ASP On-Demand versus MOTS In-House Software Solutions*
Dan Ma
School of Information Systems
Singapore Management University
Abraham Seidmann
Xerox Professor of Computer and Information Systems and Operations Management
W. E. Simon Graduate School of Business Administration
University of Rochester
Revised:
April 29, 2007
*
The authors gratefully acknowledge useful comments from Professors Rajiv Dewan, Roy Jones, Ravindra
Mantena, Edieal Pinker, Michael Raith, and Anjana Susarla, and from seminar participants at the
Universities of Michigan, Rochester, and Maryland, and the Workshops on Information Systems and
Economics (WISE).
1
ASP On-Demand versus MOTS In-House Software Solutions*
Dan Ma
School of Information Systems
Singapore Management University
Abraham Seidmann
Xerox Professor of Computer and Information Systems and Operations Management
W. E. Simon Graduate School of Business Administration
University of Rochester
Abstract
Application Service Providers (ASPs) deliver on-demand information processing services to user
firms via the Internet. Rapid technological developments in telecommunications in recent years
have made ASPs an attractive alternative to purchasing, installing, and maintaining modifiable
off-the-shelf (MOTS) software solutions. We study several critical aspects of a user firm’s choice
between an ASP and MOTS software. The competitive model considers heterogeneous users that
differ in terms of their expected transaction volumes and volatility, while ASPs and MOTS
vendors differ in terms of their pricing structure, setup cost, system customization ability,
integration and service level arrangement. Our results identify and characterize the equilibrium
conditions under which ASPs and MOTS vendors can coexist in a competitive market, and they
explain which firms could be the primary beneficiaries of each vendor type. Interestingly, we find
that the ASP' entry to the market benefits not only its own clients but also the MOTS users, and
we explain why despite that, the competitive advantage of the MOTS approach decreases
significantly as users’ transaction volatility, or business uncertainty, increases. Hence, the value
added by ASPs comes as much from the efficient pooling of transaction volatility risks as from
the reduction in IT implementation costs. We also look at the competition between MOTS and
ASO vendors in a longer time horizon with possible future changes in software quality, and we
show when the ASP can take over the whole market, even if its quality is still lower than the
MOTS software. Our findings suggest that to stay competitive, ASPs should invest in both
improving software quality and in reducing service prices over time. Surprisingly, the opposite
also holds, and along with any relative quality decreases, the ASP should raise its prices,
providing a relatively an inferior service at a higher charge. This happens because at that case the
ASP will serve mainly those firms who are priced out of the MOTS market, and can behave
monopolistically at that segment. Finally, our model explains when MOTS software users could
be better off despite an increase in transaction volatility in the market, and when they will be less
likely to switch to an ASP even as the relative quality of the ASP’s applications increases over
time. We prove that if they do switch, their primary incentive is abating higher transaction or
business volatility risks rather than reducing transactions’ costs per se.
*
The authors gratefully acknowledge useful comments from Professors Rajiv Dewan, Roy Jones, Ravindra
Mantena, Edieal Pinker, Michael Raith, and Anjana Susarla, and from seminar participants at the
University of Rochester, the University of Maryland, the Workshop on Information Systems and
Economics (WISE) 2004, and WISE 2005.
2
Introduction
Using sophisticated enterprise software such as Enterprise Resource Planning (ERP),
Customer Relationship Management (CRM), or Electronic Medical Records (EMR) systems has
become a competitive necessity. Traditionally, such software has been delivered in the form of
modifiable off-the-shelf (MOTS) products.1 The vendor sells the software application to users
and helps to customize and install it on users’ sites. The users must provide IT infrastructure,
hardware, and support services in order to enable continuous use of the software. In the past few
years, the Internet has given rise to Application Service Providers (ASPs). ASPs offer a bundle of
applications, an IT infrastructure, and all necessary support services to users across a network.
Under the ASP business model, the application and users’ data are stored off-site in a central
location run by the ASP. The ASP is in charge of all IT support services, including daily software
maintenance, data backups, software upgrades, and security.
ASPs represent an on-demand business model, where software is delivered as a service.
Unlike traditional perpetual licensing, the software is priced as a service, and typically users pay a
fee per transaction. Users’ payments are closely tied to the actual utility obtained—they pay only
when they have demand for the software.2 Such pricing shifts expenses from the capital budget to
the operating budget and significantly cuts users’ start-up costs; it also helps users to manage
peak-and-valley transaction-load problems well while keeping costs down. In many cases, using
an ASP may prove cheaper than owning and maintaining an in-house IT system. Users expect to
save money on support and upgrade costs, IT infrastructure, IT personnel, and implementation
(Lacy 2006). A recent survey shows that the top reason firms choose an ASP over traditional
software is the belief that the former is “cheaper to run” (The Economist 2006). Moreover, users
1
A MOTS product is “a commercial software application whose source code can be customized to meet a
customer’s particular requirements ... [It] is designed to be easily installed and to interoperate with existing
system components.” See http://whatis.techtarget.com for more information.
2
In this paper, we use the terms “ASP,” “on-demand software,” and “software as a service”
interchangeably.
3
can reap the benefits of the ASP’s economy of scale, access IT expertise, and gain system
flexibility (Dunn 2005; Murphy 2005; Singh et al. 2004).
ASPs are enjoying prosperous times. According to AMR Research, the on-demand software
market is growing more than 20% a year, compared with single-digit growth in traditional
software (Lacy 2006). ASPs are expected to generate $10 billion in annual revenue by 2009, up
from $1.5 billion in 2006 (Pallatto 2006). An increasing number of software vendors, including
industry giants such as IBM, SAP, Oracle, and Microsoft, are moving to such a business model.
For example, IBM is offering “IBM-on-demand,” which allows corporate users to acquire IBM’s
computing power and software applications as a service.3 Meanwhile, IBM is also providing a
package of services and incentives to help other software companies deploy their products as
hosted applications (Pallatto 2006). Oracle is ranked as one of the top ten ASPs. Its online
offering, “E-business Suite,” has successfully attracted many small companies by charging a low
transaction fee.4 In January 2006, Oracle acquired Seibel, and it now uses its fledgling CRM ondemand software to compete with Salesforce.com, the most successful ASP providing online
CRM, which reports 399,000 paying subscribers (Vara 2006). Bill Gates has proclaimed that the
emergence and rise of on-demand software will be the “next sea-change” in computing (Niccolai
2005). Many observers expect an industry-wide revolution brought on by the ASP model:
“Software as a service is the biggest thing to happen in software in 25 years” (Knorr 2004).
For firms in need of enterprise software, the ASP constitutes a viable alternative to the
traditional MOTS software solution. As an example, we studied the case of Medical Diagnostic
Imaging (MDI), an imaging center conducting about 10,000 studies annually (Martin 2004). It
was looking for a state-of-the-art PACS (Picture Archiving and Communications System) to
manage and deliver patients’ medical images. MDI looked first at a MOTS solution from such
vendors as Cannon, Onyx, and Fusion. All of them would have required MDI to build and
3
4
See http://www-1.ibm.com/services/ondemand/success.html.
See www.oracle.com.
4
maintain the PACS in-house. The bids from these vendors typically exceeded one million dollars
just for the up-front investment in the software. Additionally, MDI identified significant costs
associated with installation, customization, ongoing support, upgrades, data security and aroundthe-clock maintenance. MDI determined that an on-site MOTS system was too expensive. To
reduce the required initial costs, MDI chose an ASP, MedPACS Display, which provides viewer
software for radiologists and physicians, as well as archives, hardware, software, and
maintenance.5 This ASP charges a flat fee of $3.89 per study, including all the necessary ongoing
support (Ma 2005).6 MDI thus was also relieved of the need to hire and manage an internal IT
staff for the radiology department. Moreover, the ASP offers highly scalable processing services.
When MDI faces random fluctuations in patient volume, MedPACS responds accordingly from
the remote site that handles most of the clinical data.
Many businesses face a similar decision problem. In a web survey by ThinkStrategies, fully
one-third of 118 respondents were already using ASPs, and another third were considering using
one within the following 12 months (Kaplan 2005). A research report from Summit Strategies
indicates that small and medium businesses with limited IT resources and constrained budgets are
more likely to use such fee-per-transaction applications (Garner 2004). Some large organizations
are attracted to ASPs as well. Amazon.com, Cisco, Sprint, Morgan Stanley, Nokia, and Target are
all clients of Salesforce.com (The Economist 2006; Ricadela 2006). It is clear that ASPs are
stealing market share from vendors of MOTS software. On-demand software is a powerful trend,
one that is becoming an important disruptive force in the software industry: “It is not the end of
software. It is just another way of deploying it” (Shai Agassi, president of SAP’s product and
technology group).
