Sei sulla pagina 1di 23

MANUFACTURING & SERVICE

OPERATIONS MANAGEMENT
Vol. 12, No. 2, Spring 2010, pp. 213235
issn1523-4614[ eissn1526-5498[ 10[ 1202[ 0213
informs

doi 10.1287/msom.1090.0266
2010 INFORMS
Service Provider Competition: Delay Cost Structure,
Segmentation, and Cost Advantage
Maxim Afanasyev, Haim Mendelson
Graduate School of Business, Stanford University, Stanford, California 94305
{mafanasyev@mafanasyev.com, haim@stanford.edu}
W
e model competition between two providers who serve delay-sensitive customers. We compare a gen-
eralized delay cost structure, where a customers delay cost depends on her service valuation, with the
traditional additive delay cost structure, where the delay cost is independent of the customers service valua-
tion. Under the additive delay cost structure, service providers offer different prices and expected delays, but
customers are indifferent between the providers. Under the generalized delay cost structure, when the providers
have different capacity or operating costs, we obtain value-based market segmentation, whereby higher-value
customers choose one provider and lower-value customers choose the other. We study how the delay cost
parameters, the market size, and the service providers costs affect the structure of the equilibrium.
Key words: delay cost structure; value-based market segmentation; service competition
History: Received: May 23, 2008; accepted: March 22, 2009. Published online in Articles in Advance
September 14, 2009.
1. Introduction
In this paper we model competition between service
providers when their customers are sensitive to delay.
Our primary focus is on how the customers delay
cost structure and the asymmetry in the providers
costs affect the market equilibria. In most previous
research, customers were assumed to have an addi-
tive delay cost structure. In reality, however, we often
observe interdependence between customers service
valuations and their delay costs. We argue that this
interdependence gives rise to major changes both in
the structure of the market equilibrium and in the
levels of arrival rates, prices, and service capacities.
We show how the delay cost structure, market size,
and differences between the providers costs (capacity
costs or variable costs of providing the service) affect
service differentiation and market segmentation.
The assumption of an additive delay cost structure,
which is common in the literature on the economics
of congestion and delay, is reasonable in a variety of
settings. This assumption means that the cost of delay
is independent of the customers service valuation.
This would be the case, for example, if the delay cost
reects merely productivity losses or the value of lost
time, regardless of the nature of the service provided.
Consider, for example, the case of auto repair. While a
customer is waiting for her car to be repaired, she may
be using a rental car or public transportation. Thus,
her expected delay cost may be approximated by the
product of the expected number of days needed to
repair the car by the daily loss, the latter reect-
ing the value of lost time and the costs of alterna-
tive transportation.
1
Assume that this daily loss is
not related to the value of the repair service.
2
Then,
if a repair facility repairs the car a day faster (on
average) than its competitor but charges a premium
equal to the expected daily loss, all customers will be
indifferent between the competing facilities. Although
the repair facilities are differentiated (offering differ-
ent prices and expected delays), there is no market
segmentation.
However, there are numerous situations where the
delay cost and the value of service are interdepen-
dent. In electronic brokerage, for example, a delay in
1
Other opportunity costs of time may also be included in the daily
loss.
2
Across customers, the value of the repair service could be pos-
itively correlated with the daily loss, because higher-income cus-
tomers tend to have both higher opportunity costs and more
expensive cars.
213
Afanasyev and Mendelson: Service Provider Competition
214 Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS
trade execution deates the investors expected prot,
and customers who trade frequently are willing to
pay a larger premium for fast execution. Thus, there is
a positive relationship among an investors frequency
of trading, the total amount traded, and the delay cost
she incurs over, for example, a year, which creates an
interdependence between the value of the brokerage
service to the customer and her delay cost. A similar
relationship holds when the cost of execution delay is
proportional to the trading volume at the transaction
level (see Dewan and Mendelson 1998, 2001). Indeed,
major electronic brokerage rms (e.g., Ameritrade,
E*TRADE, and Scottrade) prominently advertise their
average or median execution speeds, and brokerages
that target frequent traders bear the additional costs
of colocating their servers inside the exchanges to
reduce delays (Martin and Malykhina 2007). In hospi-
tals, patients may incur a delay of weeks waiting for
surgery, and the longer the surgical delay, the higher
the mortality rate (Weller et al. 2005). We expect that
for patients in poor health, surgery is particularly
critical and they will suffer from a delay more than
patients whose condition is not as severe, implying
that the value of service and the delay cost are inter-
dependent. In online video streaming, transmission
delays affect the quality of the content viewed by the
customer. The higher the content quality, the more a
customer suffers if there are data transmission delays.
Thus, the delay cost is interdependent with the value
of the content to the customer.
Product development provides another example
of interdependence between service valuations and
delay costs. Adler et al. (1995) show that product
development can be modeled as a queue of projects
sharing common resources. While waiting in the
queue, new products become less attractive because
of changes in the marketplace and the competition.
A longer development lead time may result in a
loss of market leadership, a decrease in potential
revenue, and lower sales, as customers needs change
over time (see Takeuchi and Nonaka 1986, Stalk
1988, Gupta and Wilemon 1990, Mabert et al. 1992,
Millson et al. 1992). Gupta and Wilemon (1990) nd
that a six-month delay in the market introduction of
a high-tech product reduces average prot by 33%
over ve years. In the auto industry, the larger the
market potential for a new car model, the larger
the revenue loss resulting from model introduction
delays. Some well-known examples of companies
that achieved competitive advantage by reducing
their new product development times include
Toyotas introduction of the Prius, Sun Microsystems
in the workstation market, Sony in the CD market,
and Samsung in memory chips and LCD televisions
