Vous êtes sur la page 1sur 47

Electronic copy available at: http://ssrn.

com/abstract=1692252
Commercialization of Platform Technologies:
Launch Timing and Versioning Strategy

Hemant K. Bhargava

hemantb@ucdavis.edu
Graduate School of Management
University of California Davis
Byung Cho Kim

kkbbcc@gmail.com
Graduate School of Management of Technology
Sogang University
Daewon Sun

dsun@nd.edu
Department of Management
University of Notre Dame
January 11, 2012

Author names are listed alphabetically and all authors contributed equally.

Mailing Address: Hemant K. Bhargava, GH-3108, University of California Davis, Davis, CA 95616; Telephone:
530-754-5961; Fax: 530-752-2924.

Mailing Address: Byung Cho Kim, Graduate School of Management of Technology, Sogang University, 35
Baekbeom-ro (Shinsu-dong), Mapo-gu, Seoul 121-742, Korea; Telephone: 82-2-705-7986; Fax: 82-2-3274-4808.

Mailing Address: Daewon Sun, Room 359, Mendoza CoB, Notre Dame, IN 46556; Telephone: 574-631-0982;
Fax: 574-631-5127.
Electronic copy available at: http://ssrn.com/abstract=1692252
Commercialization of Platform Technologies:
Launch Timing and Versioning Strategy
Abstract
Many emerging entrepreneurial applications and services connect two or more groups of
users over Internet-based information technologies. Commercial success of such platform tech-
nology products requires adoption of astute business practices related to product line design,
price discrimination, and launch timing. We examine these issues for a platform rm that
serves two markets, labeled as user and developer markets, with the latter emerging after the
user market is proven. While the size of each market positively impacts participation in the
other, our model allows for uncertainty regarding developer participation. We demonstrate
that product versioning is an especially attractive strategy for platform rms, i.e., the tradeo
between market size and margins is tilted in the direction of more versions. However, when ex-
panding the product line carries substantial xed costs (e.g., marketing cost, cost of additional
plant, managing multiple sets of inventory, increased distribution cost) then the uncertainty in
developer participation adversely impacts the rms ability to oer multiple versions. We show
that for established rms with lower uncertainty about developer participation, the choice is
essentially between an expanded or minimal product line. Startups and rms that are entering
a new product category are more likely to benet from a wait and see deferred expansion
strategy. Still, we demonstrate that uncertainty in developer participation can make early
expansion desirable because it expands the installed base and, with the consequent increase
in developer participation levels, increases the long-term incremental gain from product line
expansion.
Keywords: Technology Commercialization, Product Launch Strategy, Platform Technology,
Versioning, Uncertainty
Electronic copy available at: http://ssrn.com/abstract=1692252
1 Introduction and Motivation
Technological innovation is an expensive and uncertain process which often requires high-end research
and development of new components, production processes, and underlying technologies. Often,
entrepreneurs and rms are unable to successfully commercialize their innovation despite having
technologically sophisticated products. Success requires clearing many hurdles and adoption of
astute business strategies (Christensen and Bower, 1996; Daneels, 2004; Moore, 1991). Challenges
include the chicken and egg problem (e.g., a new payment technology will be adopted only if
accepted by sucient number of merchants, but merchant adoption will itself depend on a sucient
installed base of users), uncertainty in product design and compatibility (e.g., shouldor willall
electric car technologies employ the same battery that can be charged at every battery station, or
will the market be fragmented among multiple technology formats?), the challenge of convincing
consumers to pay high (and denitive) up-front costs in return for small (and uncertain) benets
delivered over a long time (e.g., residential solar power), and the growth vs. protability dilemma
(e.g., should a vendor of an e-book technology sacrice margin and prots in return for high market
share, in order to entice publishers towards its technology?). This article examines this nal challenge,
i.e., the growth vs. protability dilemma, for technology goods.
Our research focuses on technology products that operate as platforms in a two-sided market.
These are products that exhibit positive cross-network eects between two distinct networks of
players, i.e., market adoption on one network inuences, and depends on, the desirability of adoption
on the other network (Eisenmann et al., 2006; Eisenmann, 2007). For example, video gaming
consoles serve (i) gamers, by giving them technology for playing complex video games and (ii)
game developers, by giving them a platform for executing such games and reaching potential buyers;
hence a console platform that attracts more game developers becomes more valuable to gamers, and
conversely, game developers are attracted to console platforms that have many gamers. Similarly,
operating system platforms connect computer users with application software developers. More
recently, smartphones have, as small computers, become platforms for connecting phone users with
1
a variety of computational software and service applications. As noted above, the launch of a
platform technology presents the rm a tough challenge of balancing customer growth and short-
term protability. Growth requires very low (or zero, or even negative) prices in order to propel
interest in the future or on the other side of the market, but these low prices lower the rms
short-term prot.
This growth vs. protability dilemma is common for many startup entrepreneurial ventures,
such as information applications (for mobile phones, tablets, and computers) which connect two or
more groups of users over the Internet. Examples include i) FiGuide.com, which provides personal
nancial services by creating a nancial planner and subscriber network, and ii) Asana.com, which
oers project management tools for a manager and employee network. To illustrate the dilemma,
consider a startup rm which aims to deliver and manage home exercise programs (HEP) on mobile
phones. In contrast to the traditional print-based programs, the use of mobile phones can deliver
multimedia content tailored to the patient, and it can also track and transfer information to the
clinician. Adoption involves a bidirectional loop between patients (as users and direct beneciary
of the program) and clinicians (who prescribe the exercise programs, customize and congure the
application for the patient, and track information about compliance and eectiveness). Because
clinicians have very thin margins and are unlikely to pay for the service, the rm intends, at least
initially, to generate revenue primarily from patients. Charging a high price to patients will generate
the revenue that is desperately needed to fund new applications but it will also restrict adoption;
that, in turn, makes it dicult to entice clinicians to participate. This is the essence of the growth
vs. protability dilemma for such startup rms.
The tension between growth and protability has been discussed in the entrepreneurship litera-
ture. Firm growth is a common measure of success (Davidsson et al., 2008), but the wisdom is that
too much growth must come at the expense of protability (Ramezani et al., 2002; Markman and
Gartner, 2002). Davidsson et al. (2009) examine the eectiveness of growth as a measure of busi-
ness success from a Resource-Based View (RBV) and argue that sound growth starts with achieving
2
protability. But generally, growth and protability are considered to be in conict especially for
products with network eects, the common belief being that rms can either have growth or achieve
protability. Many startup ventures respond to the tension between growth and protability by ini-
tially producing only a single version of their product (to avoid production complexity or perhaps to
place their best foot forward to all their customers) and selling it at a relatively low price needed to
accelerate growth. We argue that this may be an unnecessarily extreme approach, and it magnies
the conict.
Our research is founded on the proposition that growth and protability need not necessarily
operate in conict. Existing theory on market segmentation and product dierentiation suggests
versioning (i.e., an expanded product line with multiple, vertically-dierentiated, versions) as a way
out especially when network eects are present (Bhargava and Choudhary, 2004; Jing, 2007). This
strategy is captured in our model by giving the rm a choice to launch two versions of the product,
a basic and a premium one. The high-end version provides high margin, while the low-end, low-
priced version delivers high market share and installed base necessary to generate substantial network
eects. Yet this does not imply that vendors of new platform technology should launch the technology
with an expanded, rather than minimal, product line. Product line expansion is tempered by the
additional complexity and costs, including operations costs (additional plant, managing multiple
sets of inventory, increased complexity in distribution), marketing costs (data collection and price
optimization, segment development and management, and advertising to multiple customer segments
(Dhebar, 1993; Villas-Boas, 2004)), and cannibalization costs due to increased competition within
the product line. This feature is captured in our model as an incremental product line expansion
cost, specically the one-time or xed cost of executing on the expansion strategy. The third feature
of our model is common to platform goods, i.e., that developer participation involves network eects
and depends on having an installed base of end-users. The fourth and distinctive feature of our model
is that developer participation also has a random unpredictable component. While past literature
has taken a deterministic rational expectations framework to describing network eects (see e.g.,
3
Rochet and Tirole, 2006), we argue that platform rms face substantial uncertainty about whether
or not they can secure participation by the developer market.
Our research contributes to the entrepreneurship literature by highlighting the launch strategy
and timing problem for platform goods and other products that exhibit strong network eects.
Intuitively, platform rms may implement a minimal product line in order to avoid higher xed costs
at launch, and wait for substantial developer participation before expanding the product line. In
contrast, we argue that the rm should be inclined to expand the product line early in order to
increase its installed base and induce a higher level of developer participation. The novel feature
in this analysis is the role of the random component in developers participation decisions. Firms
that develop platform products often have little or no direct control over the number of third party
applications, however they can inuence developers through the design of their product launch
strategy. We show that under network eects, early expansion is generally better than deferred
expansion. The exception is that the rm should employ a wait and see (or deferred expansion)
approach when developer participation is extremely uncertain (and expansion costs high). The key
insight, however, is that early expansion can be useful even in the face of developer uncertainty: by
expanding the user market available to application developers, it can drive developer participation
high enough to the point where the gains exceed the expansion costs. In other words, platform
rms are (compared with products that do not exhibit cross-network eects) more likely to benet
from launching multiple versions simultaneously rather than sequentially after observing developer
participation.
We note that in the technology industry, many established rms also experience challenges
typically faced by startups. For instance, when Apple entered the entertainment market with the
introduction of its iPod music player in 2001, it was known as a maker of computers. It had no
footprint in home entertainment products, lacked recognition as a music retailer, and did not enjoy
supply relationships with music providers. These factors caused substantial uncertainty regarding
whether music rmshighly concerned about piracy, and worried that digital music would amplify
4
itwould in fact license their music for digital distribution through iTunes. Other times, established
rms want to retain a startup avor in order to benet from the innovations and breakthroughs that
often emerge out of new thinking. For instance, Google, Intel, eBay, and other technology rms
have internal organizational structures (such as technology incubators) and incentive schemes aimed
