-trial customers includes not
only people who are truly doubtful about their commitment
but also opportunistic consumers who would have adopted
the service anyway and only subscribe to the trial to enjoy
the free months. Future studies could further investigate the
heterogeneity in the motivations of free-trial and regular
consumers.
Third, further research could replicate this research in
other contexts. Many Internet companies currently operate
under a “freemium” model, in which customers can choose
between a free service and a premium service that comes at a
fee but provides enhanced functionality (Pauwels and Weiss
2008). For example, the on-demand streaming service Spotify
offers temporary one-month free trials for its premium service
in addition to its (permanently) free service.
Fourth, further research could give a more complete
account of the profitability implications of free-trial and
regular acquisition. For example, for a news site, usage may
be a source of advertising revenues (e.g., the number of
page views may determine advertising income). Further-
more, it might be worthwhile to also include other value-
creating behaviors (Verhoef, Reinartz, and Krafft 2010).
Notably, scholars could examine whether free-trial and
regular customers differ in the extent to which they engage
in word-of-mouth communication.
In summary, we uncover the implications of free-trial
acquisition for customer retention and CLV. We show how
service usage and marketing activities drive consumers’
retention decisions and CLVs and how free-trial acquisition
moderates these relationships. Our findings offer managers
new insights on how to retain customers acquired through
free trials.
REFERENCES
Anderson, Eric. T. and Duncan I. Simester (2004), “Long-Run
Effects of Promotion Depth on New Versus Established Cus-
tomers: Three Field Studies,” Marketing Science, 23 (1), 4–20.
Ascarza, Eva, Anja Lambrecht, and Naufel Vilcassim (2012),
“When Talk Is ‘Free’: The Effect of Tariff Structure on Usage
Under Two- and Three-Part Tariffs,” Journal of Marketing
Research, 49 (December), 882–99.
Bawa, Kapil and Robert Shoemaker (2004), “The Effects of Free
Sample Promotions on Incremental Brand Sales,” Marketing
Science, 23 (3), 345–63.
Blattberg, Robert C., Edward C. Malthouse, and Scott A. Neslin
(2009), “Customer Lifetime Value: Empirical Generalizations
and Some Conceptual Questions,Journal of Interactive Mar-
keting, 23 (2), 157–68.
Bolton, Ruth N. and Katherine N. Lemon (1999), “A Dynamic
Model of Customers’ Usage of Services: Usage as an
Antecedent and Consequence of Satisfaction,” Journal of Mar-
keting Research, 36 (April), 171–86.
Brucks, Merrie (1985), “The Effects of Product Class Knowledge
on Information Search Behavior,” Journal of Consumer
Research, 12 (1), 1–16.
Chan, Tat Y., Chunhua Wu, and Ying Xie (2011), “Measuring the
Lifetime Value of Customers Acquired from Google Search
Advertising,” Marketing Science, 30 (5), 837–50.
Cusick, Philip, Richard Choe, Eric Pan, and Ava Zhang (2014),
“DirecTV: Takeways from DTV Management Meetings—US
and Latam [J.P. Morgan Research],” Thomson ONE Banker
Analytics (accessed September 30, 2014), [available at http://
banker. thomsonib.com].
Danaher, Peter J. (2002), “Optimal Pricing of New Subscription
Services: Analysis of a Market Experiment,” Marketing Sci-
ence, 21 (2), 119–38.
Dodson, Joe A., Alice M. Tybout, and Brian Sternthal (1978),
“Impact of Deals and Deal Retraction on Brand Switching,”
Journal of Marketing Research, 15 (February), 72–81.
Dwyer, F. Robert, Paul H. Schurr, and Sejo Oh (1987), “Develop-
ing Buyer–Seller Relationships,” Journal of Marketing, 51
(April), 11–27.
Ebbes, Peter, Dominik Papies, and Harald J. Van Heerde (2011),
“The Sense and Non-Sense of Holdout Sample Validation in the
Presence of Endogeneity,” Marketing Science, 30 (6), 1115–22.