However, the long-term success of ASPs remains uncertain. In its formative stage, the ASP
market is experiencing significant consolidations. The fear that ASPs could fail is a main obstacle
5
See www.medpacs.com.
Most ASPs in the healthcare industry charge similarly for their PACS systems. Stentor, for example,
offers the image delivery software iSite on a per study basis and charges, on average, $3.50 per delivery.
6
5
to the large-scale adoption of ASPs. Data security and reliability as well as application control are
always among users’ top concerns (Bednarz 2006). In addition, although there is no doubt that the
ASP model does have numerous advantages, the traditional MOTS in-house solution can deliver
superb results. The source code of the MOTS software can be modified to meet a user’s particular
requirements, so the software is well customized and can easily interoperate with the user’s
existing system components. In contrast, the ASP offers a one-to-many solution, with limited
customization.7 Its product is not tailored to fit an individual user’s specific requirements or
unique business environment. As a result, the user may need to pay extra to make the ASP’s
standard software application work smoothly with its existing IT systems. For instance, in the
MDI case, getting the various systems to communicate and work together required some effort.
MDI had to run an additional module that “took the output from its own in-house research
information system and converted the output into an image file for storage in the archive” (Martin
2004). MDI incurs conversion costs for every transaction.
Some researchers have already looked at the ASP business model. For example, Susarla et al.
(2003) use questionnaires in order to identify the determinants of a good ASP contract. Their
work does not, however, touch on the theoretical base of the ASP business model. Cheng and
Koehler (2002) study a monopoly ASP’s pricing problem. They consider the interrelationship
between price and queuing delay and conclude that various pricing policies, such as a fixed fee, a
two-part tariff, or a metered price, are equivalent. These and other previous works, however, only
analyze ASPs in an isolated framework. They do not study a marketplace in which MOTS
vendors also participate. Hence, they overlook the in-house IT option.
Little work has been done on the competition between the ASP and MOTS solutions. It is not
clear what factors play important roles in users’ IT choices. We study a software market in which
corporate users have these two IT options for external sourcing. The software vendors are
differentiated in the following ways. First, they deliver different products: a customized software
application (from the MOTS) versus a bundle of standard software and services (from the ASP).
7
SAP believes that this lack of integration will eventually lead most large corporations that currently are
users of online CRM to move back to an in-house system (Hamm 2006).
6
Second, they adopt distinct pricing modes: an outright purchase (the MOTS) versus a “per
transaction” fee structure (the ASP). Third, they employ different delivery methods: software
installed on a user’s in-house server (the MOTS) versus an interface delivered over the Internet
remotely (the ASP). Users therefore have different cost structures and bear different risks in
dealing with each vendor type.
In practice, choosing and implementing an enterprise software system are highly complex
decisions. CIOs need to compare the two options on multiple points, such as setup costs, pricing
structure, ability to integrate data, scalability, and service level arrangements. Their decisions
involve not only economic but also technological and security issues. We do not attempt to
include all the related factors; instead, we study firms’ choices by focusing on the economics
issues only. In particular, we are interested in estimating the relative economic advantages of
using each vendor type. We assume that the users (they are firms) have made the decision to
install a computerized information system, aiming to improve internal business processes or to
increase value for end customers. The decision investigated in this paper is the choice of delivery
channel: obtaining the system through an ASP or MOTS. Strategic adoption behaviors such as
herding or delaying adoption for a better system or lower price are ignored.
ASPs and MOTS vendors price their offerings differently. The MOTS vendor sells the
software (one-time, fixed-fee pricing), and the ASP rents it (transaction-based pricing). Our work
thus is related to research on selling and renting information goods. Most previous work in this
literature examines a one-firm setting. Varian (2000), for example, considers a monopoly firm’s
choice when both selling and renting are feasible channels. Choudhary et al. (1998) analyze the
possibility of a monopoly firm selling and renting a package of software simultaneously, and
Sundararajan (2004) compares fixed-fee pricing and usage-based pricing for information goods
for a monopoly firm. Fishburn et al. (1997) are among the few who compare selling and renting
in a competitive market. They model a repeated competition game in which one firm charges a
fixed subscription fee and the other charges based on usage. In their paper, the two sellers’
offerings are homogeneous; the only difference between the two sellers is the pricing method.
7
We, however, study the competition between two vendors that are differentiated in both offerings
and pricing structures. We identify several interesting features of such a market.
First, the value added by ASPs comes as much from the efficient pooling of demand
uncertainty risks as from the reduction in significant IT implementation costs. Specifically, when
the volatility of transactions increases, more users opt for the ASP, and the relative advantage of
the ASP approach increases significantly.
Second, we find that the software market may show three different structures. Under fairly
broad conditions, the ASP and MOTS solutions will coexist. The market will be segmented in
such a way that firms with low transaction volumes opt for the ASP model because of the
cheapness and scalability, and firms with high transaction volumes prefer the MOTS model to
enjoy software that fits their specific business needs well. Under certain conditions, however, one
of the providers is unable to survive, and the software market will be dominated by one business
model alone.
Third, our findings suggest that different business models are suited for different types of
software applications. In particular, the ASP model works well for software applications such as
financial trading, electronic medical records, and payroll systems. These applications typically
serve functions with highly volatile demand but a standard process. For complex business
software, such as enterprise resource planning (ERP) systems, we still suggest the traditional inhouse model.
We also study the ASP-MOTS competition in a longer time window with potential real-time
changes. We obtain surprising yet important findings. If users are concerned about future changes
in the business environment, such as hardware upgrades on their side, which decreases the ASP
software quality, ASPs should increase their prices. They should “give up” the competition with
the MOTS vendor for high-volume users and instead focus on attracting small and medium firms.
If users expect the unfit costs of using standard software to decrease due to the advance of web
technologies, which increases the quality of ASP software, ASPs should reduce their prices to
8
compete aggressively with MOTS vendors for large corporate users. These findings help us to
suggest useful competitive strategies to providers of on-demand software.
Finally, we show that competition in a longer window gives advantages to the MOTS
solution. Once the initial implementation costs are sunk, users only need to pay ongoing
maintenance costs. Hence, users of existing in-house systems have little incentive to switch to ondemand software, even if the ASP can promise quality improvement over time. If these users do
switch, they do so in order to abate high capacity risks caused by high transaction volatility rather
than to reduce operating costs.
The rest of the paper is organized as follows. In Section 2, we describe the model. We
analyze the ASP-MOTS competition and identify the key determinants of the future of the ASP
business model in Section 3. Section 4 extends the competition analysis over a longer window.
Section 5 concludes the paper.
2
The Model
We first model the market before the ASP appears, and then the market with both vendors.
2.1 The Pre-ASP Market
Software application users in our model are firms. Before the ASP, firms get access to
software systems by installing them in-house. The MOTS vendor sells the packaged software to
its user. The vendor modifies the source code to fit the user’s specific business needs, which
assures a good integration with the user’s existing IT system. The vendor bears an operating cost
C MOTS to serve one user and receives a one-time payment P from the user. For brevity, we
assume that the MOTS vendor provides only one version of the product and that the licensing fee
is not a function of users’ transaction volumes. The MOTS vendor can monitor the number of
users or installations, but it is difficult to monitor the actual transaction volume, which varies over
time. The user must install hardware and IT infrastructures, hire IT staff, and organize an internal
IT group to provide software maintenance, data backups, and security and capacity management.
9
In order to keep these IT resources in-house, the user incurs a variety of costs. Let c denote the
service costs associated with each use of the software (i.e., the service costs per transaction).
Users face stochastic workload in terms of the number of transactions using the software. Ex
ante, users cannot precisely estimate the actual transaction volume. They therefore determine and
install the proper service capacity in-house based on the anticipated transaction volume. If the
actual volume is more than the installed capacity, the user faces an insufficient capacity problem
and loses some potential utility. If the actual volume is less than the installed capacity, the user
faces an overcapacity problem, and it still pays the costs of holding the excess IT capacity. In
both cases, the user incurs losses. It bears capacity risks caused by demand uncertainty.
Figure 2. The ASP Solution
Figure 1. The MOTS Software Solution
The ASP
The software vendor
(service cost c per
transaction)
customizes and
installs the
software with an
operating cost
C MOTS per user
pays a one-time
purchasing
price P
A user firm
stochastic
transaction
volume
with in-house
service capacity
(service cost c per
transaction)
a value of u
per
transaction
delivers the
bundle of
“standard”
software
and support
services
stochastic
transaction
volume
pays a price
pa per
transaction
A user firm
Note: The black dot indicates the location of the software system.
Figure 1 shows the MOTS solution. The vendor sells the software at a price P and bears an
operating cost C MOTS per user. The user incurs service costs c per transaction, and each
transaction creates a value of u.