at generating technology startups. Hence our research has relevance both to established technology
rms that are creating new products and entering new markets, and to startup or entrepreneurial
ventures.
However, startups in the technology industry should be cognizant about crucial dierences that
can lead to dierent strategies from those suited to established rms (Shan et al., 1994; Joglekar and
Levesque, 2006). Intuitively, the deferred (rather than early) expansion strategy is particularly suited
to startups which are more weakly positioned with respect to developer participation. Established
rms on the other handthose who have a high fraction of early adopters and/or little uncertainty
about developer participationface a clearer choice between expansion (if costs are low) or not (for
high expansion cost). Our model provides a rigorous foundation for understanding how the expansion
decision is inuenced by the interplay between intensity of network benet, adoption characteristics,
and uncertainty in developer participation. We demonstrate that despite such uncertainty, and to
some extent because of it, early expansion can be desirable for startups. This is because versioning
expands the early-stage installed base, and this increase in market adoption reduces the weight of
the uncertain component in the extent of developer participation.
2 Literature Review
We discuss three streams of research that tie into our work: product launch timing, two-sided
markets, and optimal strategies for startup ventures.
5
2.1 Product Launch Timing and Two-Sided Market
Several researchers have studied the optimal timing of product launch. Ramdas (2003) provides
a framework for examining a rms variety management and discusses strategies associated with
variety-creating decisions. Carrillo (2005) examines the impact of industry clock-speed on pacing of
new product development activities. Bag and Roy (2010) study distribution of public goods with
multiple providers and show that total contribution generated in a sequential move game may be
higher than in a simultaneous move game under incomplete information. Aoki and Prusa (1997)
nd that sequential quality choice leads to smaller quality investment and higher prot, but it
lowers consumer and social surplus. Padmanabhan et al. (1997) study a rms new product launch
strategy under consumer uncertainty regarding network externality. They argue that sequential
launch with sequential provision of quality is optimal for a high-externality rm since under-provision
of introductory quality may serve as a signal of high externality. Moorthy and Png (1992) show
that when a rm faces a serious threat from cannibalization, it may be optimal for the rm to
serve high-valuation customers rst and later introduce lower-quality version to cover low-valuation
segment. We also examine sequential product launch, but unlike these prior studies, we do so in the
context of a two-sided market, and specically in the presence of uncertainty regarding developer
participation.
The literature on platforms has recognized that several traditional business strategies such as
pricing must be modied in response to two-sided network eects. Rochet and Tirole (2003) model
platform competition with two-sided markets and study price setting and surplus sharing under
dierent governance structures. Lee and OConnor (2003) examine the consumers consumption
behavior and the corresponding new product launch strategy in the presence of network eects.
Parker and Van Alstyne (2005) provide insights to help understand interesting phenomena in the
Internet economy, such as free products and product coupling across markets. Rochet and Tirole
(2006) provide a thorough review of the growing literature on platform competition in two-sided
markets. Armstrong (2006) examines platform pricing under competition in two-sided markets and
6
identies the determinants of equilibrium prices. One of the key ndings in the monopoly platform
case is that subsidizing one user group is desirable when the groups demand elasticity is high and
the external benet realized by the other group is sucient. Eisenmann et al. (2006) provide a good
example of such a subsidy. They argue that Adobes distribution of Acrobat Reader creates large
externality from 500 million free users, which eventually incentivizes enterprises to pay $299 for the
commercial version. Liu and Chintagunta (2009) discuss pricing issues under network eects from
a marketing perspective. Eisenmann et al. (2011) examine the strategic management of platform
providers and discuss strategies for platform envelopment. While the main focus of most existing
studies is the role of network externalities in two-sided markets, our model incorporates uncertainty
in application development as well as network externalities in two-sided markets.
2.2 Optimal Strategies for Startup Ventures
A notable stream of research in the entrepreneurship literature examines various aspects of the op-
timal entry strategy of a startup venture. One interesting theme is a new ventures retail channel
selection between virtual and bricks-and-mortar networks given the proliferation of the Internet (e.g.,
Reinhardt and Levesque, 2004). Closer to our paper is entrepreneurs choice between early and de-
layed entry. Early studies nd that early entry to the emerging economy generally yields higher
prot than deferred entry (DeCastro and Chrisman, 1995). This result is somewhat consistent with
our model in the sense that early expansion has greater prot potential than deferred expansion.
More recently, Levesque and Shepherd (2004) examine a startup ventures optimal entry strategy in
emerging and developed markets, grounded on a stylized analytical model, and nd that companies
entering emerging markets have lower cost/benet ratio from using a high mimicry entry strategy
than the ones entering mature markets. Optimal timing of opportunity exploitation is another rele-
vant theme that has been extensively studied in the knowledge management literature (e.g., March,
1991; Choi et al., 2008). In general, it is believed that an optimal strategy for a startup company
is to focus on exploration until it accumulates sucient knowledge, and then move to exploitation.
7
Finally, Armstrong and Levesque (2002) extend Levesque (2000) by modeling uncertainty for the
amount of funding obtained for product development. Because the startups nancial ability is
much more limited, their results indicate that startups are much more sensitive to uncertainty than
established rms.
While numerous aspects of a new ventures business strategies have been well studied, optimal
product line expansion for platform technologies which posit two-sided markets with uncertainty has
not been examined yet, to the best of our knowledge. We aim to bridge the gap in the literature
by investigating the optimal product line expansion strategy for a startup venture with uncertainty.
Inspired by real-world technology markets, we characterize the conditions under which the optimal
timing for product line expansion is determined, and compare early and deferred expansion strategies.
Our results show that with versioning, rms can achieve both growth and protability, which will
give guidance to entrepreneurs who want to commercialize platform technologies.
3 Overview of Model
The dominant approach to modeling two-sided markets assumes simultaneous arrival of the two
sides, so that the outcomes are resolved in a simultaneous coordination game. But Hagiu (2006)
argues that this representation may not be appropriate when the order of arrival of two sides is
well dened, for example in two-sided technology platforms such as computers, gaming consoles, or
personal productivity devices. Here, the rst step tends to be device or platform adoption by end-
users (because the device has sucient standalone features to be of value even without the second
side of the market), and the second step is entry by third-party developers who provide additional
applications to extend the utility of the platform device. Given this sequential arrival of consumers
and application developers in the platform technology market, we develop a two-period model of
customer purchase and developer participation.
8
3.1 Customer and Developer Preferences
Potential buyers arrive and exist in both periods. In the rst-period, the platform product is essentially
viewed as a set of core standalone features (features which are valuable by themselves, and do not
depend on external applications) endowed by the platform rm, and without a substantial network
of third-party application developers. For example, the initial iPhone released in June 2007 was an
all-Apple product, endowed with several standalone features such as voice-calling capabilities, in-
built contact book, calendar, mail, and music capabilities. A software development kit (SDK), which
enabled the creation of third-party applications, was released only in March 2008, and the App Store
was launched in July 2008, over a year after launch of the iPhone. Thus, purchase decisions of rst-
period customers were based primarily on the products standalone features. Potential developers
observe product adoption in the rst period, and by the start of the second period the market
obtains signals about developer participation. In the second period, therefore, customers make
purchase decisions based on both the standalone features and third-party applications or product
complements. The iPhone illustrates this point well. Today, like with other platforms, customer
choice between the iPhone and similar products from competing rms (such as HTC, Google,
Motorola) depends substantially on the size of the respective applications (i.e., the App Store in the
case of iPhone) .
Customers have heterogeneous preferences for the platform product. We capture heterogeneity
with a one-dimensional type parameter v, which represents the customers marginal valuation of
product quality. Product quality may be perceived as a collection of features and the level at which
these are delivered. Higher quality may mean the inclusion of a greater number of useful features
(e.g., inclusion of a camera on a phone) or a premium level of a feature (e.g., a 5 MP camera with
zoom vs. a 2 MP camera). Customers also value the platform more if it has a greater number of
application developer participants (Eisenmann et al., 2006; Katz and Shapiro, 1992; Jing, 2007).
Thus, customers utility for the product is a combination of its standalone features and third-party
applications or complements. This feature is captured with the additive utility function employed in
9
the literature (Katz and Shapiro, 1992; Bhargava and Choudhary, 2004; Jing, 2007). Specically,
a type v customers valuation for a q-quality product when the number of complements is Q, is
v q +Q, where represents the per-complement value. For simplicity, we assume that both rst-
period and second-period customer arrivals have the same distribution of v, uniform on the [0, 1]
interval. This assumption is a simplication, but we emphasize that the main results do not change
even if the rst period customers on average have higher valuations (please refer to Appendix A in
the online supplement for the relaxation of this assumption).
Our model of customer behavior and purchase in the two periods is based on theories of tech-
nology adoption and diusion in the marketing and information systems literatures. Moore (1991)
proposed a chasm framework for technology products, in which two dierent segments of customers
are clearly dened. Customers in the early market, who are labeled technology enthusiasts and
visionaries, make an adoption decision in response to the nature and benets of the innovation.
They are more like risk takers. Their perceived value from the platform, and their purchase decision,
is based primarily on the standalone features of the product. Moreover, note that because of the
sequence of product launch, customer arrival and developer participation, the size of the developer
network is negligible at the time these early customers make their adoption decision. Further, while
such customers may anticipate developer participation in later periods, they have a substantially
high discount rate for future benets.
To summarize, the rst period early adopters willingness to pay for the product primarily de-
pends on its standalone features, while the decision-making of second-period followers involves a
combination of standalone features and developer participation. Formally, we write the net util-
ity of early adopters in the rst period and followers in the second period, U
1
(v, q) and U
2
(v, q),
respectivelyas
U
1
(v, q) = (v q) p(q) and U
2
(v, q) = (v q + Q) (q), (1)
where p(q) and (q) are rst and second period prices for a q-quality product. Let q represent the
10
rms quality vector in period 1, and p the price vector, and let D
1
(q; p) be the realized demand
for product version q. Then the total rst-period installed base of the rm in the user market is
D =