in the 2000s (Stalk and Hout 1990, Clark et al. 1987,
Smith and Reinertsen 1998, Sun et al. 2004).
In this paper we study how the delay cost structure
affects the competition between service providers and
the resulting market structure. In particular, we dis-
tinguish between service differentiation and market
segmentation (see Smith 1956, Fradera 1986). Whereas
the additive delay cost structure results in differen-
tiated services, interdependence between service val-
uations and delay costs leads to value-based market
segmentation. We give the following denitions.
Service differentiation occurs when at least one
rms services are different from those offered by its
competitor(s).
Market segmentation requires that, in addition, the
market is divided into customer segments so that
the customers in at least one segment strictly pre-
fer the services offered by one rm over the services
offered by its competitor(s).
Under service differentiation, (some) rms offer dif-
ferent service variants, typically at different prices,
but customers may be indifferent between these
alternative offerings. Under market segmentation,
different service variants attract different customer
segments. This means that some customers have a
strict preference for one rms offering. The segmen-
tation is value based if the customers are segmented
based on their valuations of the service.
We consider two competing service providers
whose customers are sensitive to delay. We model
interdependent service valuations and delay costs
using the generalized delay cost structure proposed
by Afeche and Mendelson (2004), which allows for
both additive and multiplicative delay cost compo-
nents. Earlier work assumed that customers util-
ity was additive in value and delay cost, namely
u(U, |) =U D(|), where U is the gross utility absent
delay, | is the delay in the system, equal to the sum
of the waiting time in the queue and the service
time, D(|) =d | is the expected delay cost, and u is
Afanasyev and Mendelson: Service Provider Competition
Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS 215
the customers expected net utility. Under the gen-
eralized delay cost structure, the expected utility is
u(U, |) = (|) U D(|), where (|) is a decreasing
function that deates the customers value for the
service. Each customer is self-interested and maxi-
mizes her own expected utility, which she forecasts
using the distribution of the steady-state delay; i.e.,
||u(U, |)] = ||(|) U D(|)]. In the linear case,
D(|) =d | and (|) =1 . |, so the expected utility
equals ||u(U, |)] =U (1.W) dW, where W is the
expected delay in the system.
We nd that the equilibrium is characterized by ser-
vice differentiation. Say one rm provides fast ser-
vice and charges a higher price and the other offers
slow service at a lower price. In addition, when the
service valuations and delay costs are interdependent
and the providers costs (capacity costs or operat-
ing costs) are asymmetric, we nd value-based mar-
ket segmentation into a high- and low-end customer
segment. We study how the delay sensitivity param-
eters and the total market size affect measures of
service differentiation and market segmentation. We
nd that an increase in the multiplicative delay sen-
sitivity affects service differentiation nonmonotoni-
cally and increases the level of market segmentation.
Growth in the total market size decreases both the
degrees of service differentiation and market segmen-
tation until they disappear at the limit.
Our results on value-based market segmentation
are consistent with what we observe in a number of
markets where customers are sensitive to delay. One
example is public transportation in the San Francisco
Bay Area, where Viton (1981) nds that the value of
service and the delay cost are interdependent. Con-
sidering the competition between bus service and the
Bay Area Rapid Transit system (BART), Viton (1981)
nds that the systems differentiate their services in
terms of delay, with BART having a lower average
delay than the bus system and charging a higher
fare. Viton (1981) further nds value-based market
segmentation, with BART focusing on wealthier cus-
tomers and the bus system targeting the lower-income
population.
Mortgage loan origination is another service ex-
hibiting both an interdependence between delay cost
and service value and value-based market segmen-
tation. The longer a customer waits for mortgage
approval, the more the home price and interest rates
uctuate and the higher the probability that the deal
will fall through before the loan is approved. Also,
the higher the home price, the greater the loss, so
the delay cost and the value of the origination ser-
vice are interdependent. In 1986, Citicorp launched
MortgagePower, a computer-based loan origination
system that dramatically reduced Citicorps mortgage
origination delay. In turn, Citicorp charged customers
higher processing fees, so its service was differenti-
ated by less delay and a higher price. Citicorp tar-
geted customers who borrowed larger amounts and
provided faster service in return for higher fees (Hess
and Kremer 1994, Stalk and Hout 1990). In con-
trast, traditional loan originators charged lower fees
and served lower-end customers whose loans were
smaller. A similar segmentation relating morgtgage
loan amounts, service fees, and average delay is
found by Guttentag (2001).
In the decorative laminates industry, customers
service values and delay costs are interdependent
because the disutility of construction delay is larger
for residential cabinetmakers and commercial speci-
cation customers, who typically work on expensive
projects, than for original equipment manufacturer
(OEM) direct purchasers. Accordingly, the former
are willing to pay a premium for fast delivery.
The industry has two major product offeringsfaster
but more expensive and slower but cheaperand is
also characterized by value-based market segmenta-
tion, with Ralph Wilsonart providing faster service to
residential cabinetmakers and commercial specica-
tion customers, and slower Formica dominating the
price-sensitive OEM segment (Stalk and Hout 1990,
Hamilton 2007, Feldman 2007).
In work under way, Afanasyev and Mendelson
(2009) consider movie production in Hollywood as a
special case of new product development. They esti-
mate the delay costs associated with this process and
show that value and delay cost are interdependent.
They then show how the market for studios is seg-
mentated based on movies market potential, with the
faster studios producing movies with higher expected
box ofce revenues. They also show that the lead time
in introducing movies into foreign markets exhibits
similar characteristics.
Afanasyev and Mendelson: Service Provider Competition
216 Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS
The remainder of this paper is organized as follows.
We review the literature in 2. Section 3 considers