2.2 The Dual Market—The ASP Competes with the MOTS Vendor
In the dual market, users can access software applications through an ASP.
To start doing business with a user, the ASP first moves the user’s data to its own server,
which costs the ASP a one-time setup cost S. The ASP provides IT services to enable the software
10
use, which costs c per transaction. We assume that the ASP’s service capacity is plenty. In reality,
most ASPs have established large data centers and strong network infrastructures. They design
the service systems with ample capacity as a way of reducing the risk of congestion. ASPs
therefore resemble power stations, and users buy software as a service by plugging into the “ASP
grid.”
The ASP employs a “one to many” business structure. To any individual user, the application
offered by the ASP is not well-customized. Each transaction gives the user a total value of u-t.
The parameter t measures the user’s disutility from not using its ideal product. In many cases, it
also represents the cost of extra effort to make the ASP’s product work with the user’s existing IT
components smoothly. In the rest of the paper, we call t a user’s “unfit costs.” The user pays a
price pa per transaction to the ASP. Figure 2 shows the ASP solution.
2.3 Stochastic Transaction Volume
When modeling users, we capture two characteristics. First, they have different IT needs.
Users’ IT needs are measured by the expected volume of their use of the software. Some firms
may use the software application more frequently (in expectation) than others, and these firms
have larger IT needs. Second, each user’s actual transaction volume is random.
As seen in Figure 3, users are uniformly distributed on a unit-length line normalized from 0 to
1. The location of a user on this line represents its expected transaction volume. The actual
transaction volume of a user at location d i is a random variable that is uniformly distributed on
[d i − θ , d i + θ ] , where the user’s location d i
represents its expected use of the software (in
terms of the number of transactions), and the parameter θ measures the volatility of the
transaction volume. To avoid putting any positive probability on a negative transaction volume,
we do not assume the same distribution for firms with d i < θ . In fact, we make no assumption
on the distribution of their transaction volumes. It can be shown that firms with low transaction
11
volumes are “insignificant” because they will be priced out of the market in the pre-ASP scenario
(see the Appendix).
Figure 3. Distribution of Users’ Transaction Volumes
Expected transaction
volume
di + θ
di
0
1
di − θ
We make the following general parameter assumptions:
Assumption 1. 0 ≤ θ < 1 + CMOTS ; Assumption 2. CMOTS ≤ (u − c ) − cθ (u − c ) .
u
4 4( u − c )
Assumption 1 imposes an upper bound on the transaction volatility level. Note that the
maximum expected transaction volume among all firms is 1, so this upper bound is loose.
Assumption 2 imposes an upper bound on the MOTS vendor’s costs for software
implementations. If C MOTS exceeds this upper bound, no user will be served by the MOTS
vendor in the pre-ASP market, which is a case of no interest.
3
Analysis and Results
3.1 The Pre-ASP Market (the Benchmark Case)
The MOTS vendor determines its optimal price P. Given the price, users make two decisions:
whether to buy the software, and, if so, the proper IT service capacity to install in-house.
We solve the problem backward. Denote the actual transaction volume of user i (located at
d i ≥ θ ) as Di . This user’s optimal capacity level, given the software price P and the unit service
capacity cost c (i.e., the service cost per transaction c), is obtained by solving
Max
E [u min{Di , qi }− P − cqi ] = uqi [1 − F (qi )] + uF (qi )E [Di / Di < qi ] − P − cqi ,
qi
where F (.) is the cumulative density function of the transaction volume for user i. With
probability 1 − F (qi ) , the actual transaction volume will be larger than the user’s service
12
capacity. The user runs the software up to its service capacity level and loses excess demand.
With probability F (qi ) , the actual transaction volume will be smaller than the user’s service
capacity. The user processes all transactions and carries the costs of owning excess capacity. The
first two terms in the expression above represent user i’s expected utility from using the software;
the last two terms represent the user’s costs of purchasing the software (P) and costs of
establishing in-house IT service capacity ( cqi ). Following Dewan and Mendelson (1990), we
assume a linear function for capacity costs. In practice, capacity costs could be a step function,
but a linear function is the best first-order approach for a step function, as Zimmerman (1979)
shows.
⎛
⎝
Solving the optimization problem gives qi* = d i + θ ⎜1 −
2c ⎞
⎟.
u ⎠
(1)
The optimal capacity qi* deviates from the expected transaction volume d i by a factor of
⎛
⎝
θ ⎜1 −
2c ⎞ , which may be positive or negative, depending on the relative magnitude of service
⎟
u ⎠
costs (c) and utility (u). When the transaction becomes volatile ( θ increases), the deviation
grows.
The expected total utility for user i is Eu MOTS ( d i ) = d i (u − c ) − P −
c(u − c )
θ.
u
(2)
The first term is the expected value from using the software; the second term is the user’s
one-time payment to the MOTS vendor. The last term represents the user’s direct loss due to
transaction uncertainty; its magnitude increases with the transaction volatility level ( θ ).
Users with non-negative utility choose to purchase the software. The marginal user is at
dM =
P
c
+ θ.
u−c u
(3)
Thus, the vendor’s market share is 1 − d M , and it obtains a profit of P − C MOTS per user. The
selling price P is chosen to maximize the vendor’s profit:
13
Max
∏ MOTS = ( P − CMOTS )(1 − d M ) .
P
C MOTS (u − c ) (u − c )c
+
−
θ.
2
2
2u
Consequently, the indifferent user has an expected transaction volume of
The optimal price is P M =
dM =
1 C MOTS cθ 1
+
+
> .
2 2(u − c ) 2u 2
(4)
(5)
The MOTS vendor leaves most users out of the market. This finding captures the fact that
small and medium firms typically have had no access to in-house enterprise software systems,
because of the high fixed costs of purchasing, installing, and maintaining such software.
In addition, equation (4) indicates that the MOTS vendor will reduce its price when
transaction volatility ( θ ) increases. Plugging this optimal price into equation (2) shows us that
the reduced price can only partially compensate for a user’s direct loss. The user still incurs a
total loss of c(u − c) θ due to uncertain transaction volumes. Moreover, the marginal user’s
2u
location ( d M ) will move toward the right as transaction volatility increases, which means that the
vendor’s market share decreases although its selling price has been reduced. Consequently, the
vendor’s profit decreases. In the pre-ASP market, then, the vendor also incurs losses due to the
demand uncertainty on the user’s side. Hence, when transaction uncertainty ( θ ) increases, every
market participant faces a worse situation.
Proposition 1
In the pre-ASP market, firms with low transaction volumes are not able to afford
the system; when the demand for software becomes more volatile, both users and the MOTS
vendor are worse off.
Regardless of the size of the users’ application, the fundamental challenge they face is always
capacity hedging. Users of MOTS software must set the service capacity in advance based on the
anticipated transaction volume; they therefore are unable to quickly scale their capacity up or
14
down according to changing business needs. We show in the next section that ASPs can better
address this capacity management problem by pooling the risks of many users.
3.2 The Dual Market with Competition
3.2.1
Equilibrium analysis
In the dual market, the ASP and the MOTS vendor simultaneously decide their prices, pa
and P . Given the prices, users decide to use one or the other or to stay out of the market.
Consider user i with expected transaction volume d i . Given ( pa , P ) , user i gains an
expected utility of Eu ASP ( d i ) = (u − pa − t )d i from the ASP, and
EuMOTS (d i ) = d i (u − c) − P −
∂EuMOTS ∂Eu ASP
c (u − c )
≥
≥ 0,
θ from the MOTS vendor. Since
∂d i
∂d i
u
in equilibrium the market is segmented in such a way that users with low transaction volumes
choose the ASP and users with high transaction volumes choose the MOTS software. The
indifferent user’s location, denoted as d * , is given by Eu ASP ( d * ) = Eu MOTS ( d * ) .
d* =
P
c(u − c )θ
.
+
p a − c + t u ( pa + t − c )
(6)
[ ]
The MOTS vendor serves users in d * , 1 , with a market share of 1 − d * , and the ASP serves
[
]
the users in 0, d * , with a market share of d * .
*
The MOTS vendor chooses a selling price P : Max ∏ MOTS = ( P − CMOTS )(1 − d ) . It is
P
easy to show that its profit function is concave in price. The first-order condition is
P=
pa + t − c CMOTS c(u − c)θ
+
−
.
2
2
2u
(7)
The ASP sets a price pa to maximize its profit: Max ∏ ASP =
pa
first-order condition is pa =
1
( pa − c )(d * )2 − Sd * . The
2
B(t + c ) + 2 Su(t − c )
, where B = Pu + c(u − c )θ .
B − 2 Su
(8)
15
It can be shown that the price equilibrium always exists.8
We are interested in exploring the properties of the price equilibrium, and in examining how
the MOTS vendor and users are affected by the ASP’s entry into the market.