q
D
1
(q; p). The rst-period cohort is therefore split into two parts: those, with v larger than
a threshold v who adopt the platform in the rst period, and those (with v < v) who do not.
The products adoption levels at this stage inuence and determine the extent of developer
participation. If developers observe high product popularity, they are more likely to sign on with the
platform. This relationship is normally modeled in the literature with a deterministic participation
(or variety) function of the form Q =
D

where D is the demand for the platform and is the cost


of developing applications (Shy, 2001). We maintain this classical assumption about the positive
dependence from D to Q (and we normalize to 1 since it is not the object of interest in this
paper). The novel feature of our model is the inclusion of uncertainty in the level of developer
participation, beyond the dependence of this variable on the installed base. That is, we argue that
the extent of developer participation cannot fully be predicted by the demand for the platform
product and is inuenced by other, possibly idiosyncratic, factors. Recent examples of uncertainty in
developer participation at time of launch include 3D TV. A lack of 3D content was identied as the
most signicant contributor to the slow growth of 3D TV sales (Nuttall, 2010). Despite potential
consumers belief that 3D TV will become an industry standard in the near future, 3D TV is not
much attractive to them because it does not yet have a supporting eco-system (Mitra, 2010). Thus,
uncertainty in the future application development, i.e., 3D content, still remain a serious concern for
potential buyers. Therefore, this example demonstrates that uncertainty in application development
is critical for new platform technologies.
Formally, we conceive Q as D + , where the rst component is proportional to the in-
stalled base of users and the second component is a random oset. We normalize the value of this
random variable in the rst period (where, developers observe zero installed base) to zero, hence
by convention, Q represents the developer base in the second period. For simplicity, we consider a
distribution with just two atoms, corresponding to Favorable or Unfavorable developer participation.
11
Our formalization of this uncertainty component is consistent with other recent papers (Cachon
and Lariviere, 2001; Chen, 2005; Dogan et al., 2010; Ha and Tong, 2008). To make the notation
concise, we write the two cases as B (Best case, with probability ) and W (or Worst case).
We normalize the random component in the worst-case to zero, to get
Q =
_

_
D+ if application development is High (probability )
D+ 0 if application development is Low (probability 1 )
(2)
where D is the observed installed base for the product at the start of the second period.
Additional customers enter in the second period, i.e., after developer participation is realized.
These are more risk-averse decision makers, afraid of being locked in a not-yet-standard technology,
but willing to make a decision after uncertainty about developer participation is resolved. These late
adopters or laggards are inuenced by the previous number of adopters which is a widely adopted
assumption in the literature on diusion modeling (see, e.g., Mahajan et al., 1990; Mahajan and
Muller, 1998). This assumption implies, when applied to a two-sided platform product, that second-
period customers make decisions based on the observed level of developer participation (which in
turn depends on the adoption level in the rst period). Van den Bulte and Joshi (2007), who
provide a detailed review of various theories motivating a two-segment structure, also deploy a two-
segment model containing inuentials who are more in touch with innovations than the others and
imitators whose adoption decisions are often inuenced by others decisions. The two-segment
structure was also empirically veried by many researchers (see, e.g., Moe and Fader, 2002; Joshi
et al., 2009).
We normalize a few parameters to simplify the exposition and analysis. First, like Moorthy and
Png (1992), we focus on product expansion as a business strategy rather than driven by technological
improvement, in which case the rm may introduce higher quality products over time. Thus, we
assume that the rms technological capabilities remain constant, and that cost and other parameters
induce it to oer the highest quality version in the rst period itself. Since the quality-level of the
12
high-end product is constrained exogenously, we can set this to 1. Then we denote valuations (for
just the product features) for the low quality product as v, where (0, 1) represents a quality
degradation parameter. This formulation is employed frequently in the versioning literature (see, e.g.,
Deneckere and McAfee, 1996); it implies that each user has constant marginal valuations (CMV)
for product quality, and that the type parameter v represents this marginal valuation. Second,
let denote the mass of rst-period customers, and let us normalize the mass of customers who
exogenously arrive in the second period to 1.
3.2 Structure of Game and Solution Framework
The sequence of events unfolds as follows. In the rst stage the rm chooses its initial product line
strategy. With respect to our research objectives, we limit the product qualities that the rm can
pick to two levels, L and H, where the high quality H is exogenously given and constrained by the
technology innovation level of the rm. Hence the key question for the rm is whether and when
to include an additional, lower quality, version in the product line. We assume full compatibility
between the two versions. That is, any application that works for one version works for the other.
For simplicity, we also assume that the dierence of production costs for the high and low quality
products is negligible, which is applicable to many information goods and other inferior/damaged
goods. In 4.3, we demonstrate that our main ndings still hold even with dierent marginal costs.
Hence the rms strategy space has two points, {H} and {L, H}, because launching only {L} in
the rst period is strictly dominated by launching {H} when there is no dierence in production cost
(this strategy does become feasible under positive costs, which we discuss in 4.3). At this point,
as indicated in Eq. 2 the rm is uncertain about the full extent of developer participation, though it
is aware that participation levels will depend positively on its installed base. In this stage, the rms
target market (on the user side) primarily consists of early adopters who value the product for its
technological features and care little about third-party applications. At the end of the rst period,
as adoption levels materialize, developers begin participating in the eco-system, and the extent of
13
developer participation
low/worst case (W) high/best case (B)
oer H only
W
2
(D; {H})
B
2
(D; {H})
oer {L, H}
W
2
(D; {L, H})
B
2
(D; {L, H})
Table 1: Notation for optimal operating second-period prot (not considering ).
participation level Q is observed by the rm and second-period customers. In the second-period,
the rm can recongure its product line and it sets second-period prices, targeting the product to
second-period customers or followers who make a purchase decision after observing developer
participation.
Formally, we write the strategy space for the rms rst-period decision problem as the vector
[
1
, p
L
, p
H
] where
1
is either {H} or {L, H} (as noted above, we add {H, L} in 4.3). If
1
= {H}
then p
L
is vacuous, and p
H
must satisfy the constraint 0 p
H
< 1. If
1
= {L, H} then the rm
incurs a xed product line expansion cost , and its prices must satisfy 0
p
L

<
p
H
p
L
1
< 1. In the
rst case, rst-period installed base is D = (1 p
H
), while in the second (where the rm expands
its product line) it is D =
_
1
p
L

_
. The rst-period prots for the two cases are

1
({H}, p
H
) = p
H
(1 p
H
), (3)

1
({L, H}, p
L
, p
H
) =
_
p
H
_
1
p
H
p
L
1
_
+ p
L
_
p
H
p
L
1

p
L

__
. (4)
Let
2
denote the second period product line. Given the product line it is oering, the rm
picks prices to maximize the current period prots. Let us rst consider the rms optimal operating
prot for oering a particular product line ({H} or {L, H}), i.e., prot without considering any
applicable product line expansion cost . This term is the prot after optimizing the product prices
corresponding to the level of developer participation (high or low), and is represented using the
notation given in Table 1.
14
To solve the second-period problem, the only relevant inputs from the rst period are
1
and
D. First consider the case where the rm had already chosen an expanded product line in the rst
period (i.e.,
1
= {L, H}). Should it continue oering both products in the second period?
Proposition 1 (No Product Expansion Cost) When product line expansion is costless, the rms
optimal second-period strategy, for all Q > 0, is to sell both versions. The incremental prot from
versioning (given a developer participation level Q) is
versioning
2
=

2
Q
2
(1)
4
and this incremental
prot increases in Q.
This result implies that, when
1
= {L, H}, selling both products in the second period as well
is optimal. Recall that our base-case utility function U(v, q) = v q (i.e., utility for core product
standalone features) was chosen such that versioning is not optimal in the base case (see, e.g.,
Bhargava and Choudhary, 2001; Deneckere and McAfee, 1996). It is the inclusion of utility from
third-party complements, which kick-in in the second period, that makes versioning an attractive
strategy in the second period (in the absence of additional costs for product line expansion), matching
prior results under network eects (Bhargava and Choudhary, 2004; Jing, 2007).
If, however, the rm chose a minimal rst-period product line
1
= {H}, then the additional
expansion cost destroys the inevitability of versioning in the second period. Specically, because

versioning
2
increases in Q, versioning will be attractive only when developer participation Q is
suciently high or, equivalently, expansion cost is low enough. Recall that Q is a random variable
with best-case and worst-case realizations B and W respectively, both of which are functions of
D. Let
B/W
2
(D; {L, H}) be the optimal second-period prot if the rm expands the product line
(adds L) and
B/W
2
(D; {H}) if it continues with an H-only strategy. For each of the two cases B
and W (which are resolved prior to the rms second-period action), the rms second period prot
is therefore the maximum of these two prot terms.
At the beginning of the game, therefore, the rm picks its product line and prices to maximize
its expected prot over the two periods. To complete our notation, dene E[
2
(
1
, D)] to be
15
the rms rst-period expectation of its second-period prot if it enters the second period with an
existing product line
1
and installed base D. Using the notation from Table 1, we have
E[
2
({L, H}, D)] =
B
2
(D; {L, H}) + (1 )
W
2
(D; {L, H}), (5)
E[
2
({H}, D)] = max{
B
2
(D; {L, H}) ,
B
2
(D; {H})}
+(1 ) max{
W
2
(D; {L, H}) ,
W
2
(D; {H})}. (6)
Combining the second-period actions under the B and W realizations (under
1
= {H}) the
rm will either (i) not launch L regardless of the level of Q (i.e., even if Q is high, case B), (ii)
launch L only if Q is high (note that launch L only if Q is low would trivially be inferior to launch
L even if Q is low), or (iii) launch L even if Q is low (i.e., in both B and W cases). Of these we
eliminate case (iii) because of the following result.
Proposition 2 (Benets of Early Expansion) If product expansion is foreseen as inevitable in
period 2 (i.e., launch L even if Q is low), then expanding in period 1 itself is optimal.
Eq. 6 can therefore be replaced with
E[
2
({H}, D)] = (1 )
W
2
(D; {H}) + max{
B
2
(D; {H}),
B
2
(D; {L, H}) }. (7)
Our main objective is to compare the protability of
1
= {H} with
1
= {L, H}. The total
expected prot, at the start of rst period, from a decision to set
1
, p
L
, p
H
is (
1
, p
L
, p
H
) =