our benchmark, additive delay cost model. In 4 we
study the effects of the generalized delay cost struc-
ture. In 5 and 6 we study the effects of market size
and differences in operating costs, and 7 offers our
concluding remarks.
2. Literature Review
In this section, we review the literature on delay sensi-
tivity and service provider (S) competition. Numerous
papers study markets with delay-sensitive customers
(see Hassin and Haviv 2002). Most of these papers
assume a discrete number of classes and consider the
optimal pricing or scheduling of services, or both,
among the classes, sometimes subject to incentive
compatibility constraints. Table 1 summarizes the lit-
erature on the effects of competition on queueing sys-
tems; we focus here on the rst four papers, which are
closest to ours. Unlike our model, capacity is exoge-
nous and service valuations are constant in Chen
and Wan (2003) and So (2000), service providers are
symmetric in Chen and Wan (2005), and in all four
models the service providers engage in price compe-
tition. Chen and Wan (2003) study how the market
structure changes as a function of market size. For a
small total arrival rate, the market is dominated by
one of the providers. As the arrival rate increases,
there is a unique equilibrium with both rms in the
market, and then no equilibrium, followed by a con-
tinuum of equilibria and nally, a noncompetitive
market. In a model with endogenous service capaci-
ties, Chen and Wan (2005) nd that although some of
the irregular equilibria in Chen and Wan (2003) are
eliminated, for a large market size there is a contin-
uum of equilibria. In contrast, we nd a more regular
effect of market size on market structure under our
model.
Cachon and Harker (2002) derive conditions under
which two competing service providers exist in the
market and study how scale economies affect the mar-
ket equilibrium. They nd that the lower-cost S may
have a higher market share as well as a higher price.
They also show that the full price (service fee plus
delay cost) is an increasing function of the operating
and capacity costs. To answer their research questions,
Cachon and Harker (2002) need not explicitly model
customer choice as we do. Unlike Cachon and Harker
(2002), where the rms choose prices and waiting
times and the capacities adjust in equilibrium, in our
model rms choose capacities and arrival rates and
prices adjust in equilibrium. This enables us to prove
the existence of an equilibrium.
In So (2000), heterogenous rms with exogenous
capacities compete by announcing their prices and
delivery time guarantees, again unlike our model.
The market has a xed size and all customers are
served, so each rm serves a portion of the market.
The lower a providers price and delivery time guar-
antee, the higher its market share. So (2000) nds that
providers exploit their relative advantage to differen-
tiate their services; e.g., the high-capacity S will offer
better time guarantees, whereas the lower-operating-
cost S charges a lower price.
The equilibria in Chen and Wan (2003) and So
(2000) are characterized by service differentiation but
no market segmentation, with one S offering shorter
delays and charging a price premium and the other
offering a longer delay and a lower price that exactly
offsets each customers delay cost savings. In Chen
and Wan (2005), there is no service differentiation (the
providers are symmetric), and in Cachon and Harker
(2002), there is service differentiation, but the underly-
ing customer choices are not specied in the demand
model.
With the exception of this paper, all the papers in
Table 1 that specify a delay cost structure assume
the additive cost structure, which (as we show) is
an important driver of the equilibria they obtain. To
date, few papers have considered a delay cost struc-
ture that is not additive, and none of them (to our
knowledge) studied the effects of competition. Afeche
and Mendelson (2004) present a generalized delay
cost structure and apply it to a single service facility
that has a xed capacity. They compare the revenue-
maximizing and socially optimal equilibria for a pri-
ority queue and show that some classical results may
not hold under the generalized delay cost structure.
3
They also compare the effects of bidding strategies
to uniform pricing under the generalized delay cost
3
For example, the classical results that the revenue-maximizing
admission price is higher than the socially optimal price, and the
revenue-maximizing use is lower than the socially optimal use.
Afanasyev and Mendelson: Service Provider Competition
Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS 217
T
a
b
l
e
1
C
o
m
p
a
r
i
s
o
n
o
f
T
h
i
s
P
a
p
e
r
t
o
t
h
e
R
e
l
e
v
a
n
t
L
i
t
e
r
a
t
u
r
e
S
t
u
d
y
S
e
r
v
i
c
e
p
r
o
v
i
d
e
r
s
S
e
r
v
i
c
e
p
r
o
v
i
d
e
r
d
e
c
i
s
i
o
n
s
S
e
r
v
i
c
e
v
a
l
u
a
t
i
o
n
s
D
e
l
a
y
c
o
s
t
s
t
r
u
c
t
u
r
e
R
e
s
e
a
r
c
h
q
u
e
s
t
i
o
n
s
/
R
e
s
u
l
t
s
S
o
(
2
0
0
0
)
A
a
s
y
m
m
e
t
r
i
c
P
r
i
c
e
,
t
i
m
e
g
u
a
r
a
n
t
e
e
D
e
m
a
n
d
b
a
s
e
d
o
n
p
r
i
c
e
,
t
i
m
e
g
u
a
r
a
n
t
e
e
s
N
o
t
a
p
p
l
i
c
a
b
l
e
(
t
i
m
e
g
u
a
r
a
n
t
e
e
m
o
d
e
l
)
F
i
n
d
s
t
h
a
t
p
r
o
v
i
d
e
r
s
e
x
p
l
o
i
t
t
h
e
i
r
c
a
p
a
c
i
t
y
a
d
v
a
n
t
a
g
e
t
o
d
i
f
f
e
r
e
n
t
i
a
t
e
t
h
e
i
r
p
r
i
c
e
s
a
n
d
s
e
r
v
i
c
e
s
,
a
n
d
s
t
u
d
i
e
s
d
e
t
e
r
m
i
n
a
n
t
s
o
f
e
q
u
i
l
i
b
r
i
u
m
p
r
i
c
e
s
C
a
c
h
o
n
a
n
d
H
a
r
k
e
r
(
2
0
0
2
)
2
a
s
y
m
m
e
t
r
i
c
P
r
i
c
e
,
d
e
l
a
y
D
e
m
a
n
d
f
u
n
c
t
i
o
n
A
d
d
i
t
i
v
e
F
i
n
d
t
h
a
t
s
c
a
l
e
e
c
o
n
o
m
i
e
s
i
n
c
r
e
a
s
e
i
n
c
e
n
t
i
v
e
s
t
o
o
u
t
s
o
u
r
c
e
a
n
d
t
h
e
l
o
w
e
r
-
c
o
s
t
S
h
a
s
h
i
g
h
e
r
m
a
r
k
e
t
s
h
a
r
e
a
n
d
h
i
g
h
e
r
p
r
i
c
e
C
h
e
n
a
n
d
W
a
n
(
2
0
0
3
)
2
a
s
y
m
m
e
t
r
i
c
P
r
i
c
e
C
o
n
s
t
a
n
t
A
d
d
i
t
i
v
e
S
t
u
d
y
h
o
w
m
a
r
k
e
t
s
i
z
e
a
f
f
e
c
t
s
m
a
r
k
e
t
s
t
r
u
c
t
u
r
e
C
h
e
n
a
n
d
W
a
n
(
2
0
0
5
)
2
s
y
m
m
e
t
r
i
c
P
r
i
c
e
,
c
a
p
a
c
i
t
y
C
o
n
s
t
a
n
t
A
d
d
i
t
i
v
e
S
t
u
d
y
h
o
w
m
a
r
k
e
t
s
i
z
e
a
f
f
e
c
t
s
m
a
r
k
e
t
s
t
r
u
c
t
u
r
e
L
u
s
k
i
(
1
9
7
6
)
,
L
e
v
h
a
r
i
a
n
d
L
u
s
k
i
(
1
9
7
8
)
2
s
y
m
m
e
t
r
i
c
P
r
i
c
e
C
o
n
s
t
a
n
t
A
d
d
i
t
i
v
e
/
h
e
t
e
r
o
g
e
n
e
o
u
s
C
o
n
d
i
t
i
o
n
s
f
o
r
s
e
r
v
i
c
e
d
i
f
f
e
r
e
n
t
i
a
t
i
o
n
R
e
i
t
m
a
n
(
1
9
9
1
)
A
s
y
m
m
e
t
r
i
c
P
r
i
c
e
,
c
a
p
a
c
i
t
y
C
o
n
s
t
a
n
t
A
d
d
i
t
i
v
e
/
h
e
t
e
r
o
g
e
n
e
o
u
s
C
o
n
d
i
t
i
o
n
s
f
o
r
s
e
r
v
i
c
e
d
i
f
f
e
r
e
n
t
i
a
t
i
o
n
L
o
c
h
(
1
9
9
1
)
2
s
y
m
m
e
t
r
i
c
T
w
o
m
o
d
e
l
s
:
(
1
)
p
r
i
c
e
;
(
2
)
p
r
i
c
e
a
n
d
c
a
p
a
c
i
t
y
H
e
t
e
r
o
g
e
n
e
o
u
s
A
d
d
i
t
i
v
e
C
o
n
d
i
t
i
o
n
s
f
o
r
a
s
y
m
m
e
t
r
i
c
e
q
u
i
l
i
b
r
i
a
D
e
n
e
c
k
e
r
e
a
n
d
P
e
c
k
(
1
9
9
5
)
A
s
y
m
m
e
t
r
i
c
P
r
i
c
e
,
c
a
p
a
c
i
t
y
C
o
n
s
t
a
n
t
N
o
t
s
p
e
c
i