Proposition 2
When the unfit cost t from using the ASP’s product is smaller than a threshold
value t * , where t * ≤
u−c
and t * is a decreasing function of the ASP’s setup costs S, the ASP’s
2
entry into the market results in a lower price, a smaller market share, and less profit for the
MOTS vendor. Meanwhile, each user is better off, and the total consumer surplus increases.
<Proof> See the Appendix.
The unfit cost t, which measures users’ disutility (per transaction) from using a lesscustomized software system, plays an important role in the competition. When t is smaller than
the given threshold value, a common competition outcome is reached: the ASP’s entry into the
market has negative effects on its competitor (i.e., the MOTS vendor) and positive effects on
users. The ASP benefits users in different ways: users with low transaction volumes, which
otherwise will be out of the market, are able to afford the software through the “pay-as-you-go”
business scheme; users with medium transaction volumes, which otherwise will install an inhouse system, opt for the ASP because the ASP represents a cheaper solution; users with high
transaction volumes stick to the in-house solution but pay less to the MOTS vendor because of
the ASP’s competition. Therefore, each user is better off, and the total consumer surplus is strictly
improved.
However, such effects appear only when t ≤ t * . As Proposition 3 shows, when t increases
and exceeds the threshold value t * , the competition exhibits totally different features.
8
To see this, note that equation (8) is the only solution to the ASP’s first-order condition. If we plug it into
the second-order condition, we get an expression that is always negative. Thus equation (8) gives us the
unique profit-maximizing price. Besides, there is no better corner solution: for very low prices, the ASP
will take over the entire market, as shown in Proposition 5, and for very high prices, the ASP’s profit
eventually falls to zero, since the ASP will be driven out of the market, as shown in Proposition 6.
16
Proposition 3
When t ≥ t * , the ASP will charge a price pa = u − t . The ASP’s entry into the
market has no effect on either the MOTS vendor or users.
<Proof> See the Appendix.
When t * < t < u − c , the entry of an ASP does not affect the MOTS vendor’s price, market
share, or profit. The MOTS vendor still behaves as if it were the only solution provider in the
market, as the ASP in fact is not competing with it. Although full market coverage is achieved,
the ASP extracts all consumer surplus, so users gain zero utility. In such cases, the ASP enters the
[
]
market, serves “residual” customers (in 0, d M ), and gains non-zero profit, but it otherwise has
no effect. If the unfit cost t increases further, t > u − c , the ASP business model is not able to
survive. The ASP’s revenue from serving residual users is not enough to offset its initial setup
costs. As a result, the ASP can not make a profit. We analyze such cases in detail in Section 3.2.2.
Next we show how the relative competitive power of the two vendors changes with the
transaction volatility level θ . Lemma 1 states that both vendors will reduce prices as θ
increases, and that the lowered prices influence their respective customers in different ways.
All inequalities hold when t < t * , and all equalities hold when t ≥ t * .
<Proof> See the Appendix.
First, we show that the ASP’s users are better off when the transaction volatility level
increases. The on-demand feature of the ASP business model allows users to enjoy full capacity
scalability and to handle possible demand fluctuation at no risk. Hence, the users of ASPs are not
affected negatively by demand volatility. Moreover, the ASP will reduce its price due to the
competition from the MOTS (i.e., dpa* dθ ≤ 0 ), which benefits the users. Second, the MOTS
vendor, when t < t * , is willing to share more uncertainty risks than it would in the pre-ASP
market; i.e., dP * dθ > dP M dθ =
c(u − c )
. In other words, the MOTS vendor’s price elasticity
2u
with regard to demand volatility increases. As a result, the users of the MOTS software are better
17
off in a relative sense: although they still incur total losses due to transaction volatility (implied
by dP * dθ <
c(u − c )
), they are better off because they now bear less uncertainty risk. The ASP
u
not only removes its own users’ uncertainty risks by doing business “on-demand,” but also helps
to reduce the other users’ uncertainty risks through competition. Finally, we find that the
indifferent customer’s position ( d * ) will shift toward the right when θ increases. When
transaction volatility is high, more users tend to choose the ASP.
We reach the following conclusion:
Proposition 4
When the ASP exists, all users are able to abate the risks brought on by uncertain
transaction volumes. When transaction volumes become more volatile, the ASP gains more
market share.
3.2.2 One business model dominates
Here, we investigate whether a single software business model, ASP or MOTS, can dominate
the whole market. We provide answers for questions such as: Is it possible that on-demand
software providers will completely replace MOTS vendors and become the only delivery
channel? Or will ASPs finally fall out of the market and become only a historical footnote?
Propositions 5 and 6 state our findings.
Proposition 5
t1 =
There is a threshold value for the unfit cost parameter t,
c(u − c)θ C MOTS
+
− S . When t ≤ t1 , the ASP serves the whole market using the price
2u
2
pa =
C MOTS u + c (u − c )θ
+c+S .
2u
<Proof> See the Appendix.
When t ≤ t1 , the ASP drives the MOTS vendor out of the market. In this case, all users will
access the software through the ASP. Since S << C MOTS , t1 > 0 always. Hence, in the extreme
case of t=0, the ASP is always able to dominate the whole market.
18
On the other hand, the MOTS vendor can block the ASP’s entry into the market when t is
large enough, as stated in the following proposition.
Proposition 6
t2 = u − c −
There is a threshold value for the unfit cost parameter t,
4u(u − c ) S
. When t ≥ t2 , the ASP is not able to survive.
C MOTS u + (u + cθ )(u − c )
<Proof> See the Appendix.
When t ≥ t2 , at any price the ASP might offer, one of the two constraints—(1) users obtain
nonnegative utility and (2) the ASP obtains a nonnegative profit—must be violated. The ASP
business model therefore is not viable. In this case, only the MOTS solution survives.
Furthermore, we can show that both t1 and t2 increase in θ . As the transaction volatility
increases, it becomes easier for the ASP to dominate the market, and more difficult for the MOTS
vendor to do so.
3.3 The Key Determinant(s) of an ASP’s Competitive Ability
This subsection summarizes and provides a complete view of the ASP-MOTS competition.
t1 ≤ t * ≤ t2 .
Lemma 2
Lemma 2 shows the order of critical values of t. The proof of Lemma 2 is straightforward.
We therefore have four competition regimes, as shown in Figure 4. Our model predicts that in
each regime the price competition will occur as follows.
Figure 4. A Complete Diagram of the Price Competition
1
2
t1
3
t*
4
unfit cost t
t2
Regime 1. t ∈ [0, t1 ] : The ASP serves the whole market. All software is delivered by the ASP.
Regime 2. t ∈ [t1 , t * ] : The ASP and MOTS business models coexist in the software market, and
the competition between the two vendors benefits each user in the market.
19
Regime 3. t ∈ [t * , t2 ] : The ASP and MOTS business models coexist in the software market. The
ASP is not competing with the MOTS but just serves “residual” users. There is no competition,
and the MOTS vendor still acts as a monopolist.
Regime 4. t > u − c : The ASP is not able to survive. The MOTS model is the only software
solution.
In Regimes 1, 2, and 3, full market coverage is achieved, while in Regime 4 (when the MOTS
solution dominates), some users remain out of the market.
In different regimes of t, the competition exhibits different features. When t is low, the ASP
solution dominates, revealing the possibility of “all software in ASPs.” When t is moderate, the
ASP shares the market and competes with the MOTS vendor. When t grows further, the ASP
loses its competitive advantages and even may not be able to survive. Thus, we conclude:
Proposition 7
As users’ unfit costs from using a not fully customized system decrease, the ASP’s
ability to compete increases monotonically.
It is an interesting but not straightforward conclusion. To some extent, our model resembles
the duopoly competition model with vertically differentiated products, with the ASP as the lowquality provider and the MOTS as the high-quality provider. The unfit cost t measures the quality
difference between the MOTS and ASP products. In the vertical differentiation model, when the
quality difference decreases, the low-quality provider initially gets better off because its product
becomes more attractive. But it gets worse off as the quality difference decreases further, because
the competition becomes more intensive and eventually hurts both vendors. However, in our
paper, we find that reducing the value of t benefits the low-quality provider (i.e., the ASP)
monotonically. Moreover, when the two vendors’ products are close enough, the high-quality
provider (i.e., the MOTS vendor) could be squeezed out of the market. Our findings deviate from
traditional vertical differentiation literature because the ASP and the MOTS vendors use different
pricing strategies. What we find here is more like what Varian (2000) showed in a monopoly
20
setting: when the quality difference between selling and renting a product is small enough, the
renting strategy dominates. We show that a similar conclusion holds in duopoly competition.