1
(
1
, p
L
, p
H
)+E[
2
(
1
, D)] where D = D(p
L
, p
H
) =
_
1
p
L

_
or (1p
H
) as appropriate. For
the two possible rst-period product line decisions, we have the optimal prot under each strategy
as
Early expansion of product line (
1
= {L, H}), then

early
= max
p
L
,p
H
({L, H}, p
L
, p
H
) = max
p
L
,p
H
_

1
({L, H}, p
L
, p
H
) + E[
2
({L, H}, D)]
_
. (8)
16
Defer decision to expand product line (
1
= {H}), then

defer
= max
p
H
({H}, p
H
) = max
p
H
_

1
({H}, p
H
) + E[
2
({H}, D)]
_
. (9)
3.3 Optimal Product Line and Prices
This section species the optimal solutions employing the solution framework described in 3.2.
While we provide a complete analysis of this two-period problem using the solution framework
specied above in Appendix B in the online supplement, these solution details are needed in order to
analyze the impact of network eects and uncertainty on the rms product line expansion strategy.
Solving separately the three optimization problems (one for Eq. 8 and two implied by the max
term inside Eq. 9), we obtain
Lemma 1 (Optimal Prices and Prot of Early Expansion) Under a rst-period product line
{L, H}, optimal rst-period prices are
p
early
L
=
(2
2

2
( + ))
4
2

2

2
,
p
early
H
=

2
(4 2)
2

2

2
( + 2)
8
2
2
2

2
,
and the optimal total prot under early expansion is

early
=
(4(1 4 + + + 2) +
2
(
2
(1 + ) + 4 + 4
2
))
16
2
4
2

2
((1 4 + ) + (1 )
2

2
)
16
3
4
2

2
. (10)
Lemma 2 (Optimal Prices and Prots of Deferred Expansion) For the two deferred expan-
sion strategies:
1. Under a defer, H-only strategy, optimal rst-period price and total expected prots are
p
defer, H-only
H
=
2 (1 + + )
4
2

2
, (11)

defer, H-only
=
4 + 4(2 + ) + (4 + (4 + (4 (1 )
2
)))
4 (4
2

2
)
.
(12)
17
2. Under a defer, expand strategy, optimal rst-period price and total expected prots are
p
defer, expand
H
=
(2 (1 + (1 )))
2
( + )
4
2
((1 ) + )
2
, (13)

defer, expand
=

2
(4 + 4
2

2
(1 4 + (1 )( 2)))
16 4
2
((1 ) + )
2
+
(4(1 + + ) + (
2

2
+ 2 (4
2
(1 )
2
) ( 2)))
16 4
2
((1 ) + )
2
.
(14)
The optimal launch strategy can now be determined by comparing the total prots under the three
strategies specied in Eqs. 10, 12, and 14. We do this next.
4 Results
Lemma 1 and 2 provided the optimal prices and prots conditional on each of the three product line
strategies, (i) expand early ({L, H} in period 1 itself), (ii) expand late (H in period 1, add L in period
2), and (iii) sell H only. Intuitively, as explained earlier, the second-period network eects make
versioning an attractive strategy, but high product line expansion costs can force the rm to follow
an H only strategy. Our goal in this section is to examine and elaborate on these ideas with rigor and
precision. Moreover, higher network benets (e.g., through greater value per-developer or through
higher levels of developer participation) make versioning more attractive in the second period; but
at the same time, early expansion increases the levels of developer participation. Additionally, we
seek to inquire about the impact of uncertainty in developer participation on the expansion strategy.
On the one hand, the rm might wish to delay the expansion decision until the level of developer
participationand, consequently, the level of consumer surplus available to extractbecomes better
known. Alternately, the rm may want to expand the product lineand installed baseearly, in
order to drive higher levels of developer participation and increase the available surplus in the second
period. This section investigates these two forces.
18
4.1 Optimal Product Expansion Strategy
Proposition 3 (Optimal Product Expansion Strategy) Pairwise comparison of the three ex-
pansion strategy yields cuto points (
HD
,
DE
, and
HE
specied in Eqs. 25, 26, & 27) such
that

defer, expand
>
defer, H-only
<
HD
, (15)

early
>
defer, expand
<
DE
, (16)

early
>
defer, H-only
<
HE
. (17)
Combining these results yields the optimal expansion strategy as
1. If < min{
HD
,
DE
} (low expansion cost), then early expansion is optimal.
2. When expansion cost is moderate, i.e., min{
HD
,
DE
} < < max{
HD
,
DE
},
(a) If
HD
< <
DE
,
i. if <
HE
, early expansion is optimal,
ii. otherwise (i.e., >
HE
), defer, H-only is optimal.
(b) If
DE
< <
HD
, the rms optimal strategy is to defer the expansion decision to the
second period, and then expand only if developer participation is high (case B).
3. If > max{
HD
,
DE
}, the optimal strategy is to sell product H only.
When xed cost of launching L is very small, then the rm knows that having L in the second-
period product line improves second-period prot. For a traditional good, the rm would have no
reason to launch L early (in the rst period) as discussed in the Benchmark case in Appendix C in the
online supplement . However, the presence of network eects induces the rm to launch L early in the
rst period itself and benet from higher installed base which causes greater developer participation
and creates greater value for consumers in the second period. For instance, after launching the
relatively expensive iPhone at the end of June 2007, Apple faced a relatively low expansion cost
of adding the iPod Touch, which is just the iPhone minus the calling feature. Importantly, such a
device had the promise of increasing the overall installed base of devices that could run iPhone apps,
which made the platform very attractive to potential application developers. Indeed after launching
iPhone at the end of June 2007 (and with no prior footprint in the world of mobile phones), Apple
quickly added an iPod Touch to the product line in September 2007. The iPod Touch was priced
19
much lower than the iPhone and also did not carry a recurring monthly cellular service fee. Industry
estimates (around January 2010) were that the installed base of the iPhone OS platform was nearly
doubled by addition of the iPod Touch, referred to as a stealth device for the platform.
1
As becomes higher, however, the rms expansion strategy becomes more conservative: it
defers expansion in the rst period, observes developer participation and then incurs expansion costs
only if Q is high enough to guarantee high gains from versioning. This leads to a sequential or
delayed expansion of the product line, if favorable market circumstances emerge. The evolution
of Apples iPod music player is a good example, because of the higher expansion costs associated
with a Windows version of the product and with multiple form factors for the product. Apple
launched iPod in October 2001 with a minimal product line, a Mac-only iPod in a single design
(with 5GB and 10GB disks). Only after observing iTunes roaring success did Apple branch into
an expanded product line with a Windows version of the iPod and additional form factors such as
the iPod Mini and iPod Shue. These moves, which involved substantial xed costs of product line
expansion, enormously increased the iPod installed base but were deferred until Apple had observed
high developer participation and the consequent assurances of a successful product category.
Pushing further into the impact of uncertainty, we examine how the rms strategy shifts as the
degree of uncertainty in developer participation changes. To do this, we frame the cuto points
for the optimal strategies as a combination of and ; this is because a change in alone (which
measures dierence in Q between the W and B scenarios) aects both reservation prices and the
extent of uncertainty in participation. Figure 1(a) illustrates the impact of uncertainty. At = 0,
the rms optimal strategy is either to expand early if is very low or not to expand at all if is
higher; the wait and see approach of deferring expansion has no value due to lack of uncertainty,
and early expansion is always superior to deferred expansion. This same policy remains in forces as
increases beyond 0 (because the uncertain component of Q remains small relative to the overall
value), except that the early expansion is optimal for a higher range of expansion costs due to
1
See http://gigaom.com/apple/ipod-touch-now-outselling-iphone/.
20
Early
Expansion
Early
Expansion
No
Expansion
No
Expansion
Defer,
Expand
Expansion Cost h
R
a
n
d
o
m
O
f
f
s
e
t




Early
Expansion
Early
Expansion
No
Expansion
No
Expansion
Defer,
Expand
Expansion Cost h
P
r
o
b
a
b
i
l
i
t
y


(a)
Early
Expansion
Early
Expansion
No
Expansion
No
Expansion
Defer,
Expand
Expansion Cost h
R
a
n
d
o
m
O
f
f
s
e
t




Early
Expansion
Early
Expansion
No
Expansion
No
Expansion
Defer,
Expand
Expansion Cost h
P
r
o
b
a
b
i
l
i
t
y


(b)
Figure 1: Impact of Uncertainty about Developer Participation on Expansion Strategy
increase in second-period reservation prices. But, as depicted in Figure 1(a), as increases even
further, the rms optimal policy shifts and introduces a new element: defer and expand (only if
developer participation is high) for moderate expansion costs. The reason is that for such high ,
the uncertain component of second-period reservation prices is substantial (relative to the mean),
hence the uncertainty eect becomes dominant in the expansion policy and leads to the use of a
wait and see approach to product line expansion. The formal result is stated below.
Proposition 4 (Cuto Point for Defer, Expand Strategy) There exists a unique

> 0 such
that for low level of uncertainty in developer participationi.e., <

the optimal strategy
is either to expand the product early (if expansion cost is very low, <
DE
()) or not at all.
For higher levels of uncertainty in developer participation, however, there is a moderate region of
expansion cost such that the optimal expansion strategy is a wait and see approach (expand in
the second period only if high developer participation is observed); hence for >