Gedenk, Karen and Scott A. Neslin (1999), “The Role of Retail
Promotion in Determining Future Brand Loyalty: Its Effect on
Purchase Event Feedback,” Journal of Retailing, 75 (4), 433–
59.
Gensler, Sonja, Peter Leeflang, and Bernd Skiera (2012), “Impact
of Online Channel Use on Customer Revenues and Costs to
Serve: Considering Product Portfolios and Self-Selection,” Inter-
national Journal of Research in Marketing, 29 (2), 192–201.
—, —, and— (2013), “Comparing Methods to Sepa-
rate Treatment from Self-Selection Effects in an Online Banking
Setting,” Journal of Business Research, 66 (9), 1272–78.
Gilbert, Daniel T. and Jane E.J. Ebert (2002), “Decisions and Revi-
sions: The Affective Forecasting of Changeable Outcomes,
Journal of Personality and Social Psychology, 82 (4), 503–514.
—, Timothy D. Wilson, Elizabeth C. Pinel, Stephen J. Blum-
berg, and Thalia P. Wheatley (1998), “Immune Neglect: A
Source of Durability Bias in Affective Forecasting,” Journal of
Personality and Social Psychology, 75 (3), 617–38.
Gilbride, Timothy J., Greg M. Allenby, and Jeff D. Brazell (2006),
“Models for Heterogeneous Variable Selection,” Journal of
Marketing Research, 43 (August), 420–30.
Heilman, Carrie M., Kent Nakamoto, and Ambar G. Rao (2002),
“Pleasant Surprises: Consumer Response to Unexpected In-
Store Coupons,” Journal of Marketing Research, 39 (May),
242–52.
Heiman, Amir, Bruce McWilliams, Zhihua Shen, and David Zil-
berman (2001), “Learning and Forgetting: Modeling Optimal
Product Sampling over Time,” Management Science, 47 (4),
532–46.
Hoch, Stephen J. and John Deighton (1989), “Managing What
Consumers Learn from Experience,” Journal of Marketing, 53
(April), 1–20.
Iyengar, Raghuram, Kamel Jedidi, Skander Essegaier, and Peter J.
Danaher (2011), “The Impact of Tariff Structure on Customer
Retention, Usage, and Profitability of Access Services,” Market-
ing Science, 30 (5), 820–36.
Jain, Dipak, Vijay Mahajan, and Eitan Muller (1995), An
Approach for Determining Optimal Product Sampling for the
Diffusion of a New Product,Journal of Product Innovation
Management, 12 (2), 124–35.
Johnson, Michael D. and Fred Selnes (2004), “Customer Portfolio
Management: Toward a Dynamic Theory of Exchange Relation-
ships,” Journal of Marketing, 68 (April), 1–17.
Kumar, V., Rajkumar Venkatesan, Tim Bohling, and Denise Beck-
mann (2008), “Practice Prize Report—The Power of CLV: Man-
aging Customer Lifetime Value at IBM,” Marketing Science, 27
(4), 585–99.
Lemmens, Aurélie and Christophe Croux (2006), “Bagging and
Boosting Classification Trees to Predict Churn,” Journal of
Marketing Research, 43 (May), 276–86.
Lemon, Katherine N., Tiffany Barnett White, and Russell S. Winer
(2002), “Dynamic Customer Relationship Management: Incor-
porating Future Considerations into the Service Retention Deci-
sion,” Journal of Marketing, 66 (January), 1–14.
Lewis, Michael (2006), “Customer Acquisition Promotions and
Customer Asset Value,” Journal of Marketing Research, 43
(May), 195–203.
Luo, Xueming, Sascha Raithel, and Michael A Wiles (2013), “The
Impact of Brand Rating Dispersion on Firm Value,” Journal of
Marketing Research, 50 (June), 399–415.
Mitchell, Andrew A. and Jerry C. Olson (1981), “Are Product
Attribute Beliefs the Only Mediator of Advertising Effects on
Brand Attitude?Journal of Marketing Research, 18 (August),
318–32.