This result is very important and provides useful managerial implications. It tells us that
ASPs must offer software applications with a lower value of t in order to survive and succeed in
the competitive marketplace. Therefore, not every software application can be delivered through
an ASP. The on-demand business model is not simply a new distribution channel for existing
software; instead, it requires a fundamentally new set of products and technology architectures.
Applications that are suited to the ASP channel are those most compatible with other systems and
programs. On the vendor’s side, ASPs should write their applications using an open language (for
example, XML), in a proper modular structure, and with a loosely coupled interface with other
applications (Dzubeck 2004; Thibodeau 2004). All these strategies would help to reduce users’
unfit costs. In addition, industry-wide adoption of software standards and protocols, once
achieved, would reduce the unfit costs across different applications and therefore affect ASPs
positively. Some efforts already have been made in this direction. Microsoft, Oracle, and Sun
have joined forces with IBM and Hewlett-Packard to facilitate technical standards to govern how
commercial software should be written. The ASP Industry Consortium (ASPIC) and the
Information Technology Association of America (ITAA) are working closely to make ASPs more
standardized. These strategic moves should benefit ASPs’ competitiveness in the future.
Another factor that affects the performance of ASPs is the volatility of demand, θ . A highly
*
∂t
volatile transaction volume enhances the ASP’s competitive advantages: ∂t > 0 , 1 > 0 , and
∂θ
∂θ
∂t2
> 0 . As we have shown above, the ASP’s entry into the market not only allows its own
∂θ
clients to “go on demand” and therefore shift away their uncertainty risks, but it also helps users
of the MOTS software to abate their losses caused by demand uncertainty. The ASP business
arrangement is appreciated by users to a large extent because it offers highly scalable systems.
21
Users can promptly scale their capacity up or down at no additional cost, because there is no fixed
IT investment made on the users’ side.
Moreover, when it becomes expensive to provide on-site IT support services (a large value of
c), user firms will find the ASP solution more attractive. As is shown by
C MOTS
θ [(u − 2c )( pa + t − c ) + c(u − c )]
∂d *
=
+
> 0 , when users face a large c, they prefer to
∂c
2( p a + t − c ) 2
2u ( p a + t − c ) 2
get their IT support services from an outside provider, rather than install a system and maintain it
internally. In such a situation, the ASP is expected to take a significant market share.
3.4 Robustness Check
Although we make some simplifying assumptions to reduce the complexity of the analyses,
our results are robust to these assumptions. For example, we assume that the MOTS software is
highly customized, while the ASP tends to come up with much higher unfit costs because of its
one-to-many structure. Such an assumption does not imply that the MOTS product must be a
perfectly fit system. In practice, an in-house software system could also impose lack-of-fit costs
on users. All that matters for modeling purpose is the marginal disutility for users between the
MOTS solution and the ASP. In fact, our model addresses the case in which the MOTS software
has unusually high unfit costs, i.e., t < 0 . This leads to the trivial outcome that the ASP
dominates the marketplace.
In addition, we assume that both vendors use basic pricing strategies: the ASP charges a fee
per transaction, and the MOTS vendor requires an outright purchase of the system. Admittedly,
there are numerous combinations of pricing strategies in the competitive environment.
Practically, nonlinear pricing can be considered.9 Since our objective is to characterize the nature
of the competition between the ASP and the MOTS vendor, we focus on the most salient
9
An analysis of the “fixed fee + variable charge” by the ASP can be found in the paper by Cheng and
Koehler (2002).
22
differences between the two delivery channels. Assuming complex pricing strategies will not
discredit our findings here.
Our main results can also be extended to the scenario in which the user firm’s transaction
volatility is proportional to its expected transaction volume. For instance, we may assume that the
actual transaction volume of a user located at d i is uniformly distributed on [(1 − θ )d i , (1 + θ )d i ] ,
with θ < 1 . In this case, most of our lemmas and propositions still hold.10
4. Two-Stage Competition with Uncertainty
So far, we have analyzed a market with static unfit costs (t). In practice, however, these costs
could change during an ASP’s contract period, which typically lasts for 3 to 5 years. Unfit costs
could grow over time given software or hardware changes on the users’ side, or decrease over
time due to improvements in the ASP’s software and related technology. For example, if the ASP
uses a browser interface that is dependent on nonstandard aspects of IE6 but business
circumstances faced by users drive a demand for the latest IE or Firefox, or if the ASP’s interface
involves a module built on top of a program that only works in a pre-Vista MS Windows
environment but hardware replacement at the user’s site leads to multiple PCs with the Vista OS,
unfit costs may increase. In such cases, an existing ASP user faces higher unfit costs t in its later
usage. On the other hand, if the ASP vendor continuously invests in improving its system
integration features, users’ unfit costs may be decreasing over time. For instance, Salesforce.com
developed and launched AppExchange in January 2006. AppExchange is an online marketplace
for on-demand business software. Currently it includes over 150 applications, and Adobe, Skype,
and Factiva are among the various partners. AppExchange allows Salesforce.com and other ASPs
to establish across-application integration and therefore provides software users seamless
extension of their existing systems (Cowley 2005; Kuchinskas 2006). In this case, users expect to
have reduced unfit costs because a uniform platform eases collaboration across applications.
10
The only exception is Lemma 1, which can be proven under the special case with negligible setup costs,
but for the more general cases, it gets too complicated mathematically.
23
To capture how competition takes place in a longer window with possible changes in unfit
costs over time, we build a two-stage competition model. In the first stage, the vendors choose
their prices ( pa , P ) simultaneously. We assume that their prices remain unchanged over time.
The ASP’s application imposes unfit costs t1 in the first stage. Users could have certain
expectations about a future change in unfit costs. They may expect unfit costs to increase if they
anticipate changes in demand or hardware upgrades, or to decrease if they anticipate
technological advances that favor the shared software business model. In the second stage, such a
change is realized. Users consider switching from the ASP’s software to the MOTS software or
the reverse.
We make two simplifying assumptions. First, users and vendors weight utilities or profits
obtained from both stages equally. The time discount factor for stage two is 1. Second, the ASP’s
initial setup costs to serve a new client are negligible; i.e., S = 0 . These two assumptions help to
ease the analytical analysis without changing the results qualitatively. Our study on the two-stage
model will only focus on the scenario with both vendors coexisting. Scenarios in which one
vendor fails are of little interest in this context.
4.1 The Two-Stage Model with Increased Unfit Costs
Figure 5a shows the two-stage competition with an (expected) increased t. In the first stage, with
unfit costs t1 , users in [0, d1 ] choose the ASP and in [d1 ,1] use the MOTS software. In the
second stage, users face changing business circumstances, and the unfit costs of using a standard
ASP application increase from the initial t1 to t H > t1 . Existing ASP users may consider
switching to the MOTS solution, which provides customized software with a better fit.
An existing ASP user i compares the utility from the ASP, (u − pa − t H )d i , and the utility
from the MOTS software, (u − c)d i − P − c(u − c)θ . The “marginal” switcher, denoted by d S , is
u
given by
dS =
P
c (u − c )θ .
+
pa + t H − c u ( pa + t H − c )
(9)
24
Figure 5a. Competition with Increased Unfit Costs
dS
Expected transaction
volume
d1
0
1
Users choose the ASP
initially and stay with
the ASP afterward.
Users choose the ASP
initially and switch to the
MOTS software in the second
stage.
Users choose the MOTS vendor
initially and stay with the MOTS
software afterward.
As shown in Figure 5a, the user d1 , which is indifferent between the two solutions in the first
stage, gains the same total utility. If it chooses the ASP and switches to the MOTS software later
on, its total utility is {(u − p a − t1 )d1 } + ⎧⎨(u − c)d1 − P − c(u − c)θ ⎫⎬ . If it chooses the MOTS
u
⎩
⎭
(
)
θ
c
u
−
c
c(u − c)θ ⎫ . Note that
⎧
⎫
⎧
software initially, its total utility is
⎨(u − c)d1 − P −
⎩
u
⎬ + ⎨(u − c)d1 −
⎭ ⎩
u
⎬
⎭
once the user installs the MOTS system in the first stage, it owns the system and only needs to
pay ongoing maintenance costs in the later stage. By equating these two utilities, we get
d1 =
c(u − c)θ
.
u ( p a + t1 − c)
(10)
Equations (9) and (10) imply that the MOTS vendor’s price P only affects d S , not d1 .
When the MOTS vendor’s price increases, all else being equal, the number of users that choose
the in-house system in the first stage remains unchanged. This is because the price P is
cancelled out in the indifferent user d1 ’s utility function (equation (10)). In the second stage,
however, the number of switchers (from the ASP to the MOTS solution) decreases as P
increases. A high MOTS software price, representing the large up-front costs of an internal
system, reduces existing ASP users’ incentive to move the software in-house.