, early expansion
is optimal if <
DE
(), H-only is optimal if >
HD
(), and deferred expansion is optimal when
[
DE
(),
HD
()].
21
What is notable about this analysis is that there exists a range of expansion costs for which
early expansion is optimal even though, had the rm followed an H-only strategy in the rst period,
expansion would not have been optimal in the second-period despite the removal of uncertainty
in developer participation. This result is surprising because intuition suggests that uncertainty
in developer participation (combined with relatively high expansion costs) makes early expansion
unattractive; if the rm expands at all, it should do so in the second period and only if developer
participation is high. The reason for the result is the intricate feedback loop between the rms
product line, installed base and developer participation. Note that the incremental second period
prot gain from versioning is inuenced by the installed base D. Under an H-only product line
compared with early expansionD is relatively small, leading to smaller incremental gain and hence
versioning is attractive only for relatively low . If, however, the rm suboptimally expanded the
product line early and has higher installed base in period 1, then the second-period incremental gain
from versioning is higher than before, justifying the decision to incur the expansion cost in period
1. Implementation of this strategy can raise a startups short-term cash needs. But the costs of
nancing these cash requirements can be more than oset by the spillover eect of increased D
on second-period prots that can generate enough gains to pay back the loan (and interest) in the
second period, even though this same action is unprotable in the full-information setting of the
second period.
An alternative way to examine the impact of uncertainty is to alter and at the same time,
because changing alone implies higher developer market size on average. We conducted a numerical
study, maintaining the average market size xed but changing the uncertainty parameter . Note
that if the average market size is kept xed (by adjusting ), a lower corresponds to a higher
standard deviation of Q. Therefore, Figure 1(b) shows the pure eect of uncertainty about developer
participation, unconpounded by the eect of market size. As depicted in Figure 1(b), there exists
a threshold

such that (i) when >

, the optimal product launch strategy is either early or no
expansion, and (ii) when <

, the deferred expansion strategy becomes attractive. To understand
22
(i), consider an extreme case where is very high (close to 1) and recall Proposition 2. Since high
developer participation is very likely, the rm might as well expand early if expansion is optimal at all
(i.e., product line expansion cost is low enough). For (ii), consider a low . Now, is relatively large
because we maintained the average market size. Hence, the rm is better o with early expansion
when product line expansion cost is low, and no expansion when it is very high. For moderate
expansion cost, the wait and see (deferred strategy) becomes very attractive because the payo
is relatively large.
4.2 Managerial Implications for Startups vs. Established Firms
While many of our examples have focused on established rms entering new product categories
(e.g., Apple iPhone), our ndings also have important implications for a startup that needs to
commercialize a platform product. Now we extend our discussion to technology startups and explain
how they can interpret our ndings. As discussed, many technology-oriented rms frequently face
uncertainty in application development in two-sided markets. For instance, OpenTable has created a
market for connecting restaurants and diners; it provides technology to enable restaurant discovery,
reservations, and other applications. A big challenge for OpenTable was to obtain sucient numbers
of restaurants (correspondingly, diners) into the network, in order to convince diners (correspondingly,
restaurants) to use the system.
In order to dierentiate and compare startups and established rms, we analyze the eect of two
parameters in our model: , the ability to attract developers and , the size of early adopters. First, in
our model, the likelihood of participation is captured via the parameter in the participation function
Q = D+ . It is widely accepted in the literature on software development that the reputation
of a software founder is critical to recruiting developers (West and OMahony, 2005). Therefore, we
assume that is lower for startups relative to the value for established rms: an OpenTable faces
greater challenges in obtaining participants than an Amazon might face in convincing publishers to
provide e-books for the Kindle (or, e.g., Apple to convince developers to write applications for the
23
iPad).

No
Expansion
Defer,
Expand
Early
Expansion
Ability to Attract Developers g
E
x
p
a
n
s
i
o
n
C
o
s
t




No
Expansion
Early
Expansion
Defer,
Expand
Size of Early Adopters k
E
x
p
a
n
s
i
o
n
C
o
s
t



T
o
t
a
l
E
x
p
e
c
t
e
d
P
r
o
f
i
t
Sizeof EarlyAdopters
Early
Defer
No

(a)

No
Expansion
Defer,
Expand
Early
Expansion
Ability to Attract Developers g
E
x
p
a
n
s
i
o
n
C
o
s
t




No
Expansion
Early
Expansion
Defer,
Expand
Size of Early Adopters k
E
x
p
a
n
s
i
o
n
C
o
s
t



T
o
t
a
l

E
x
p
e
c
t
e
d

P
r
o
f
i
t
Size of Early Adopters
Early
Defer
No

(b)

No
Expansion
Defer,
Expand
Early
Expansion
Ability to Attract Developers g
E
x
p
a
n
s
i
o
n
C
o
s
t




No
Expansion
Early
Expansion
Defer,
Expand
Size of Early Adopters k
E
x
p
a
n
s
i
o
n
C
o
s
t



T
o
t
a
l
E
x
p
e
c
t
e
d
P
r
o
f
i
t
SizeofEarlyAdopters
Early
Defer
No

(c)
Figure 2: Optimal Expansion Strategies for Startups vs. Established Firms
Figure 2(a) illustrates the variation in product line expansion strategies for startups vs. established
rms. In all three gures, startups are to the left side of the x axis, i.e., lower ability to attract
developers and smaller size of early adopters, while established rms are to the right side of the x
axis, i.e., greater ability to attract developers and larger size of early adopters. When expansion costs
are low, then even startups should consider early expansion as a way to increase the installed base and
position itself better for the second period. OpenTable addressed this problem by having multiple
levels of nonlinear pricing structures for small vs. large restaurants, both the initial one-time costs and
the continuing fees for providing customers to the restaurant. It also oers restaurants a choice (and
dierent price levels) between using their own reservation technology and paying only for customer
reservations vs. paying for reservation software and hardware as well. This example also indicates a
useful strategy that many rms can follow: create product variety and segment customers through
dierentiation in pricing (which is relatively less expensive to administer) rather than designing
multiple physical products. For the HEP example discussed in the Introduction section, our results
suggest that the rm can oer i) a low-end patient HEP with a minimal fee (or even free) that
might have fewer features and ii) a high-end with a higher fee that could communicate more data,
and more real-time communication, to clinicians. But the result also suggests that when product
line expansion costs are higher, early expansion is less attractive to startups than to established
24
rms. As an additional consequence, this nding also suggests that a startup should carefully build
its business strategy to attract more developersi.e., raise its by providing development tools
or incentives.
The second dierentiating parameter for startups and established rms is . Specically, startups
and established rms may also dier in their ability to attract early adopters who purchase the product
based solely on core standalone features and do not require substantial developer participation before
their purchase decision. Therefore, we assume that startups are likely to have lower relative to
established rms, consistent with the existing theory in the marketing literature that a rm with a
better reputation has a higher chance to get early adoptions of its product (Herbig and Milewicz,
1995). Figure 2(b) illustrates the eect of on the expansion strategy. First, note that as
increases, total prot increases in all of the strategies because of the increase in total market size.
However, compare deferred expansion against no expansion. In both cases, the rst period action
is the same (H-only), but deferred expansion can exploit higher more because higher leads to
higher D and higher Q, hence higher reservation prices in the second period. Thus, the deferred
expansion prot grows faster than no expansion prot as increases. For the early expansion
protwhich partially sacrices short-term (rst-period) prot in order to have better position in
the second periodhigher increases the rst-period sacrice, but it also delivers higher D and Q,
leading to greater gain in second period. Therefore, the desirability of early expansion increases with
. This point is reinforced in Figure 2(c) which demonstrates that an established rm is more likely
to follow early expansion; and, if expansion costs are too high, it might just choose not to expand at
all (this is because with higher , Q becomes more certain, reducing the benet from a wait and
see approach). In contrast, a startup is more likely to nd deferred expansion attractive because
the uncertain component of Q carries greater weight when is small.
25
4.3 Role of Marginal Costs
While the assumption of same (or zero) marginal cost is realistic for information goods and widely
accepted in the literature, relaxing this assumption may add reality since some hardware platforms
may have dierent marginal costs, e.g., Sonys inclusion of a Blu-Ray player in the PS3 gaming
console substantially raised the per-unit costs of the console. In this section, we investigate whether
our main analysis and ndings can be applicable to dierent marginal costs by relaxing same marginal
cost assumption. We start with checking the robustness of our main ndings, i.e., comparison
between early and deferred expansion strategies. In order to better reect reality that marginal
cost increases with product quality, we assume dierent levels of marginal costs for H and L, then
normalize Ls marginal cost to be zero. We denote Hs positive marginal cost with c. The result is
summarized in the following proposition.
Proposition 5 (Optimal Strategy with Dierent Marginal Costs) There exists a unique cut-
o point such that
1. When < , early expansion outperforms defer, expand strategy,
2. When > , there exists a unique cuto point

such that defer, expand strategy is preferred
when the level of favorable developer participation is low ( <

) and early expansion is optimal
when it is high ( >

).
This result demonstrates that deferred expansion could be a viable strategy even with dierent
marginal costs. Next we examine the impact of dierent marginal costs on optimal product launch
strategy. Specically, we compare two dierent defer, expand strategies, i.e., launch H rst then
expand later (H, Then L) vs. introduce L rst, then expand later (L, Then H). With same
marginal cost, trivially we see that H, Then L dominates L, Then H. But, more generally, the
optimal expansion strategy depends on the dierence in marginal cost structures between the low
and high quality versions. Let
H, Then L
(
L, Then H
) be the prot function of H, Then L
(L, Then H).
26
Proposition 6 (Optimal Sequential Launch with Dierent Marginal Costs) Let g() be the
prot dierence (
L,Then H