e
d
S
t
u
d
y
e
x
i
s
t
e
n
c
e
o
f
e
q
u
i
l
i
b
r
i
u
m
L
e
d
e
r
e
r
a
n
d
L
i
(
1
9
9
7
)
L
a
r
g
e
A
,
a
s
y
m
m
e
t
r
i
c
S
c
h
e
d
u
l
i
n
g
p
o
l
i
c
y
D
e
m
a
n
d
f
u
n
c
t
i
o
n
N
o
t
s
p
e
c
i

e
d
S
h
o
w
t
h
a
t
l
o
w
e
r
-
c
o
s
t
a
n
d
l
o
w
e
r
-
v
a
r
i
a
b
i
l
i
t
y
S
h
a
v
e
l
a
r
g
e
r
m
a
r
k
e
t
s
h
a
r
e
,
h
i
g
h
e
r
c
a
p
a
c
i
t
y
u
t
i
l
i
z
a
t
i
o
n
,
a
n
d
h
i
g
h
e
r
p
r
o

t
s
G
i
b
b
e
n
s
e
t
a
l
.
(
2
0
0
0
)
2
s
y
m
m
e
t
r
i
c
P
r
i
c
e
,
o
p
t
i
o
n
t
o
c
r
e
a
t
e
s
u
b
n
e
t
w
o
r
k
C
o
n
s
t
a
n
t
A
d
d
i
t
i
v
e
/
h
e
t
e
r
o
g
e
n
e
o
u
s
F
i
n
d
t
h
a
t
u
n
d
e
r
c
o
m
p
e
t
i
t
i
o
n
w
i
t
h
t
w
o
c
l
a
s
s
e
s
,
e
a
c
h
S
f
o
c
u
s
e
s
o
n
l
y
o
n
o
n
e
c
l
a
s
s
A
r
m
o
n
y
a
n
d
H
a
v
i
v
(
2
0
0
3
)
2
s
y
m
m
e
t
r
i
c
P
r
i
c
e
C
o
n
s
t
a
n
t
w
i
t
h
i
n
e
a
c
h
o
f
t
w
o
c
u
s
t
o
m
e
r
c
l
a
s
s
e
s
A
d
d
i
t
i
v
e
w
i
t
h
i
n
e
a
c
h
c
l
a
s
s
A
n
a
l
y
z
e
c
o
m
p
e
t
i
t
i
o
n
a
n
d

n
d
s
e
r
v
i
c
e
d
i
f
f
e
r
e
n
t
i
a
t
i
o
n
A
l
l
o
n
a
n
d
F
e
d
e
r
g
r
u
e
n
(
2
0
0
6
)
2
,
d
i
f
f
e
r
e
n
t
m
a
r
g
i
n
a
l
c
o
s
t
s
T
h
r
e
e
m
o
d
e
l
s
:
(
1
)
p
r
i
c
e
,
(
2
)
c
a
p
a
c
i
t
y
,
(
3
)
s
i
m
u
l
t
a
n
e
o
u
s
p
r
i
c
e
a
n
d
c
a
p
a
c
i
t
y
D
e
m
a
n
d
f
u
n
c
t
i
o
n
N
o
t
s
p
e
c
i

e
d
F
i
n
d
t
h
a
t
o
u
t
s
o
u
r
c
i
n
g
(
s
e
r
v
i
c
e
p
o
o
l
i
n
g
)
i
s
b
e
n
e

c
i
a
l
A
l
l
o
n
a
n
d
F
e
d
e
r
g
r
u
e
n
(
2
0
0
7
)
2
a
s
y
m
m
e
t
r
i
c
T
h
r
e
e
m
o
d
e
l
s
:
(
1
)
d
e
l
a
y
f
o
l
l
o
w
e
d
b
y
p
r
i
c
e
,
(
2
)
p
r
i
c
e
f
o
l
l
o
w
e
d
b
y
d
e
l
a
y
,
(
3
)
s
i
m
u
l
t
a
n
e
o
u
s
p
r
i
c
e
a
n
d
d
e
l
a
y
D
e
m
a
n
d
f
u
n
c
t
i
o
n
N
o
t
s
p
e
c
i

e
d
C
o
m
p
a
r
e
t
h
e
t
h
r
e
e
m
o
d
e
l
s
,
s
h
o
w
i
n
g
t
h
a
t
m
o
d
e
l
1
l
e
a
d
s
t
o
h
i
g
h
e
s
t
w
a
i
t
i
n
g
t
i
m
e
a
n
d
l
o
w
e
s
t
p
r
i
c
e
s
P
e
k
g
u
n
e
t
a
l
.
(
2
0
0
8
)
2
a
s
y
m
m
e
t
r
i
c
P
r
i
c
e
,
l
e
a
d
t
i
m
e
(
c
e
n
t
r
a
l
i
z
e
d
v
s
.
d
e
c
e
n
t
r
a
l
i
z
e
d
)
D
e
m
a
n
d
f
u
n
c
t
i
o
n
A
d
d
i
t
i
v
e
S
t
u
d
y
e
f
f
e
c
t
s
o
f
d
e
c
e
n
t
r
a
l
i
z
a
t
i
o
n
o
n
p
r
i
c
e
a
n
d
l
e
a
d
-
t
i
m
e
d
e
c
i
s
i
o
n
s
T
h
i
s
p
a
p
e
r
2
a
s
y
m
m
e
t
r
i
c
A
r
r
i
v
a
l
r
a
t
e
a
n
d
c
a
p
a
c
i
t
y
H
e
t
e
r
o
g
e
n
e
o
u
s
/
d
e
m
a
n
d
f
u
n
c
t
i
o
n
G
e
n
e
r
a
l
i
z
e
d
E
f
f
e
c
t
s
o
f
g
e
n
e
r
a
l
i
z
e
d
d
e
l
a
y
c
o
s
t
s
t
r
u
c
t
u
r
e
;
c
o
n
d
i
t
i
o
n
s
f
o
r
a
n
d
i
m
p
l
i
c
a
t
i
o
n
s
o
f
m
a
r
k
e
t
s
e
g
m
e
n
t
a
t
i
o
n
Afanasyev and Mendelson: Service Provider Competition
218 Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS
structure. Katta and Sethuraman (2005) consider a
queue with priorities and customers with a gener-
alized delay cost structure. They propose an algo-
rithm that sorts customers into priority groups to
maximize the revenues of a monopoly. Kalvenes and
Keon (2007) consider an M/M/m loss system and
customers with a multiplicative delay cost structure
and propose a mechanism to reallocate customer ser-
vice times from periods when the system is highly
congested to periods when it is less congested. Our
paper studies the effects of interdependence between
customers service valuations and delay costs on S
competition. We show that when the delay cost and
service valuations are interdependent and the rms
costs are different, the equilibria must be character-
ized by value-based market segmentation.
3. Basic Model
We start with a traditional model of competition
between two Ss who serve customers with heteroge-
neous service valuations that are subject to an addi-
tive delay cost. We rst present our model of user
behavior and then proceed with a description of the
rms and the competition between them. This model
forms a foundation for the analyses that follow.
3.1. User Behavior
We rst present the underlying user behavior model
for a single system with no delays. Potential cus-
tomers arrive following a Poisson process with rate A.
The values they derive from receiving the ser-
vice when there is no delay are modeled as an
independent and identically distributed (i.i.d.) sam-
ple from a random variable U with cumulative dis-
tribution function (c.d.f.) . The probability that an
arriving customer values the service at or higher is