The MOTS vendor gets [d1 ,1] users in the first stage and [d S , d1 ] users in the second stage.
Its profit comes from users’ one-time payment. The optimal price is given by
Max ∏ MOTS = ( P − CMOTS )[1 − d S ] .
(11)
P
The ASP serves [0, d1 ] users in the first stage and [0, d S ] users in the second stage. It gains
profit from users’ every use of the software. The optimal price is given by
25
dS
⎡ d1
⎤ 1
Max ∏ ASP = ( pa − c )⎢ ∫ xdx + ∫ xdx ⎥ = ( pa − c )(d12 + d S2 ) .
pa
0
⎣⎢ 0
⎦⎥ 2
(12)
If users do not anticipate future changes in unfit costs, this reduces to a one-stage static game
with invariant unfit costs t1 , which we have analyzed in Section 3. Here, we take such a case as
the benchmark to investigate how users’ expectation that unfit costs will increase affects the
competition equilibrium outcome.
Let pa
t1 ,t H
and P t ,t be the prices of the ASP and the MOTS vendor when users expect
1 H
unfit costs to increase from t1 to t H , and let d1 t ,t be the indifferent user defined by equation
1 H
(10). Let pa
t1
and P t be the prices when users believe unfit costs t1 will not vary, and let d t*1
1
be the indifferent user defined by equation (6) in that instance.
Proposition 8
When users anticipate a future increase in unfit costs, t1 → t H , both vendors will
increase their prices; i.e., pa t ,t > pa t and P t ,t > P t . More users will choose the MOTS
1 H
1
1 H
1
software initially (i.e., d1 t ,t < d t* ), and the ASP will lose existing clients to the MOTS vendor
1 H
1
once the increase occurs.
<Proof> See the Appendix.
Proposition 8 states three important findings. First, although increased unfit costs imply a
decrease in product quality of u − t , the ASP should nevertheless increase its price:
pa t ,t > pa t . By charging a high price, the ASP gives up competing for high-volume users with
1 H
1
the MOTS vendor; it instead concentrates on exploiting low-volume users that are unable to
afford the MOTS solution anyway.11 Second, the MOTS vendor also raises its price, which is
intuitive because its product becomes more attractive. Interestingly, we find that the MOTS
The ASP’s price cannot exceed the upper bound of u − t H . Otherwise low-end users have no incentive
to use the ASP’s product, and they will opt to stay out of the market.
11
26
seller’s best response function ( P = pa + t H − c + C MOTS − c(u − c)θ , see the proof of Proposition 8
2
2
2u
in the Appendix) is the same as that in a static competition with unfit costs t = t H (equation (7)).
This means that the MOTS vendor in practice will adopt a simple pricing strategy and price its
software as if it were in a one-stage competition with invariant unfit costs. Finally, we conclude
that a belief that unfit costs will increase benefits in-house solution providers but hurts ASPs.
4.2 The Two-Stage Model with Decreased Unfit Costs
Figure 5b shows the two-stage competition when t decreases. In the first stage, with unfit
costs t1 , users in [0, d1 ] choose the ASP and in [d1 ,1] buy from the MOTS vendor. In the second
stage, the unfit costs of using the ASP application decline to t L < t1 . Such a change could be the
result of advances in web technology, adoption of software standards and protocols, or the ASP’s
efforts to create a uniform software platform.
Figure 5b. Competition with Decreased Unfit Costs
d1
Expected transaction
volume
dS
0
1
Users choose the ASP
initially and stay with
the ASP afterward.
Users choose the MOTS
software initially and switch
to the ASP in the second
stage.
Users choose the MOTS
software initially and stay with
it afterward.
When unfit costs decrease in the second stage, an existing MOTS software user i compares
the utility available from the ASP product, (u − pa − t L )d i , and the utility from the MOTS
software, (u − c)d i − c(u − c)θ . Note that because the in-house system has already been installed,
u
the one-time payment to the vendor P has been sunk. This user only needs to pay ongoing
maintenance costs. The “marginal” switcher, denoted by d S , is given by
dS =
c (u − c)θ .
u ( p a + t L − c)
(13)
27
As Figure 5b shows, the user d1 is indifferent between the two solutions at the first stage. If
it chooses the MOTS and then switches to the ASP later on, its total utility is
c(u − c)θ ⎫
⎧
⎨(u − c)d1 − P −
⎬ + {(u − pa − t L )d1 } . If it chooses the ASP initially, its total utility is
u
⎩
⎭
{(u − pa − t1 )d1 }+ {(u − pa − t L )d1 }. Hence, we get d1 =
P
c(u − c)θ .
+
p a + t1 − c u ( p a + t1 − c)
(14)
Equations (13) and (14) show that the MOTS price P only affects d1 but not d S . When the
MOTS price increases, all else equal, the number of users that choose the in-house system in the
first stage decreases. However, the number of switchers (from the MOTS to ASP, in the second
stage) is not affected by the price increase. A high price for the MOTS software does not give
existing users a greater incentive to switch to on-demand software because it is considered a sunk
cost.
The MOTS vendor and the ASP choose the following profit-maximizing prices, respectively:
Max ∏ MOTS = ( P − C MOTS )[1 − d1 ] ;
(15)
P
dS
⎡ d1
⎤ 1
Max ∏ ASP = ( pa − c )⎢ ∫ xdx + ∫ xdx ⎥ = ( pa − c )(d12 + d S2 ) .
pa
0
⎣⎢ 0
⎦⎥ 2
(16)
We take the static competition with invariant unfit costs t1 as the benchmark case. Let
pa
t1 ,t L
and P t ,t be the prices of the ASP and MOTS products when users expect unfit costs to
1 L
decrease from t1 to t L , and let d1 t ,t be the indifferent user defined by equation (14). Let pa
1 L
t1
and P t be the prices when users believe unfit costs t1 will not vary, and let d t*1 be the
1
indifferent user defined by equation (6).
Proposition 9
When users anticipate a decrease in unfit costs, t1 → t L , both vendors will
reduce their price; i.e., p a
t1 ,t L
< p a t , P t ,t < P t . More users will choose the ASP solution
1
1 L
1
initially; i.e., d1 t ,t > d t* . Existing clients of the MOTS vendor have little incentive to switch to
1
L
1
28
on-demand software even if unfit costs decrease, but they may do so if transaction volatility is
high.
<Proof> See the Appendix.
Proposition 9 states three important findings. First, the ASP’s response to users’
anticipation of decreased unfit costs is to reduce its price, despite an increase in the quality of its
product, in order to compete for the more profitable high-volume users.12 Second, the MOTS
vendor once again sticks with a simple pricing strategy. Its best response function
(P =
pa + t1 − c C MOTS c(u − c)θ
, see the proof of Proposition 9 in the Appendix) is the same
+
−
2
2
2u
as that in a static competition with unfit costs t = t1 (equation (7)). The MOTS vendor therefore
can just ignore the expected change in unfit costs on the users’ side and price the software as if it
were in a one-stage competition. Third, our analysis shows that existing users of MOTS software
are unlikely to move to on-demand software. These users have two choices: stay with the MOTS
solution, with a utility of (u − c)d i − c(u − c)θ , or switch, with a utility of (u − pa − t L )d i . Since
u
pa > c always, the user switches only if its transaction volatility ( θ ) is high. Hence, we
conclude that once an in-house system has been installed, users have little incentive to switch to
an ASP unless they need to manage high capacity risks.
5
Discussion and Conclusion
We examine firms’ choices between the ASP and MOTS solutions when they face stochastic
demand for the software. Our findings show that the ASP on-demand model is superior when the
user faces a low transaction volume, high transaction volatility, or expensive in-house IT services,
or when the user’s unfit costs from using the ASP’s standardized product are low. Under broad
market conditions, the ASP and MOTS business models will coexist, so it is likely that the
12
Since the ASP gets paid based per transaction, users with high transaction volumes are
considered more profitable.
29
software industry will not be dominated by just one. The ASP’s entry, at the expense of the
traditional MOTS vendor, may benefit every user firm. It offers small and medium firms costsaving access to software, competitively reduces large firms’ implementation costs, and mitigates
each firm’s transaction uncertainty risks. Although the ASP model has several advantages, its
long-run viability largely depends on the magnitude of users’ unfit costs. As such costs increase,
the ASP’s ability to compete with the MOTS vendor monotonically decreases. This important
finding implies that the ASP model requires a fundamentally new set of products and technology
architectures. To achieve profits, the new generation of ASPs should leverage an open language,
a proper modular structure, and a loosely coupled interface with other applications, and thus offer
products with low unfit costs. Today’s ASPs are developing and delivering their own applications
using pure Web architectures, in a standard XML language, and in discrete units, which will
greatly increase their chance of success. Our work provides a theoretical base for these strategic
moves.