H,Then L
). If g()|
c=1
> 0, there exists a unique cuto point c
where H, Then L is preferred if c < c. Otherwise (c > c), L, Then H is preferred. If g()|
c=1
< 0,
H, Then L is always preferred.
This result indicates that marginal costs primarily inuence the sequence of product launch and
versioning, rather than the timing of the expansion decision. Specically, when the incremental cost
of the H version is quite high, then the rm may employ a sequential expansion strategy when it
launches the lower-quality version L rst, and then launches the more expensive product if positive
market conditions emerge. This is in contrast to the equal-cost case where launching H rst is
always optimal. The reason why the rm might want to launch L rst is that it wants to sustain
enough market share in the rst period in order to create enough incentives for high developer
participation. Launching the higher-cost version, H, rst, would force the rm to either sacrice
margin in the rst period or obtain a much lower market share if it sets a high price.
5 Concluding Remarks
Many innovative platform products have been launched in the last two decades, including Xbox,
PlayStation, Palm Pilot, Microsoft Platform products, iPhone, iPod, and iPad. Firms have deployed
dierent ways of introducing new platform products. This observation inspired us to investigate
optimal product launch and pricing strategies for a rm that wants to launch a new platform product
when it is uncertain about third-party application development. While prior studies mostly considered
uncertainty about user adoption, our paper is novel in its consideration of uncertainty in application
development. This factor is relevant because the level of developer participation plays a critical role
in consumers purchase decisions. The consideration of developer participation uncertainty leads
to the novel nding that deferred expansion can often be the optimal product launch strategy.
We demonstrated that a technology-oriented startup needs to pay special attention to the level
27
of expansion cost, the number of early adopters (or technology enthusiasts), and the likelihood of
application developers participation. For a startup with a technology product that wants to expand
rapidly in a two-sided market, our study suggests some important practical guidelines, including
i) increase the awareness of the product to make customers purchase it in early stages and ii)
provide some incentives or convenient development tools to application developers for fast-growing
applications.
Our analysis has several limitations that demand some consideration. First, we modeled cus-
tomers as arriving in two periods, and constrained the rst-period customers to either buy in the rst
period or vanish from the market. Relaxing this assumption might reduce second period prots, and
it appears to shift the strategic choice of expansion timing away from deferred expansion. However,
we believe it does not materially aect the results. Second, in computing the rst-period customers
utility for the product, our model ignored the anticipated value from having a collection of third-
party developers in the later period. This view is reasonable if the rst-period customers apply a
very high discount rate for future benets of this sort, e.g., when they are technology innovators
who purchase a new technology based primarily on visible standalone features. The assumption was
also motivated by reasons of computational tractability (which is harmed by the inclusion of such
anticipated benets). However, we conducted numerical experiments and conrmed that our main
ndings still hold: with anticipated benets, more early adopters would purchase the product in the
rst period, but this does not change a rms selection of optimal product launch strategy. Despite
these limitations, we hope that the conceptual insights provided in this paper will be of value to a
spectrum of rms that develop new platform technologies.
This paper can be improved in additional ways, which motivate a few directions for future re-
search. First, future research can focus more on technology startups. It is generally more dicult for
startups to attract developers and early adopters. Therefore, many startups would be keenly inter-
ested in the eect of other pricing/marketing strategies that could help attract more early adopters
(e.g., free distribution, free trial-version) and/or improve developers participation (e.g., increasing
28
technology investment for App development, providing incentives to developers, and identifying op-
timal contract mechanism). Second, while we assumed that the manufacturer already acquired the
innovative technology with R&D investment (i.e., development cost was sunk), it would be useful
to examine how uncertainty in developer participation impacts the level of innovation. This problem
might be particularly important to many startups because of their limited resources. Note that
with limited resources, identifying an appropriate level of innovation is a very important decision
problem. Third, further investigation of the two product line expansion strategies with more gen-
eral assumptions (e.g., continuous distribution for uncertainty, relaxing two assumptions mentioned
above) would improve our understanding of the dynamics of platform product launch strategies.
Fourth, product line expansion is often dictated by technological improvement, a factor ignored in
our model and other studies of sequential vs. simultaneous versioning (Moorthy and Png, 1992).
It would be useful to examine how the expansion strategy is impacted by the interplay between
technological improvement and other factors considered in this paper. Considering the fact that
many startups gradually improve their technologies over time, this extension will shed more light on
startups technology commercialization strategies.
Acknowledgement
The authors thank the guest editors (Prof. Moren Levesque & Prof. Nitin Joglekar), senior editor,
and two referees for their contribution in improving the clarity and quality of this paper.
References
Aoki, R. and T. J. Prusa (1997): Sequential versus simultaneous choice with endogenous
quality, International Journal of Industrial Organization, 15, 103121.
Armstrong, M. (2006): Competition in two-sided markets, RAND Journal of Economics, 37,
668691.
Armstrong, M. and M. L evesque (2002): Timing and Quality Decisions for Entrepreneurial
Product Development, European Journal of Operational Research, 141, 88106.
29
Bag, P. K. and S. Roy (2010): On sequential and simultaneous contributions under incomplete
information, International Journal of Game Theory, Forthcoming.
Bhargava, H. and V. Choudhary (2001): Information good and vertical dierentiation,
Journal of Management Information Systems, 18, 89106.
(2004): Economics of an information intermediary with aggregation benets, Information
Systems Research, 15, 2236.
Cachon, G. and M. Lariviere (2001): Contracting to assure supply: how to share demand
forecasts in a supply chain, Management Science, 47, 629646.
Carrillo, J. (2005): Industry clockspeed and the pace of new product development, Production
and Operations Management, 14, 125141.
Chen, F. (2005): Salesforce Incentives, Market Information, and Production/Inventory Planning,
Management Science, 51, 6075.
Choi, Y., M. L evesque, and D. Shephert (2008): When should entrepreneurs expedite or
delay opportunity exploitation, Journal of Business Venturing, 23, 333355.
Christensen, C. and J. Bower (1996): Customer power, strategic investment, and the
failure of leading rms, Strategic Management Journal, 17, 197218.
Daneels, E. (2004): Disruptive technology reconsidered: A critique and research agenda,
Journal of Product Innovation Management, 21, 246258.
Davidsson, P., P. Steffens, and J. Fitzsimmons (2008): Performance assess-
ment in entrepreneurship research: is there a pro-growth bias? Downloadable at from
http://eprints.qut.edu.au/archive/00012040.
(2009): Growing protable or growing from prots: Putting the horse in front of the cart?
Journal of Business Venturing, 24, 388406.
DeCastro, J. and J. Chrisman (1995): Order of market entry, competitive strategy and
nancial performance, Journal of Business Venturing, 33, 165177.
Deneckere, R. and R. McAfee (1996): Damaged goods, Journal of Economics and Man-
agement Strategy, 5, 149174.
Dhebar, A. (1993): Cambridge Software Corp. Harvard Business School.
Dogan, K., Y. Ji, V. S. Mookerjee, and S. Radhakrishnan (2010): Managing the
versions of a software product under variable and endogenous demand, Information Systems
Research, Forthcoming.
Eisenmann, T. (2007): Platform-Mediated Networks: Denitions and Core Concepts, Harvard
Business School.
30
Eisenmann, T., G. Parker, and M. Van Alstyne (2006): Strategies for two-sided mar-
kets, Harvard Business Review, 84, 92101.
(2011): Platform Envelopment, Strategic Management Journal, Forthcoming.
Ha, A. Y. and S. Tong (2008): Contracting and Information Sharing Under Supply Chain
Competition, Management Science, 54, 701715.
Hagiu, A. (2006): Pricing and commitment by two-sided platforms, RAND Journal of Eco-
nomics, 37, 720737.
Herbig, P. and J. Milewicz (1995): The Relationship of Reputation and Credibility to Brand
Success, Journal of Consumer Marketing, 12, 510.
Jing, B. (2007): Network Externalities and Market Segmentation in a Monopoly, Economic
Letters, 95, 713.
Joglekar, N. and M. L evesque (2006): Marketing, R&D, and startup valuation, IEEE
Transactions on Engineering Management, 56, 229242.
Joshi, Y. V., D. J. Reibstein, and J. Zhang (2009): Optimal entry timing in markets with
social inuence, Management Science, 55, 926939.
Katz, M. L. and C. Shapiro (1992): Product Introduction with Network Externalities, The
Journal of Industrial Economics, 40, 5583.
Lee, Y. and G. OConnor (2003): New product launch strategy for network eects products,
Journal of the Academy of Marketing Science, 31, 241255.
L evesque, M. (2000): Eects of Funding and Its Return on Product Quality in New Ventures,
IEEE Transactions on Engineering Management, 47, 98105.
L evesque, M. and D. Shepherd (2004): Entrepreneurs choice of entry strategy in emerging
and developed markets, Journal of Business Venturing, 19, 2954.
Liu, H. and P. Chintagunta (2009): Pricing under network eects, in Handbook of pricing
research in marketing, Edward Elgar Publishing, Inc., MA, USA, 435450.
Mahajan, V. and E. Muller (1998): When is it worthwhile targeting the majority instead of
the innovators in a new product launch? Journal of Marketing Research, 35, 488495.
Mahajan, V., E. Muller, and F. M. Bass (1990): New product diusion models in
marketing: A review and directions for research, Journal of Marketing, 54, 126.
March, J. (1991): Exploration and Exploitation in Organizational Learning, Organization Sci-
ence, 2, 7187.
Markman, G. and W. Gartner (2002): Is extraordinary growth protable? A study of Inc.
500 high-growth companies, Entrepreneurship Theory and Practice, 27, 6575.
31
Mitra, K. (2010): One D too many, Business Today. October 3, 2010.
Moe, W. W. and P. S. Fader (2002): Using advance purchase orders to forecast new product
sales, Marketing Science, 21, 347364.
Moore, G. (1991): Crossing the chasm, New York: Harper Business.
Moorthy, K. S. and I. P. L. Png (1992): Market segmentation, cannibalization and the
timing of product introductions, Management Science, 38, 345359.
Nuttall, C. (2010): Lack of content leaves 3D TV sales at, Financial Times. December 16,
2010.
Padmanabhan, V., S. Rajiv, and K. Srinivasan (1997): New products, upgrades, and
new releases: A rational for sequential product introduction, Journal of Marketing Research, 34,
456472.
Parker, G. and M. Van Alstyne (2005): Two-sided network eects: A theory of information
product design. Management Science, 51, 14941504.
Ramdas, K. (2003): Managing product variety: An integrative review and research directions,
Production and Operations Management, 12, 79100.
Ramezani, C., L. Soenen, and A. Jung (2002): Growth, corporate protability, and value
creation, Financial Analysts Journal, 62, 5667.
Reinhardt, G. and M. L evesque (2004): A new entrants decision on virgual versus bricks-
and-mortar retailing, Journal of Electronic Commerce Research, 3, 136152.
Rochet, J. and J. Tirole (2003): Platform competition in two-sided markets, Journal of
the European Econmic Association, 1, 9901029.
(2006): Two-sided markets: a progress report, RAND Journal of Economics, 37, 645667.
Shan, W., G. Walker, and B. Kogut (1994): Interrm cooperation and startup innovation
in biotechnology industry, Strategic Management Journal, 15, 387394.
Shy, O. (2001): The economics of network industries, Cambridge University Press.
Van den Bulte, C. and Y. V. Joshi (2007): New product diusion with inuentials and
imitators, Marketing Science, 26, 400421.
Villas-Boas, J. M. (2004): Communication Strategies and Product Line Design, Marketing
Science, 23, 304316.
West, J. and S. OMahony (2005): Contrasting Community Building in Sponsored and Com-
munity Founded Open Source Projects, Proceedings of the 38th Hawaii International Conference
on System Sciences.
32
Online Supplement for
Commercialization of Platform Technologies:
Launch Timing and Versioning Strategy
A Asymmetric Valuations Across Time
Our formulation assumed that the aggregate customer valuations were identical in both periods, with
v being uniformly distributed on [0, 1] in both periods. However, it could be also true that the early
adopters in the market are likely to be the ones with higher valuations for the product (especially so,
if they are willing to purchase without any third-party applications); therefore assuming that early
adopters and followers have the same valuation for the product is a simplication. In this section,
we relax this assumption to see if the main results still hold.
Consider the case where valuations of rst period customers (early adopters) are higher than that
of second period customers (followers). This can be achieved by setting the rst period customers
valuations as uniformly distributed on [0, b] where b > 1. We examine the impact on our earlier
nding about the linkage between the expansion strategy and the uncertainty about developer
participation (recall Figure 1(a)).
Early
Expansion
Early
Expansion
No
Expansion
No
Expansion
Defer,
Expand
Expansion Cost h
R
a
n
d
o
m
O
f
f
s
e
t