() =1 (). If only customers who value the ser-


vice at or higher join the system, the effective arrival
rate is \ = A

(), which gives rise to the inverse
demand function =

1
(\,A), or V
/
(\) =

1
(\,A)
(see Mendelson 1985, Dewan and Mendelson 1990),
where V(\) is the expected total value created for
users per unit of time when the (effective) arrival rate
is \, and V
/
(\) is the corresponding marginal value,
which is also equal to the cutoff value . Given the
c.d.f. , we can calculate V(\) and V
/
(\), the demand
curve when there is no delay.
Example. Let the distribution of consumer values
U be uniform on |0, 100] and the total arrival rate be
A = 300. Then the expected number of arriving cus-
tomers with service valuations above per unit of
time equals 300 Pr{U ], which gives rise to the
linear demand curve
V
/
(\) =100 \,3. (1)
We use this demand curve in our numerical examples
throughout the paper.
We assume that V
/
(\) is strictly decreasing, cor-
responding to the usual assumption of a monotone
decreasing demand curve, i.e.,
V
//
(\) -0 for \ 0, (2)
and that
V
//
(\
1
\
2
) \
1
V
///
(\
1
\
2
) -0 for \
1
, \
2
0. (3)
Assumption (3) is needed to ensure that the rms
prot functions are supermodular and is satised
by the commonly used linear V
/
(\) = o p\ (o >0,
p >0) and constant elasticity V
/
(\) = |\
p
(| > 0,
0 -p -1) demand curves.
Customers are sensitive to delay, and in this section
we follow the common assumption that the delay cost
is d per customer per unit of time. We assume that
the queue length is unobservable and that customers
make their decisions based on their expected through-
put time W. Let P be the price charged per job. It is
well known that the analogue of the demand curve
when delay costs are taken into account then becomes
P =V
/
(\) d W. (4)
This modeling approach is common in the economics
of queues literature. Note that the model can be spec-
ied in terms of the distribution of user valuations
or the inverse demand curve V
/
(\). If the model is
specied in terms of probabilities, the effective arrival
rate is A Pr{U d W P], reecting the fact that as
the expected delay W increases, customers are willing
to pay less for the service, hence the expected number
of customers decreases.
Afanasyev and Mendelson: Service Provider Competition
Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS 219
3.2. Service Providers
We consider a market with two competing Ss, S
1
and S
2
, with respective constant marginal capacity
costs g
1
and g
2
. We assume g
1
g
2
, so S
1
is the lower-
cost S. Each service provider S

is modeled as an
A,G,1 queue with arrival rate \

. We assume that a
customers service time is 8,j

, where 8 is a random
variable with coefcient of variation r (the same for
both rms) and mean 1, and j

is the service capac-


ity of S

. Thus, larger investments in service capacity


stochastically decrease each providers service time.
We further assume the rst-come, rst-served queue
discipline and unobservable customer service valua-
tions. The expected delay at S

with capacity j

expe-
riencing arrival rate \

is given by the Pollaczek-


Khinchin formula for the expected total time in the
system,
4
W

r
2
j

1
2j

. (5)
The expected delay at each S

increases in \

and r
and decreases in j

. We further assume that there is a


direct operating cost c of serving each customer.
5
We
assume that each S charges a single price and that the
demand is high enough for both Ss to be protable at
zero delay cost; i.e.,
V
/
(0) >c max{g
1
, g
2
]. (6)
To ensure internal solutions, we also assume that for
a high enough arrival rate \, it is unprotable to serve
additional customers; i.e., there is a \ satisfying
V
/
(\) -c min{g
1
, g
2
], (7)
where c is the rms unit operating cost and g

( =1, 2) is S

s unit capacity cost.


3.3. Game Structure and Equilibria
We model competition as a Cournot game between
the two rms with two stages: the rms choice of \

and j

( =1, 2), followed by customer choice (which,


as we show below, is degenerate in equilibrium).
In the rst stage of the game, each S

, =1, 2 decides
on its capacity j

and arrival rate \

, analogous to the
4
See, e.g., Equation (3.17) in Taylor and Karlin (1998) with =1,j
and t =r,j.
5
In 6 we allow the cs to be different across rms.
traditional Cournot model, where the rms rst set
their respective quantities. In the second stage, given
\

and j

for =1, 2, customers decide which (if any)


S to join, and the equilibrium prices are adjusted
accordingly. This ensures the existence of an equi-
librium. If the rms choose their prices and wait-
ing times rst and capacities then adjust to satisfy
the equilibrium conditions, an equilibrium does not
always exist (Cachon and Harker 2002).
What happens in the second stage of the game
depends on whether there are one or two rms in the
market.
6
If only S

is in the market, each customer


calculates the expected delay W

and service fee P

,
given \

and j

. The net utility of a customer with


service valuation U is U d W

if the customer is
served, or zero if the customer is not served. The cus-
tomer thus chooses to receive the service if and only
if U d W

> 0. In this case, there is obviously


a

U such that all customers with U >

U choose to
receive service, and all customers with U

U decline
the service. Because customer choice must satisfy the
individual rationality constraint, S

s prot maximiza-
tion problem in the rst stage is
max
\

, j

|\

c\

]
s.t. V
/
(\

) d W

=P

.
(8)
Next, consider the customer choice when two
providers are in the market. Now each customer has
three options to choose from:
(a) join S
1
(net utility equals U d W
1
P
1
),
(b) join S
2
(net utility equals U d W
2
P
2
), or
(c) do not join (net utility equals 0).
The customer decides among these alternatives by
comparing the net utilities in (a), (b), and (c) above.
When both providers serve customers, all cus-
tomers must be indifferent between S
1
and S
2
.
7
Indeed, if there exists a customer who strictly prefers
S