ASPs are actively investing in reducing users’ unfit costs. It is interesting and necessary to
analyze the price competition over a longer horizon in which users’ unfit costs may change over
time. Our findings suggest that investments in improving ASP software quality should be
accompanied with a downward price adjustment. In the future, we expect to see that ASPs offer
better products at lower prices. On the other hand, MOTS vendors should always adopt a simple
pricing strategy: they should price their software as if they were involved in a static competition
with invariant unfit costs. Hence, we predict that no price change will be observed for traditional
packaged software. In addition, our analysis shows that existing MOTS software users switch to
the ASP solution only to eliminate capacity, or business volume risks rather than to reduce
operating costs. These results have direct implications for the kind of businesses that ASP will
target in the future.
There are many possibilities for further ASP studies. It would be interesting to examine
the role of the Service Level Agreement (SLA). When users have demand for different levels of
30
IT service, the SLA constitutes a way to segment the market and improve the ASP’s profit, and
maybe to increase users’ surpluses as well. Another possible research topic is the software
vendor’s choice. Some major software companies have changed their business models to operate
as an ASP as well. For example, Oracle now has more than 400 customers using its ASP
products and a growing share of its revenues from its ASP unit. It is interesting to see that many
software providers are doing business both ways: selling a sophisticated version as a MOTS
product and leasing a simplified version as an ASP. Future research may investigate whether a
software vendor should add the ASP as an additional channel, switch to it exclusively, or eschew
that approach entirely.
31
References
Bednarz, A., “Manufacturers Eye on On-Demand Software,” Network World, Apr 24, 2006
Cheng, H.K. and Koehler, G.J., “Optimal Pricing Policies of Web-enabled Application Services,”
working paper, 2002
Choudhary, V., Tomak, K., and Chaturvedi, A., “Economics Benefits of Software Renting,”
Journal of Organizational Computing and Electronic Commerce 8, 277-305, 1998
Cowley, S., “Salesforce.com Makes Platform Move with AppExchange,” InfoWorld, Sep 2005
Dewan, S., and Mendelson, H., “User Delay Costs and Internal Pricing for a Service Facility,”
Management Science, Vol. 36, No. 12, Dec. 1990
Dunn, D., “Systems Management on Call,” Information Week, Jan 10, 2005
Dzubeck, F., “XML and Management Team up,” Network World, Vol. 21, Iss. 27, Jul 2004
Fishburn, P., Odlyzko, A., and Siders, R., “Fixed Fee versus Unit Pricing for Information Goods:
Competition, Equilibria, and Price Wars,” First Monday 2(7), 1997
Garner, R., “Software for Hire,” CRN, Nov 1, 2004
Hamm, S., “SAP Gets On-Demand Religion,” Business Week, Feb 2, 2006
Kaplan, J., “Sorting Through Software As A Service,” Network World, Nov 21, 2005
Knorr, E., “Software on Demand: the End of IT as We Know It?” InfoWorld, Vol. 26, Iss. 48,
Nov 29, 2004
Kuchinskas, S., “Salesforce Finally Ships AppExchange,” Ecommerce, Jan 17, 2006
Lacy, S., “The On-Demand Software Scrum,” Business Week, April 17, 2006
Ma, D., private communications with the representatives of MedPACS (phone: 262-367-0181),
2005
Martin, C., “Fee-per-Study Approach Can Cut Start-up Costs,” Radiology Today, Vol.5, No. 19,
Sep 2004
Murphy, D., “As Easy as ASP,” Direct Response. London, Jan 2005
Niccolai, J., “Gates Memo Puts Services at the Heart of Microsoft,” Network World, Nov 2005
32
“Proponents of Software as a service say it will wipe out traditional software,” The Economist,
Apr 21, 2006, article available at
http://globaltechforum.eiu.com/index.asp?layout=rich_story&doc_id=8439&title=Universal+serv
ice%3&categoryid=4&channelid=3
Pallatto, J., “IBM Recruiting ISVs, Partners to SAAS,” Channel Insider, Feb 23, 2006
Ricadela, A., “Does Salesforce.com Have the Chops to Become a Billion-Dollar Company?”
Information Week, May 2006
Singh, C., Shelor, R., Jiang, J., and Klein, G., “Rental Software Valuation in IT Investment
Decisions,” Decision Support Systems, Vol. 38, Iss. 1, Oct 2004
Sundararajan, A., “Nonlinear Pricing of Information Goods,” Management Science 50 (12), 2004
Susarla, A., Barua, A., and Whinston, A., “Understanding the Service Component of Application
Service Provision: An Empirical Analysis of Satisfaction with ASP Services,” MIS Quarterly
27(1), 2003
Thibodeau, P., “Tech Standards Set Tone for On-Demand Systems,” Computer World, Vol. 38,
Iss. 44, Nov 1, 2004
Vara, V., “Web Services Face Reliability Challenges,” Wall Street Journal, Feb 23, 2006
Varian, H., “Buying, Selling, and Renting Information Goods,” Journal of Industrial Economics
XLVIII, 473-488, 2000
Zimmerman, J., “The Costs and Benefits of Cost Allocations,” The Accounting Review, Vol. 54,
No. 3., pp. 504-521, Jul., 1979
33
Appendix
Proofs for Lemmas and Propositions
Prove Proposition 1: “users with low transaction volumes ( d i < θ ) are not served by the
MOTS vendor.”
We prove by contradiction. Suppose that the MOTS vendor does serve some users with low
transaction volumes. Let d h , d l be the marginal users in [ θ , 1] and [0, θ ] respectively. First, we
examine the extreme case that users in [0, θ ] face a deterministic transaction volume. The
vendor’s profit-maximization problem is Max ( P − C MOTS )(1 − d h + θ − d l ) , where
P
dl =
cθ
θ cθ
P
1 C
and d h = d l +
. Solving it gives d l0 = + MOTS + −
> θ , which
(u − c )
u
4 2(u − c ) 4 4u
leads to a contradiction. Next, we look at the general distribution case. For users in [0, θ ], we
make no assumption on the distribution of their transaction volumes. Denote a user’s utility loss
due to transaction uncertainty by f (δ ) . It is easy to get P = (u − c )d l − f (δ ) and
d h= d l +
cθ
f (δ )
. The software vendor’s problem is
−
u (u − c )
⎛
cθ
f (δ ) ⎞ . The profit function is concave, and the
Max ((u − c)d l − CMOTS − f (δ ) )⎜⎜1 − 2d l −
+θ +
⎟
dl
u
(
u
− c) ⎟⎠
⎝
first-order condition is ∂ = ⎧⎨(u − c ) − (u − c )cθ − 4(u − c )d l + (u − c )θ + 2C MOTS ⎫⎬ + 3 f (δ ) . The
∂d l ⎩
u
⎭
terms in the parenthesis are just the first-order condition when the transaction volume is
deterministic. Hence, this first-order condition is positive at d l0 . Making use of the concavity, we
conclude that the optimal solution d l* > d l0 > θ , which is again a contradiction.
(
Q.E.D.
)
<Proof of Proposition 2> Denote by P * , pa* the price pair at the intersection point of equations
(7) and (8). Consider the boundary conditions for the vendors’ prices: the MOTS vendor’s price
has a lower bound P ≥ C MOTS , and the ASP’s price has an upper bound pa* ≤ u − t . The ASP
34
must offer a price that gives users enough incentive to use it. A price larger than the upper bound
u − t will attract no users. On the other hand, the ASP’s price must be high enough to cover the
(
ASP’s initial setup cost to give the ASP an incentive to serve the market.i The price pair P * , pa*
)
is an equilibrium outcome if it satisfies all these boundary conditions. In the special case with
S=0, we get (PS*=0 , p a*,S =0 ) = ⎛⎜ max ⎧⎨C MOTS , C MOTS + t − c(u − c )θ ⎫⎬, t + c ⎞⎟ , which is a feasible
2
2u
⎭
⎩
⎝
⎠
outcome if t + c ≤ u − t ⇔ t ≤
∂pa*
u−c *
> 0 . In the general case
≡ ts = 0 . It is easy to show that
∂S
2
with S ≥ 0 , the ASP’s price is pa* = c + t + ∆ ( S ) , where ∆( S ) ≥ 0 , ∂∆( S ) > 0 . Thus, the critical
∂S
value t * must satisfy t * = u − c − ∆( S ) ≤ u − c , and t * is decreasing in S .
2
2
Compare the MOTS vendor’s best response function in the dual market (equation (7)) with its
optimal price P M in the pre-ASP market (equation (4)). Since pa* < u − t , the vendor’s price
decreases after the ASP enters the market: P * < P M .
Plugging equation (7) into equation (6) gives the location of the indifferent user in the dual
market:
d* =
1
C MOTS
cθ ( u − c )
.