Mundlak, Yair (1978), “On the Pooling of Time Series and Cross
Section Data,” Econometrica, 46 (1), 69–85.
Palmatier, Robert W., Mark B. Houston, Rajiv P. Dant, and Dhruv
Grewal (2013), Relationship Velocity: Toward a Theory of
Relationship Dynamics,” Journal of Marketing, 77 (January),
13–30.
Park, Sungho and Sachin Gupta (2012), “Handling Endogenous
Regressors by Joint Estimation Using Copulas,” Marketing Sci-
ence, 31 (4), 567–86.
Pauwels, Koen and Allen Weiss (2008), “Moving from Free to
Fee: How Online Firms Market to Change Their Business
Model Successfully,” Journal of Marketing, 72 (May), 14–31.
Polo, Yolanda, F. Javier Sese, and Peter C. Verhoef (2011), “The
Effect of Pricing and Advertising on Customer Retention in a
Liberalizing Market,” Journal of Interactive Marketing, 25 (4),
201–214.
Prins, Remco, Peter C. Verhoef, and Philip Hans Franses (2009),
The Impact of Adoption Timing on New Service Usage and
Early Disadoption,” International Journal of Research in Mar-
keting, 26 (4), 304–313.
Reinartz, Werner J., J.S. Thomas, and V. Kumar (2005), “Balanc-
ing Acquisition and Retention Resources to Maximize Customer
Profitability,” Journal of Marketing, 69 (January), 63–79.
Risselada, Hans, Peter C. Verhoef, and Tammo H.A. Bijmolt
(2014), “Dynamic Effects of Social Influence and Direct Mar-
keting on the Adoption of High-Technology Products,” Journal
of Marketing, 78 (March), 52–68.
Rust, Roland T., V. Kumar, and Rajkumar Venkatesan (2011),
“Will the Frog Change into a Prince? Predicting Future Cus-
tomer Profitability,” International Journal of Research in Mar-
keting, 28 (4), 281–94.
and Peter C. Verhoef (2005), “Optimizing the Marketing
Interventions Mix in Intermediate-Term CRM,” Marketing Sci-
ence, 24 (3), 477–89.
Schary, Philip B. (1971), “Consumption and the Problem of
Time,” Journal of Marketing, 35 (April), 50–55.
Schmitt, Philipp, Bernd Skiera, and Christophe Van den Bulte
(2011), “Referral Programs and Customer Value,” Journal of
Marketing, 75 (January), 46–59.
Schweidel, David A., Peter S. Fader, and Eric T. Bradlow (2008),
“Understanding Service Retention Within and Across Cohorts
Using Limited Information,” Journal of Marketing, 72 (Janu-
ary), 82–94.
— and George Knox (2013), “Incorporating Direct Marketing
Activity into Latent Attrition Models,” Marketing Science, 32
(3), 471–87.
Scott, Carol A. (1976), “The Effects of Trial and Incentives on
Repeat Purchase Behavior,” Journal of Marketing Research, 13
(August), 263–69.
Steffes, Erin M., B.P.S. Murthi, and Ram C. Rao (2008), “Acquisi-
tion, Affinity and Rewards: Do They Stay or Do They Go?”
Journal of Financial Services Marketing, 13 (3), 221–33.
—, ——, and —— (2011), “Why Are Some Modes of
Acquisition More Profitable? A Study of the Credit Card Indus-
try,” Journal of Financial Services Marketing, 16 (2), 90–100.
Thomas, Jacquelyn S. (2001), “A Methodology for Linking Cus-
tomer Acquisition to Customer Retention,” Journal of Market-
ing Research, 38 (May), 262–68.
Train, Kenneth (2009), Discrete Choice Methods with Simulation,
2nd ed. Cambridge, UK: Cambridge University Press.
Van Doorn, Jenny, Katherine N. Lemon, Vikas Mittal, Stephan
Nass, Doreen Pick, Peter Pirner, et al. (2010), “Customer
Engagement Behavior: Theoretical Foundations and Research
Directions,” Journal of Service Research, 13 (3), 253–66.