Early
Expansion
Early
Expansion
No
Expansion
No
Expansion
Defer,
Expand
Expansion Cost h
P
r
o
b
a
b
i
l
i
t
y



Solid: b=1
Dashed: b=3
Expansion Cost h
R
a
n
d
o
m
O
f
f
s
e
t

Figure 3: Eect of Asymmetric Customer Valuations on Optimal Expansion Strategy


1
Figure 3 illustrates the impact of this relaxation: shift in the cuto lines for the dierent strategies
(the solid lines represent the base case b = 1 while the dashed lines represent b = 3). The main
impact is that early expansion becomes more attractive. The intuitive explanation for this observation
is that, as b increases above 1, the product becomes more attractive in the rst period, leading to
higher market share and increasing the likelihood of high developer participation. Therefore, early
expansion is preferred for a higher threshold of expansion cost. Overall, our main nding holds with
the relaxation, and the impact is in a predictable way.
B Technical Details for Lemma 1 and 2
We organize our analysis in terms of the two high-level strategic choices that the rm faces at the
start of the game, whether to expand the product line early or to defer the expansion decision to
the second period after uncertainty about developer participation is resolved.
B.1 Early Expansion of Product Line
In this case, because the rm incurs its product line expansion cost in the rst period and enters
the second period with {L, H}, the second period decision is simply a pricing problem for the
two products. The optimal second-period prices will depend on the realized level of developer
participation, hence we obtain
1. Case B (high level of developer participation): Taking derivatives of the second-period prot
term against
L
and
H
, the second-period prices are solutions to the system of simultaneous
equations representing the rst-order conditions. We have the optimal prices

early
L
=
+
_

_
1
p
L

_
+
_
2
,

early
H
=
1 +
_

_
1
p
L

_
+
_
2
,
2
and they yield the second-period prot as

B
2
(D, {L, H}) =
+ 2
_

_
1
p
L

_
+
_
+
2
_

_
1
p
L

_
+
_
2
4
. (18)
2. Case W (low level of developer participation): By setting = 0 for the optimal solutions
of Case B, we can obtain
L
and
H
, and the system of simultaneous equations yields the
second-period prot as

W
2
(D, {L, H}) =
+ 2
_

_
1
p
L

__
+
2
_

_
1
p
L

__
2
4
. (19)
Combining these two prot terms (Eq. 18 and Eq. 19) into Eq. 5 and then into Eq. 8, the optimal
expected total prot, conditioned on early expansion, is

early
= max
p
L
,p
H
_
_
_
_
_
_

_
p
H
_
1
p
H
p
L
1
_
+ p
L
_
p
H
p
L
1

p
L

__

+

B
2
(D; {L, H}) + (1 )
W
2
(D; {L, H})
_
_
_
_
_
_
(20)
where D =
_
1
p
L

_
. Solving it optimally yields Lemma 1.
B.2 Deferred Expansion Decision
Now, consider the case where the rm launches only H in the rst period. In order to compute
expected second-period prot under this strategy (E[
2
({L, H}, D)], Eq. 7), we need to solve
second-period price-optimization problems and compute the following three terms,
W
2
(D; {H}),
3

B
2
(D; {H}), and
B
2
(D; {L, H}).

B
2
(D; {H}) =
(1 + (D + ))
2
4
, (21)

B
2
(D; {L, H}) =
(1 + 2(D + )) +
2
(D + )
2
4
, (22)

W
2
(D; {H}) =
(1 + (D))
2
4
. (23)
Combining these three terms into Eq. 7 and then substituting into Eq. 9, the total expected
prot from a deferred expansion strategy is (after replacing D with (1 p
H
))

defer
= max
p
H
_
_
_
_
_
_
_
(p
H
(1 p
H
))) + (1 )
(1+((1p
H
)))
2
4
+ max
_

_
(1+((1p
H
)+))
2
4
,
(1+2((1p
H
)+))+
2
((1p
H
)+)
2
4

_
_
_
_
_
_
_
. (24)
Solving separately the two optimization problems implied by the max term inside Eq. 24, we
obtain Lemma 2.
C Benchmark: No Network Eect
The main goal of our paper is to isolate the impact of the platform characteristics (i.e., cross-network
eects) and developer participation uncertainty on the rms product launch and timing decision.
In order to see this, we describe two benchmark cases representing zero network eects and/or zero
uncertainty about developer participation.
1. First, set = 0, i.e., consumer valuations for the product do not depend in any way on
developer participation (this setting also, in eect, makes developer uncertainty irrelevant).
Due to this, the two periods collapse into one, and with the linear utility form U(v, q) = v q,
it follows from past literature that versioning is not optimal for the rm (see, e.g., Bhargava
4
and Choudhary, 2001; Deneckere and McAfee, 1996).
2. Second, set = 0, i.e., developer participation does not depend on installed base. Moreover,
let = 1, i.e., there is no uncertainty about developer participation. Now, second-period
customers utility has the form vq + k due to which it follows from Bhargava and Choudhary
(2004) that versioning is optimal in period 2, i.e., the rms second period product line is
{L, H}. Normally, the uncertainty about developer participation should make the rm cautious
about incurring xed costs needed to expand its product line early. But, it can trivially be
seen that, even in the absence of this uncertainty (and even though the cost of launching L
is the same in either period) the rm has no incentive to launch L in the rst period.
What these two benchmark cases demonstrate is that (a) when the user side of the platform
does not perceive cross-network benets (i.e., users do not care about developer participation), then
versioning is not optimal for the rm, and (b) when the developer side does not care about installed
base, then the rm has no reason to expand its product line early despite its fore-knowledge about
developer participation and about the positive impact of developer participation on valuations of
second-period consumers.
D Other Proofs
Proof of Proposition 1. Note that the incremental prot from versioning is

versioning
2
=

2
Q
2
(1 )
4
> 0.
By taking rst derivative of this prot with respect to Q, we have

versioning
2
Q
=

2
Q(1 )
2
> 0.
5
Proof of Proposition 2. We rst consider early expansion. Recall from Lemma 1 that the
prot under early expansion is as follows:

early
=
(4(1 4 + + + 2) +
2
(
2
(1 + ) + 4 + 4
2
))
16
2
4
2

2
((1 4 + ) + (1 )
2

2
)
16
3
4
2

2
.
Suppose a rm foresees product expansion inevitable in period 2. Then, following the same logic
described in Appendix B, we can derive total expected prot as

defer
=

2
(4 16 + (2 + )
2
+ 8)
2
(1 )
4

2
4(4
2

2
)
+

2
(
2
(4 1) + 4 + 4
2
)
4(4
2

2
)
.
By taking dierence of these two prots, we have

early

defer
=
(1 )
2

2
(2 + + 2)
2
4(4
2

2
)(4
2

2

2
)
.
Because 4
2
>
2
is a necessary condition for positive prices, the prot dierence is always
positive. Therefore, it is optimal to expand in period 1 when product expansion is inevitable.
Proof of Proposition 3. Recall from Lemma 1 that the prot under early expansion is as
follows:

early
=
(4(1 4 + + + 2) +
2
(
2
(1 + ) + 4 + 4
2
))
16
2
4
2

2
((1 4 + ) + (1 )
2

2
)
16
3
4
2

2
.
6
Also, recall from Lemma 2 that the prots are

defer, H-only
=
4 + 4(2 + ) + (4 + (4 + (4 (1 )
2
)))
4 (4
2

2
)
,

defer, expand
=

2
(4 + 4
2

2
(1 4 + (1 )( 2)))
16 4
2
((1 ) + )
2
+
(4(1 + + ) + (
2

2
+ 2 (4
2
(1 )
2
) ( 2)))
16 4
2
((1 ) + )
2
.
We rst compare defer, expand with defer, H-only. We nd the cuto point in terms of at
which
defer, expand

defer, H-only
= 0. Note that the rst derivative of prot dierence with
respect to is , implying that prot dierence is monotonically decreasing in . The detailed
expression of
HD
is

HD
=
(1 )
2
(4 + (2 + (1 (1 ))))
2
4(4
2

2
)(4
2
((1 ) + )
2
)
. (25)
Because 4
2
>
2
(a necessary condition for positive prices), we see that 4 >
2