+
+
2 2 ( p a + t − c ) 2u ( p a + t − c )
(A1)
Compare this with the marginal user’s location in the pre-ASP market ( d M , equation (5)).
We find that d * > d M : the MOTS vendor’s market share decreases. Together, a lower price and
a decreased market share result in less profit for the MOTS vendor.
Figures A1(a) and (b) show the pre-ASP and dual markets respectively. When the ASP is an
option, full market coverage is achieved. Small users with low transaction volumes, located in
[0, d ] , are able to afford the software through the ASP, because the ASP does not impose high
M
[
]
implementation and maintenance costs on them. Users in d M , d * will purchase the MOTS
35
software in the pre-ASP market but choose to use the ASP in the dual market. For these users, the
ASP represents a cheaper solution than the MOTS option. Since
[
] [
] [
]
E u MOTS ( P = P M ) ≤ E u MOTS ( P = P * ) ≤ E u ASP ( pa = pa* ) , they are better off as well. Finally,
[
]
users in d * ,1 are firms with a high volume of transactions. They choose the MOTS solution in
both markets, but their payments are lower in the dual market because of the ASP’s competition.
The ASP’s entry into the market benefits these firms as well, although they are not its clients.
Each user therefore is better off, and the total consumer surplus is strictly improved.
Figure A1. The Pre-ASP Market and the Dual Markets When t ≤ t *
(a) The pre-ASP market (the MOTS software solution alone)
users purchase the MOTS software at P M
1
0
2
1
dM
users are priced out of the market
(b) The dual market
users purchase the MOTS
software at P * < P M
0
users choose the ASP
d
M
d*
1
Q.E.D.
<Proof of Proposition 3> When t > t * , the ASP’s unrestricted optimal price (from equation (8))
is pa > u − t . This price is too high to attract any user. The ASP will set its price at the restricted
*
optimal pa'' = u − t . The ASP’s users therefore get zero surplus. Plug pa'' into equation (7). The
MOTS vendor’s optimal selling price, denoted by P " , is the same as in the pre-ASP market
( P M , given by equation (4)). Plugging pa'' and P " into equation (6) gives the MOTS vendor’s
market share, 1 − d " =
1 C MOTS cθ
, which again is the same as in the pre-ASP market
−
−
2 (u − c ) 2u
36
( 1 − d M , given by equation (5)). The conclusion that the MOTS vendor’s profit does not change
follows.
When the value of t increases further, and especially when t > u − c , it is easy to see that the
ASP is not able to operate profitably.
Q.E.D.
<Proof of Lemma 1> Using equation (8), we get dp a* dθ =
dB
1
[ −4 Sut ] , where
dθ ( B − 2 Su ) 2
dB dP *
=
u + c (u − c ) .
dθ
dθ
To judge the sign of dpa* dθ , consider two possible cases: (1)
and (2)
dB
< 0 <=> dpa* dθ > 0 ,
dθ
dP *
c(u − c )
dB
.
<−
> 0 <=> dp a* dθ < 0 . If the first case is true, dB < 0 implies
dθ
u
dθ
dθ
However, equation (7) indicates that
dP * 1 dp a* c(u − c )
c(u − c )
. Hence, we eliminate
=
−
>−
dθ
2 dθ
2u
2u
dP *
the possibility of case (1). Consider the second case. If dB > 0 , either
> 0 or
dθ
dθ
−
dpa*
c(u − c ) dP *
< 0 , which means that
<
< 0 . On the other hand, equation (8) implies
dθ
u
dθ
*
dP *
c(u − c ) . It therefore must be the case that dpa
dP *
< 0,
<−
< 0 , and
dθ
dθ
2u
dθ
c(u − c )
c(u − c )
. This proves parts (1) and (2). Use equation (A1):
< dP * dθ <
2u
u
*
dp a* 2uc (u − c )⎛⎜ p * + t − c − dp a θ ⎞⎟
− 2C MOTS
⎜ a
dθ ⎟⎠
dd *
⎝
dθ +
> 0 . This proves part (3).
=
4( p a* + t − c )
4u 2 ( p a* + t − c ) 2
dθ
Q.E.D.
<Proof of Proposition 5> First we consider the case of t = t1 . We can show that the price pair
( P, pa ) = (C MOTS , pa ) is the equilibrium. Note that with this price pair, the whole market is
37
served by the ASP alone: d * ( P = C MOTS , pa = pa ) = 1 . Given the MOTS vendor’s price C MOTS ,
the ASP has no incentive to deviate, because the price pa is obtained using the ASP’s best
response function (equation (8)). This price pa can give the ASP positive profit, because it is
high enough to cover the ASP’s setup costs.ii On the other hand, given the ASP’s price pa , the
MOTS vendor has no incentive to deviate as well: a price increase will not attract any user, and a
price decrease will not bring positive profit. Thus we conclude that the price pair
( P, pa ) = (C MOTS , pa ) is the Nash equilibrium, and the ASP uses the price pa to serve the
whole market when t = t1 . For cases with t < t1 , a similar analysis applies. Since decreasing its
price will not bring the ASP more users, the ASP will keep its price at pa .
Q.E.D.
<Proof of Proposition 6> As shown in the proof of Proposition 3, when t increases, the ASP’s
competition becomes infra-marginal. The MOTS vendor will set the price at the monopoly price
(the same as in the pre-ASP market): P =
C MOTS (u − c ) (u − c )c
+
−
θ . The ASP’s best response
2
2
2u
price is bounded by the upper bound u-t. In this case, the MOTS vendor will take the users
⎡1
C
cθ
⎤
located in ⎢ + MOTS +
, 1⎥ . When t ≥ t2 , it is not profitable for the ASP to take the
⎣ 2 2(u − c ) 2u ⎦
2
residual users, since Π ASP
⎛1 C
1⎛1 C
cθ ⎞
cθ ⎞
= ⎜⎜ + MOTS +
⎟⎟ S ≤ 0 ,
⎟⎟ ( pa − c ) − ⎜⎜ + MOTS +
2 ⎝ 2 2(u − c ) 2u ⎠
⎝ 2 2(u − c ) 2u ⎠
∀p a ≤ u − t .
Q.E.D.
<Proof of Proposition 8> The optimization problems for the MOTS vendor and the ASP are
given by equations (11) and (12).
The best response functions (BRFs) are P = pa + t H − c + C MOTS − c(u − c)θ (for the MOTS
2
2
2u
c 2 (u − c) 2 θ 2 c + t1 − p a
+ (t H + c − p a ) = 0 (for the ASP). The proof of
vendor) and
2u 2
( p a + t1 − c) 3
38
Proposition 2 shows that in a one-stage static competition with invariant unfit costs t1 and
negligible ASP setup costs ( S = 0) , the ASP’s price is p a
t1
= c + t1 . It is straightforward to
check the ASP’s BRF and conclude that any price smaller than c + t1 is not optimal. In fact, we
can show that the optimal price must be in the range of c + t1 < p a
t 1 ,t H
< c + tH .
Since the ASP’s price increases, the MOTS vendor’s price will increase as well (using the
MOTS vendor’s BRF); i.e., P t
d1 t
,t
1 H
1
,t H
. It is easy to show that d1
> P t . Plugging the two prices into equation (10) gives
1
t 1 ,t H
< d t*1 , where d t*1 is given by equation (A1).
Q.E.D.
<Proof of Proposition 9> The optimization problems for the MOTS vendor and the ASP are
given by (15) and (16). The best response functions (BRFs) are P = p a + t1 − c + C MOTS − c(u − c)θ
2
(for the MOTS vendor) and
2
2u
c + t L − pa
c (u − c) θ
+ (t1 + c − p a ) = 0 (for the ASP).
2
2u
( p a + t L − c) 3
2
2
2
Check the ASP’s BRF. It is straightforward to see that the optimal price must satisfy
c + t L < p a t ,t < c + t1 = p a t . Since the ASP’s price decreases, the MOTS vendor’s price will
1
L
1
also decrease (using the MOTS vendor’s BRF); i.e., P t
(14) and compare to equation (A1); the conclusion d1
1
,t L
t 1 ,t L
< P t . Plug these prices into equation
1
> d follows.
*
t1
Q.E.D.
i
This boundary condition may be reached when the unfit cost t is very small. As t decreases, the price from
equation (8) also decreases. It may reach a level that is not high enough to make the ASP profitable,
especially when the setup cost S is large. In this case, the ASP’s best response price will not be the price
given by equation (8), which is analyzed in Proposition 6.
ii
When the ASP serves the whole market, the minimum price required to cover the ASP’s setup costs is
2S + c . Since S << C MOTS (specifically, 2S < CMOTS ), the ASP’s price satisfies pa > 2 S + c .
39