Verhoef, Peter C. and Bas Donkers (2005), “The Effect of Acquisi-
tion Channels on Customer Loyalty and Cross-Buying,” Journal
of Interactive Marketing, 19 (2), 31–43.
—, Werner J. Reinartz, and Manfred Krafft (2010), “Customer
Engagement as a New Perspective in Customer Management,”
Journal of Service Research, 13 (3), 247–52.
Villanueva, Julian, Shijin Yoo, and Dominique M. Hanssens
(2008), “The Impact of Marketing-Induced Versus Word-of-
Mouth Customer Acquisition on Customer Equity Growth,”
Journal of Marketing Research, 45 (February), 48–59.
234 JOURNAL OF MARKETING RESEARCH, APRIL 2015
... More effective are user test programs. Company can offer the target group a version of the product that is limited in capabilities or in validity [5]. ...
Chapter
Online free trial weakens information asymmetry to some extent. By analyzing online free trial data of Taobao and Tmall, the author discusses the influences of product attributes and trial application activities on trial product sales volume. The results show that product evaluation score, product evaluation quantity, and free trial samples have the significantly positive influences on product sales volume, while consumers’ proportion of successful trial products has the significantly negative influences on product sales volume.
Article
Full-text available
Many firms capitalize on their customers' social networks to improve the success rate of their new products. In this article, the authors analyze the dynamic effects of social influence and direct marketing on the adoption of a new high-technology product. Social influence is likely to play a role because the decision to adopt a high-involvement product requires extensive information gathering from various sources. The authors use call detail records to construct ego networks for a large sample of customers of a Dutch mobile telecommunications operator. Using a fractional polynomial hazard approach to model adoption timing and multiple social influence variables, they provide a fine-grained analysis of social influence. They show that the effect of social influence from cumulative adoptions in a customer's network decreases from the product introduction onward, whereas the influence of recent adoptions remains constant. The effect of direct marketing is also positive and decreases from the product introduction onward. This study provides new insights into the adoption of high-technology products by analyzing dynamic effects of social influence and direct marketing simultaneously.
Article
Full-text available
The dynamic components of relational constructs should play an important role in driving performance. To take an initial step toward a theory of relationship dynamics, the authors introduce the construct of commitment velocity— or the rate and direction of change in commitment—and articulate its important role in understanding relationships. In two studies, the authors demonstrate that commitment velocity has a strong impact on performance, beyond the impact of the level of commitment. In Study 1, modeling six years of longitudinal data in a latent growth curve analysis, the authors empirically demonstrate the significance of commitment velocity as a predictor of performance. In Study 2, the authors use matched multiple-source data to investigate the drivers of commitment velocity. Both customer trust and dynamic capabilities for creating value through exchange relationships (i.e., communication capabilities for exploring and investment capabilities for exploiting opportunities) affect commitment velocity. However, trust and communication capabilities become less impactful as a relationship ages, while investment capabilities grow more important. The authors offer three post hoc tenets that represent initial components of a theory of relationship dynamics that integrates two streams of relationship marketing research into a unified perspective.
Article
Market response models based on field-generated data need to address potential endogeneity in the regressors to obtain consistent parameter estimates. Another requirement is that market response models predict well in a holdout sample. With both requirements combined, it may seem reasonable to subject an endogeneity- corrected model to a holdout prediction task, and this is quite common in the academic marketing literature. One may be inclined to expect that the consistent parameter estimates obtained via instrumental variables (IV) estimation predict better than the biased ordinary least squares (OLS) estimates. This paper shows that this expectation is incorrect. That is, if the holdout sample is similar to the estimation sample so that the regressors are endogenous in both samples, holdout sample validation favors regression estimates that are not corrected for endogeneity (i.e., OLS) over estimates that are corrected for endogeneity (i.e., IV estimation). We also discuss ways in which holdout samples may be used sensibly in the presence of endogeneity. A key takeaway is that if consistent parameter estimates are the primary model objective, the model should be validated with an exogenous (rather than endogenous) holdout sample.