( =1, 2), namely, for some valuation U,


U d W

>U d W

, (9)
then all customers will strictly prefer S

(because
(9) will hold for all U), so S

will serve no cus-


tomers, which is a contradiction. This means that the
6
If there is no rm in the market, the solution is trivial.
7
We will show later that this is the direct result of the assumed
additive delay cost structure.
Afanasyev and Mendelson: Service Provider Competition
220 Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS
customer choice problem is degenerate (as all cus-
tomers are indifferent between the two rms). This
is often modeled by randomizing the choice between
the providers. Thus, there exists a service valua-
tion threshold

U such that the customers with ser-
vice valuations above

U are randomly allocated and
those below

U are not served (because their utilities
upon being served by either provider would be nega-
tive). Although customers are indifferent between the
two providers, consistency with the rst-stage deci-
sions requires that the probability of joining S

equals
\

,(\
1
\
2
) ( =1, 2).
To complete the specication of the game, recall
that each S

, = 1, 2 decides on j

and \

, building
on the customers decision criteria (which, as shown
above, are degenerate). Hence, a necessary condition
for an equilibrium is that each S

solves
max
\

, j

|\

c\

]
s.t. V
/
(\
1
\
2
) d W

=P

,
(10)
where W

is the expected delay given by Equation (5).


If S

decides to stay out of the market, it sets \

=
j

=0, and S

solves (8). In a mixed-strategy equilib-


rium, customers with valuations above some thresh-
old are randomly allocated between the two rms,
and customers with valuations below this threshold
are not served. It is clear from the above discussion
that under the additive delay cost structure, the only
possible equilibria are mixed-strategy equilibria.
The Cournot game is often presented as a
game with explicit or implicit probabilistic customer
choices. In our game, the probability that a customer
with service valuation U is served by S

equals
p

(U) =

\
1
\
2
if U >

U, =1, 2,
0 otherwise.
(11)
This also translates directly to probabilistic choice
models often used in the marketing literature (see
Currim 1982, Kamakura and Srivastava 1984, Grover
and Srinivasan 1987, Kamakura and Russell 1989).
The marketing literature also allows for more gen-
eral forms than (11), which is implied by our model.
This suggests that a more general model specication
might lead to a generalization of (11), such as that
corresponding to the multinomial logistic model.
3.4. Equilibrium
Intuitively, an increase in the delay sensitivity d in-
creases customers delay costs and lowers the effective
total demand. As a result, for small d, both providers
are in the market, but as d increases, only one S sur-
vives. For larger values of d, there is no S in the mar-
ket. Proposition 1 makes this intuition precise. Proofs
of propositions and corollaries are in the appendix.
Proposition 1. There exist constants 0 - d
1
d
2

d
3
d
4
such that the following applies.
(1) Duopoly solution: For d - d
1
, there is a mixed-
strategy equilibrium with both S
1
and S
2
in the market,
where the capacity j

and market size \

of S

are given by
the solution to the equations
8

(\

r
2
j

)j

(j

)
2

\

r
2
j

1
2j
2

=
g

d
, =1,2,
V
/
(\
1
\
2
)\

V
//
(\
1
\
2
)dW

d\

r
2
1
2(j

)
2
=c, =1,2,
(12)
where W

are given by (5).


(2) Single monopoly solution: For d |d
1
, d
2
)|d
3
, d
4
),
there is an equilibrium with only the low-cost provider, S
1
,
in the market. The equilibrium capacity j
1
and market size
\
1
are given by the solution to the equations

(\
1
r
2
j
1
)j
1
(j
1
\
1
)
2

\
1
r
2
j
1
\
1
1

1
2j
2
1
=
g
1
d\
1
,
V
/
(\
1
) \
1
V
//
(\
1
) dW
1
d\
1
r
2
1
2(j
1
\
1
)
2
=c,
(13)
where W
1
is given by (5) (with =1).
(3) Multiple monopoly solutions: For d |d
2
, d
3
), there
are two equilibria: one with only S
1
in the market and one
with only S
2
in the market. In the equilibrium with S
m
in
the market, the capacity j
m
of S
m
and its market size \
m
are given by the solutions to the equations

(\
m
r
2
j
m
)j
m
(j
m
\
m
)
2

\
m
r
2
j
m
\
m
1

1
2j
2
m
=
g
m
d\
m
,
V
/
(\
m
) \
m
V
//
(\
m
) dW
m
d\
m
r
2
1
2(j
m
\
m
)
2
=c,
(14)
where W
m
is given by (5) (with =m).
8
For d = 0, the equations are j