2
and
4 >
2
((1 ) + )
2
). Therefore,
HD
> 0, implying that
defer, expand
>
defer, H-only
when <
HD
, while
defer, expand
<
defer, H-only
when >
HD
.
We now compare early expansion with deferred expansion. The prot dierence is denoted with
f() (i.e., f() =
early

defer, expand
). Note that
f()

= 1 + < 0, meaning that f()


is a monotonically decreasing in . Next, we nd the cuto point of at which prot dierence
becomes zero. We denote this cuto point with
DE
. Note that

DE
=
(1 )
2

2
(4
3
(1 + (1 )(1 + )) +
2
(1 )
2

2
)
4(1 )(4
2

2

2
)(4
2
((1 ) + )
2
)
+
(1 )
2

2
(
2
((4 + (1 )(1 + )) + 4(2 + (1 )))
4(1 )(4
2

2

2
)(4
2
((1 ) + )
2
)
+
(1 )
2

2
(
2
(4 + 4
2
+
2
( + 2(1 ))))
4(1 )(4
2

2

2
)(4
2
((1 ) + )
2
)
> 0. (26)
Therefore,
early
>
defer, expand
when <
DE
, while
early
<
defer, expand
when >
DE
.
7
Finally, we compare early expansion with defer, H-only. We dene prot dierence as
early

defer, H-only
. Note that the rst derivative of the prot dierence with respect to is 1,
implying that prot dierence is monotonically decreasing in . We nd the cuto point of which
leads to zero protm denoted with
HE
. Note that

HE
=
(1 )
2
(
2
(4(1 + + ) + (4 + (1 + ))))
4(4
2

2

2
)(4
2

2
)
+
(1 )
2
(4(2(2 + ) + (2 + )))
4(4
2

2

2
)(4
2

2
)
+
(1 )
2
((16
2
4
2
(1 + ((1 ) ))
2
+
4
(1 )
2

4
)
2
)
4(4
2

2

2
)(4
2

2
)
> 0.
(27)
This implies that
early
>
defer, H-only
when <
HE
, while
early
<
defer, H-only
when
>
HE
. Combining all three cases completes the proof.
Proof of Proposition 4. Recall from the proof of Proposition 3 that
HD
and
DE
are as
follows:

HD
=
(1 )
2
(4 + (2 + (1 (1 ))))
2
4(4
2

2
)(4
2
((1 ) + )
2
)
,
and

DE
=
(1 )
2

2
(4
3
(1 + (1 )(1 + )) +
2
(1 )
2

2
)
4(1 )(4
2

2

2
)(4
2
((1 ) + )
2
)
+
(1 )
2

2
(
2
((4 + (1 )(1 + )) + 4(2 + (1 )))
4(1 )(4
2

2

2
)(4
2
((1 ) + )
2
)
+
(1 )
2

2
(
2
(4 + 4
2
+
2
( + 2(1 ))))
4(1 )(4
2

2

2
)(4
2
((1 ) + )
2
)
.
8
First, we show that
HD
<
DE
when = 0. Note that

HD
|
=0

DE
|
=0
=
(1 )
2

2
4(4
2
((1 ) + )
2
)(1 )(4
2

2
)(4
2

2

2
)

((4 + (1 )(1 + ))(4
2

2
) +
4
2
(4
2

2
(2 + (1 ) + (1 ))) +

2
(4 + (4(1 ) + (1 (1 + ))))).
Since the rst fraction is negative, we want to investigate the sign of the remaining part. Let
= (4 + (1 )(1 + ))(4
2

2
) +
4
2
(4
2

2
(2 + (1 ) + (1 ))) +

2
(4 + (4(1 ) + (1 (1 + ))))).
The rst derivative of with respect to is

= (1 )
2

2
(4(1 + + ) + (4 + (1 + ))) > 0,
implying that is monotonically decreasing in . Thus, has its minimum value at = 1,
which is (2 + )
2
(4
2

2
) > 0. This proves that the entire equation is negative, i.e.,

HD
|
=0
<
DE
|
=0
.
Second, we show that there exists a unique intersection between
HD
and
DE
. For each of
these two functions, we want to show that (i) they are convex, (ii) stationary points are negative,
and (iii) they are positive when = 0. Consider
HD
rst. Note that

HD

2
=
(1 )(4
2
(1 )
3
)
2
2(4
2

2
)(4
2
((1 ) + )
2
)
> 0,
which proves convexity of
HD
. Also note that the rst derivative of
HD
with respect to becomes
9
zero at
=
(2 + )
4
2
(1 )
2
< 0.
Finally,

HD
|
=0
=
(1 )
2

2
(2 + )
2
4(4
2

2
)(4
2
((1 ) + )
2
)
> 0.
Next, we examine
DE
. Note that

DE

2
=
(1 )
2

4
(4 +
2
(1 )
2
)
2(4
2

2

2
)(4
2
((1 ) + )
2
)
> 0,
implying convexity of
DE
. The rst derivative of
DE
with respect to becomes zero at
=
((2 + (1 )) + 2(2 + (1 )))
4 +
2
(1 )
3
< 0.
Also we have

DE
|
=0
=
(1 )
2

2
(
2

2
+ (4 + (1 )(1 + )) + 4
2
((1 )(1 + )))
4(1 )(4
2

2

2
)(4
2
((1 ) + )
2
)
> 0.
Next, we examine tangent of
HD
and
DE
at = 0. Note that

HD

=0


DE

=0
=
(1 )
2
2(4
2
((1 ) + )
2
)

_
(2 + )(4
2
(1 )
2
)
4
2

2

((2 + (1 )) + 2(2 + (1 )))
(1 )(4
2

2

2
)
_
.
Note that the rst fraction is positive, we focus on the second part. Let
=
(2 + )(4
2
(1 )
2
)
4
2

2
.
10
Then

=

2

2
(2 + )
4
2

2
> 0.
Since is increasing in , has its minimum value at = 0, which is 2 + . Let
= 2 + +
((2 + (1 )) + 2(2 + (1 )))
(1 )(4
2

2

2
)
.
We have

=
2(4(2 + (1 )) + 4
2
(2 + (1 )) +
2

2
(2 + (1 )))
(1 )(4
2

2

2
)
2
> 0,
implying that has its mimimum value at = 0. Note that |
=0
=
2
1
+ (1 + ) > 0.
Therefore,

HD

=0
>

DE

=0
. Combining all results, i.e., (i) two functions are convex, (ii) the
only stationary points are negative, (iii) at = 0,
HD
<
DE
, and (iv) at = 0, tangent of
HD
is larger than tangent of
DE
, completes the proof.
Proof of Proposition 5. We dene f() be the dierence of the two prots, i.e., f() =

early

defer, expand
. Note that

2
f()

2
=
(1 )
2

4
(4 +
2
(1 )
2
)
2(4
2

2

2
)(4
2
((1 ) + )
2
)
.
Under the condition, 4
2

2

2
> 0 which is necessary for positive prices, the second derivative
of f() with respect to is positive, implying convexity of f(). We then set the rst derivative of
f() with respect to , i.e.,
f()

equal to zero and solving it, we have

=
(8c
2
+ 4(1 (1 c(1 )))
(1 )
2
(4 +
2

2
(1 ))

2
(2(1 c
2
) 2(1 ) + (1 c )(1 ))
(1 )
2
(4 +
2

2
(1 ))
.
11
Since the denominator of

is positive, we examine the sign of its numerator. Now we use 4


2

2
> 0 to show that
8c
2
+ 4(1 (1 c(1 ))) +
2

2
(2(1 c
2
) 2(1 ) + (1 c )(1 ))
> 2c
2

2
+
2

2
(2(1 c
2
) 2(1 ) + (1 c )(1 )) + 4(1 + c(1 ))
=
2

2
(2 2
2
2(1 ) + (1 c )(1 )) + 4(1 + c(1 ))
>
2

2
(2 2
2
2(1 ) + (1 c )(1 )) +
2

2
((1 + c(1 )))
=
2

2
(2(1 )(1 + (1 )) + (1 )(1 ) + (1 )(1 c(1 ))) > 0,
which proves that the only stationary point in terms of is negative. Combining this result with
convexity of f() concludes that for any > 0, f() is monotonically increasing. Next, we consider
f() |
=0
. Note that
f() |
=0

= 1 + < 0.
Thus, there exists which leads to the following results. When > , we have f() |
=0
< 0. For
this case, we see that there exists a unique cuto point

which satises f() |
=

= 0. Therefore,
if <

, then f() |

< 0, which implies that defer, expand generates higher prot. Otherwise, i.e.,
if >

, then early expansion is optimal. When < , f() |
=0
> 0, which implies f() > 0 for
any > 0 (recall that f() is monotonically increasing in ). Thus, defer, expand launch always
outperform early expansion, because f() > 0 implies
early
>
defer, expand
. This completes
the proof.
Proof of Proposition 6. First, we use construction approach to show that defer, expand H
is always better than defer, expand L, when c = 0. Note that the rst period prots from defer,
expand H and defer, expand L are (p
H
c)(1p
H
) and p
L
_
1
p
L

_
, respectively. Thus, holding
12
the price, the prot dierence function in period 1 becomes
(p c)(1 p) p
_
1
p

_
=
p
2
(1 )

> 0,
which implies that the rst period prot is always higher with defer, expand H than with defer,
expand L if prices are set at the same level. We then note the demand dierence as
(1 p)
_
1
p

_
= p
_
1 +
1

_
> 0.
This implies that demand is also higher with defer, expand H than with defer, expand L at the same
price. Therefore, we see that when versioning occurs in the second period, the total prot is also
higher with defer, expand H than with defer, expand L, because demand in the rst period is higher.
Thus, we conclude that defer, expand H is dominating defer, expand L when c = 0.
Next, we note that

2
g(c)
c
2
=
4(1 ) + (4 + (1 )(4 ))
8 2
2
((1 ) + )
2

< 0,
implying that g() is a concave function of c. Combining these two results with preconditions of this
proposition completes the proof.
13

Vous aimerez peut-être aussi