Article
In this study, the authors develop an inexpensive method to help firms assess the relative effectiveness of multiple advertising media. Specifically, they use a firm's loyalty program database to capture media exposure, through an online media survey, for all the media in which the firm advertises. In turn, the exposure data are matched with the purchase history for these same respondents, thereby creating single-source data. The authors illustrate their method for a large retailer that undertook a short-term promotional sale by advertising in television, radio, newspaper, magazine, online display ad, sponsored search, social media, catalog, direct mail, and e-mail channels. In this case, seven of the ten media significantly influence purchase outcomes. Finally, the authors demonstrate how to use their advertising response model to determine the optimal budget allocation across each advertising media channel.
Article
Although relatively under-researched, behavioral influence strategies commonly are used by marketers to modify consumer demand. Reported here are the results of a field experiment which tested the effectiveness of trial and incentive behavioral influence strategies. The efficacy of using self-perception theory to explain and predict behavior is assessed.
Article
The authors argue that what consumers learn from the experience of using products is not a simple matter of discovering objective truth. They frame the problem of learning from experience as a fourstage process (hypothesizing-exposure-encoding-integration) with three moderating factors (familiarity with the domain, motivation to learn, and the ambiguity of the information environment). The framework is used to identify where learning from product consumption experience is most open to managerial influence. The authors discuss strategic tools for managing experiential learning and consider applications to the simulation of learning in concept and pre-test-market product testing.
Article
The author discusses the ways in which time enters into consumers' product choices, how the consumer perceives the time involved in these choices, and the significance of adding consideration of time for marketing. Time is a neglected dimension in the study of consumer behavior in this article. The implications of time for marketing strategy are described, and new ways to look at both products and markets are suggested.
Article
A key challenge in financial services marketing is attracting good customers to the firm. For most financial services firms, including credit card firms, a good customer is also a profitable customer. Managers would like to use marketing tactics that attract the most profitable customers while closely monitoring and perhaps limiting expenditures on marketing tactics that tend to attract relatively less profitable customers. Therefore, managers need to understand the relative effectiveness of different modes of new account acquisition and the impact that the various modes of acquisition may have on overall account profitability. To date, there have been very few studies that have calculated individual level customer profitability and then investigated the relationship between new customer acquisition source and customer profitability. That is, how do modes of acquisition differ in their ability to attract profitable customers? We answer this question using a proprietary and novel data set from the credit card industry. Of the four modes of acquisition used in this industry, we find that Internet and direct mail efforts generate more profitable customers than telemarketing and direct selling. We provide possible explanations for these findings. Our work adds to the growing literature in customer relationship management and our results have important managerial implications for resource allocation among acquisition strategies.
Article
Past research in marketing and psychology suggests that pricing structure may influence consumers' perception of value. In the context of two commonly used pricing schemes, pay-per-use and two-part tariff, we evaluate the impact of pricing structure on consumer preferences for access services. To this end, we develop a utility-based model of consumer retention and usage of a new service. A notable feature of the model is its ability to capture the pricing structure effect and measure its impact on consumer retention, usage, and pricing policy. Using data from a pricing field experiment for a new telecommunication service, we find that consumers derive lower utility from consumption under a two-part tariff than pay-per-use pricing, resulting in lower retention of customers and lower usage of the service. Specifically, our demand analysis shows that a two-part tariff structure leads to an average decline of 10.5% in the annual retention rate and an average decrease of 38.7% in yearly usage relative to pay-per-use pricing after controlling for income effects. Despite the higher customer churn and lower usage, we find that the two-part tariff is still the profit-maximizing pricing structure. However, our results show that if firms ignore the pricing structure (or access fee) effect, then they would overcharge customers for the access fee and undercharge them for the per-minute price. Translated in terms of profitability, the failure to account for the access fee effect leads to a reduction of 11% in firm profit.