= \

and V
/
(\
1
\
2
)
\

V
//
(\
1
\
2
) =c.
Afanasyev and Mendelson: Service Provider Competition
Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS 221
(4) Market is not served: For d d
4
, there is no S in
the market, and \
1
=\
2
=j
1
=j
2
=0.
The sequence of market structures in Proposition 1
shows that for small delay sensitivities, both rms
are in the market. As the delay sensitivity increases,
the higher-cost S leaves the market. This happens
because, as a monopoly, S
2
can protably serve fewer
customers than S
1
; hence the entry barrier for S
1
is
lower than for S
2
. As the delay sensitivity increases
further, only one S can survive. As the delay sensitiv-
ity increases further, S
2
cannot obtain positive prots.
Because of its cost advantage, S
1
stays in the market
longer but eventually, for larger values of d, leaves the
market as well. Proposition 1 allows for multiple equi-
libria, so the constants d
1
through d
4
in Proposition 1
need not be unique. Furthermore, the interval |0, d
1
)
consists of up to three subintervals:
(i) |0, d
D
1
), where only duopoly equilibria exist;
(ii) |d
D
1
, d
A
1
), where duopoly and S
1
monopoly
equilibria exist; and
(iii) |d
A
1
, d
1
), where duopoly and two, S
1
or S
2
,
monopoly equilibria exist.
In 5 we study the progression of the different equi-
libria as a function of the market size A and discuss
these subintervals in more detail.
Proposition 1 also shows that when the two rms
are in the market, they may offer differentiated ser-
vices, but customers are always indifferent between
them; i.e., there is a mixed strategy equilibrium. If the
providers capacity costs are equal, the expected delay
and price are determined, so there is neither service
differentiation nor market segmentation. If g
1
-g
2
,
S
1
has a cost advantage: it sets a lower expected delay
than S
2
but in equilibrium charges a price premium
that exactly equals the delay cost savings of each cus-
tomer; i.e.,
P
1
P
2
=d (W
2
W
1
). (15)
This means that while the Ss offer differentiated ser-
vices, there is no market segmentation.
4. Generalized Delay Cost Structure
In 3 we found that under the additive delay cost
structure, there is no market segmentation. In this
section we show that under the generalized delay
cost structure, we may obtain value-based market seg-
mentation, where one S serves customers with higher
service valuations (the high-end segment) and the
other S serves customers with lower service valua-
tions (the low-end segment). In this equilibrium, the
high-end S offers a smaller delay but at a higher
price than the low-end S. Unlike in 3, high-value
customers strictly prefer the high-end S and low-
value customers strictly prefer the low-end S. More-
over, if the rms have different capacity costs, only an
equilibrium with value-based market segmentation is
possible.
The generalized delay cost specication allows the
delay cost of a customer to depend on his or her
service valuation. As described in the introduction,
such dependence is common in a variety of settings.
Under the generalized delay cost structure, the utility
derived by a customer with service valuation U and
delay | is (|) U D(|), where (|) is a decreasing
function that deates the customers service valuation
U, and D(|) is the additive component of the delay
cost. This means that the customers service valua-
tion and delay cost are interdependent. Each customer
with service valuation U maximizes his or her own
expected utility; i.e., ||u(U, |)] =||(|) U D(|)].
We assume for the most part that (|) is linear and
decreasing, (|) = 1 . |. Under the linear speci-
cation, a customer with service valuation U has util-
ity ||u(U, |)] = U (1 .W) dW, where W is the
expected delay in the system. Hence, the customer
incurs a delay cost of (.U d) W, where . > 0 is
a multiplicative coefcient and d 0 is an additive
coefcient of the delay cost. The demand curve (4)
becomes
P =V
/
(\) (1 . W) d W. (16)
Similar to Equation (8), if only S

is in the market, it
solves
max
\

, j

|\

c\

]
s.t. V
/
(\

) (1 . W

) d W

=P

.
(17)
When both rms are in the market, we structure the
game as a Cournot model, as before, where in the rst
stage S

decides on j

and \

( = 1, 2), and in the


second stage the customers decide which S to join,
if any. In 3.3 we showed that under the additive
delay cost structure, there was only a mixed-strategy
equilibrium; i.e., there was a service valuation

U such
Afanasyev and Mendelson: Service Provider Competition
222 Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS
that customers with service valuations above

U were
randomly allocated between the two rms, and those
below

U were not served. Another type of equilib-
rium is the value-based market segmentation equilibrium,
where customers are divided into three groups based
on their service valuations. The lowest value group
receives no service, the highest value group is served
by one S, and the intermediate group is served by the
other S. Formally, this means that

U
1
>

U
2
exist such
that a customer with service valuation U (a) joins S

,
if U

U
1
, (b) joins S

, if U |

U
2
,

U
1
), or (c) does not
join either S, if U -

U
2
.
Hence, all customers with service valuations above

U
1
form the high-end segment, served by S

, and
all customers with service valuations in the interval
|

U
2
,

U
1
) form the low-end segment, served by S

.
In our model, the only equilibrium structures pos-
sible are the mixed-strategy and value-based market
segmentation equilibria that we identied. We have
already shown in 3.3 that for . =0, we obtain only
the mixed-strategy equilibrium under a duopoly. If
for . >0 some customers with different valuations U

and U
8
are indifferent between S
1
and S
2
, then
U

(.U

d)W
1
P
1
=U

(.U

d)W
2
P
2
, and
U
8
(.U
8
d)W
1
P
1
=U
8
(.U
8
d)W
2
P
2
.
(18)
Then, by subtracting the two equations, we obtain that
W
1
= W
2
, which also implies that P
1
= P
2
. Thus, any
customer U
C
will also be indifferent between the two
providers. Similarly, if customers with valuations U

and U
8
strictly prefer S

, then any customer with valu-


ation U
C
(U

, U
8
) will also prefer S

. Indeed, if U
C

(U

, U
8
), then there exists an 0 -o -1 such that U
C
=
oU

(1 o)U
8
. Because U

and U
8
strictly prefer S

,
U

(.U

d)W

>U

(.U

d)W

, and
U
8
(.U
8
d)W

>U
8
(.U
8
d)W

.
(19)
Giving the two equations weights o and (1 o),
respectively, and summing up, we obtain that
U
C
(.U
C
d)W

>U
C
(.U
C
d)W

. (20)
We thus conclude that only our mixed-strategy
or value-based market segmentation equilibria are
possible.
We next derive the structure of the segmented equi-
libria for a market with asymmetric providers and a
generalized delay cost structure with . > 0 (symmet-
ric providers are considered later in this section). We
show that in such a market, there is no equilibrium
without value-based market segmentation; i.e., among
the customers who are served, those with service val-
uations above a threshold

U
1
choose the fast S, and
those with service valuations below that threshold but
above another threshold

U
2
choose the slow S. We
focus on the equilibrium in which the lower-cost S
1
serves the high-value segment
9
and show that for suf-
ciently small delay sensitivities, both Ss are in the
market and the market is segmented.
Proposition 2. There exist nested sets, with (0, 0)
+
1
, +
1
+
2
+
3
+
4
in (., d) space such that the fol-
lowing applies.
(1) Duopoly solution: For (., d) +
1
and . > 0, there
is an equilibrium with both rms in the market, where S
1
serves the high-end customers and S
2
serves the low-end
customers. The capacity j

and the arrival rates \

to each
S

are given by the solution to the following equations.

(\
1
r
2
j
1
)j
1
(j
1
\
1
)
2

\
1
r
2
j
1
\
1
1

1
2j
2
1
=
g
1
\
1
|V
/
(\
1
). d]
,
|V
/
(\
1
) \
1
V
//
(\
1
)].(W
2
W
1
)
|V
/
(\
1
\
2
) \
1
V
//
(\
1
\
2
)](1 .W
2
)
dW
1
\
1
|V
/
(\
1
). d]
r
2
1
2(j
1
\
1
)
2
=c
(21)
and

(\
2
r
2
j
2
)j
2
(j
2
\
2
)
2

\
2
r
2
j
2
\
2
1

1
2j
2
2
=
g
2
\
2
|V
/
(\
1
\
2
). d]
,
V
/
(\
1
\
2
) \
2
V
//
(\
1
\
2
)(1 .W
2
) dW
2
|V
/
(\
1
\
2
). d]
r
2
1
2(j
2
\
2
)
2
=c,
(22)
where W

are given by (5).


9
Under certain conditions, an equilibrium exists in which the
higher-cost S, S
2
, serves the high-value segment. It can be shown
that whenever such an equilibrium exists, there always exists an
equilibrium in which the lower-cost S serves the high-end segment.
Afanasyev and Mendelson: Service Provider Competition
Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS 223
(2) Single monopoly solution: For (., d) (+
2
+
1
)
(+
4
+
3
), there is an equilibrium with only S
1
in the mar-
ket. The equilibrium capacity j
1
and market size \
1
of S
1
are given by the solution to the following equations.

(\
1
r
2
j
1
)j
1
(j
1
\
1
)
2

\
1
r
2
j
1
\
1
1

1
2j
2
1
=
g
1
(V
/
(\
1
). d)\
1
,
|V
/
(\
1
) \
1
V
//
(\
1
)](1 .W
1
) dW
1
\
1
|V
/
(\
1
). d]
r
2
1
2(j
1
\
1
)
2
=c,
(23)
where W
m
is given by (5).
(3) Multiple monopoly solutions: For (., d) +
3
+
2
,
there are two equilibria: one with only S
1
in the market,
and one with only S
2
in the market. In the equilibrium with
S
m
in the market (m=1, 2), the equilibrium capacity j
m
and market size \
m
of S
m
are given by the solution to the
following equations.

(\
m
r
2
j
m
)j
m
(j
m
\
m
)
2

\
m
r
2
j
m
\
m
1

1
2j
2
m
=
g
2
(V
/
(\
m
). d)\
m
,
|V
/
(\
m
) \
m
V
//
(\
m
)](1 .W
m
) dW
m
\
m
|V
/
(\
m
). d]
r
2
1
2(j
m
\
m
)
2
=c,
(24)
where W
m
is given by (5).
(4) Market is not served: For (., d) ; +
4
, there is no S
in the market and \
1
=\
2
=j
1
=j
2
=0.
Proposition 2 shows that for small values of the
delay sensitivitiesi.e., in the set +
1
where we obtain
the duopoly solutionthere is a value-based mar-
ket segmentation equilibrium, where S
1
serves the
high-end segment and S
2
serves the low-end segment,
and customers are not indifferent between the two
providers. Market segmentation requires delay costs
that are low enough to support two rms in the mar-
ket. As the delay sensitivities increase, so (., d) is in
the set +
2
+
1
(the single monopoly solution region),
the only possible equilibrium is one with S
1
being
a monopoly. In this set, the lower-cost S enters and
crowds out the higher-cost S. As the delay sensitiv-
ities grow furtheri.e., in the set +
3
+
2
(the multi-
ple monopoly solutions region)two equilibria exist
with either of the Ss being a monopoly. This hap-
pens because as the delay sensitivities increase, S

( = 1, 2) cannot be protable if it enters the market


with S

already there. As . and d increase further, S


2
cannot survive even as a monopoly. Because of its cost
advantage, S
1
can survive for larger values of . and d,
but ultimately it cannot be protable and leaves the
market, too, so the market is not served. Note that
Proposition 2 allows for multiple equilibria, e.g., for
(., d) +
1
, there may also be monopoly solutions, and
for (., d) +
3
+
2
there are two monopoly equilibria.
Corollary 1. For . > 0 and g
1
- g
2
, any duopoly
equilibrium must be a value-based market segmentation
equilibrium.
Corollary 1 shows that market segmentation is
driven by the generalized delay cost structure as
well as by the rms cost asymmetry. Because the
providers capacity costs are different, we expect them
to choose different delays, and, because . >0, differ-
ent customers value the delays differently.
Similar to what we found in Proposition 1, the set
+
1
consists of up to three subsets:
(i) +
D
1
, where only duopoly equilibria exist;
(ii) +
A
1
+
D
1
, where duopoly and S
1
monopoly equi-
libria exist; and
(iii) +
1
+
A
1
, where duopoly and two, S
1
or S
2
,
monopoly equilibria exist.
We may also have in +
1
several duopoly equilibria.
For example, for sufciently small g
2
g
1
, there is
an equilibrium where S
2
serves the high-end segment
and S
1
focuses on the low-end segment. It can be
shown that if we have an equilibrium with the higher-
cost S serving the high-end segment, there is always
an equilibrium with the lower-cost S serving the high-
end segment.
Importantly, the structure of the equilibria obtained
under the generalized delay cost structure (Proposi-
tion 2) is substantively different from the structure
we obtained under the additive delay cost struc-
ture. In particular, the framework of an additive
delay cost structure does not adequately explain the
prevalence of value-based segmentation in markets
ranging from transportation to mortgage loan origi-
nation. With generalized delay costs, one of the rms
serves the higher-end segment and the other serves
Afanasyev and Mendelson: Service Provider Competition
224 Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS
Figure 1 Distortions in Total Industry Prots (Solid Line, Left Axis)
and the Difference Between the Capacity of the High-End
and Low-End Service Providers (Broken Line, Right Axis) in
the Duopoly Equilibrium as Functions of the Multiplicative
Delay Sensitivity t
0 1 2 3 4
15
15
10
5
0
D
i
s
t
o
r
t
i
o
n

i
n

t
o
t
a
l

p
r
o
f
i
t
s

(
%
)
D
i
s
t
o
r
t
i
o
n

i
n

c
a
p
a
c
i
t
y

d
i
f
f
e
r
e
n
c
e

(
%
)
20
20
25
25
30
35
40
V
Notes. The distortions are caused by assuming that the delay cost is addi-
tive and the average delay cost remains the same. The distortion in total
prots is the change in total prot, I
1
I
2
, caused by not accounting for
the multiplicative structure of the delay sensitivity relative to the correct total
prots. The distortion in capacity differences is the change in capacity dif-
ference (p
1
p
2
) caused by not accounting for the multiplicative structure
of the delay sensitivity relative to the correct capacity difference. For both
line graphs, the service valuations u are uniformly distributed over the inter-
val |0, 100] and the market size is A =300 (which corresponds to l
/
(') =
100 ',3),
1
=29,
2
=31, c =5, and J =0.
the lower-end segment, resulting in a natural value-
based segmentation. In a market characterized by
multiplicative delay sensitivity, assuming an additive
delay cost structure leads to signicant deviations not
only in qualitative structure but also in the quanti-
tative results. Clearly, these deviations are increasing
in .. However, as shown in Figure 1, the distortion in
the capacity differences between the service providers
tends to plateau as . increases, whereas the distor-
tion in total industry prot increases as long as both
service providers are in the market.
Our results link the structure of delay costs to the
existence of service differentiation or market segmen-
tation. Quantitatively, they show how service differ-
entiation and market segmentation depend on the
additive (d) and multiplicative (.) components of
delay cost. Our analysis shows how increasing each
of the two components of delay cost affects market
behavior. To translate this into empirically meaningful
results, we need concrete measures of service differ-
entiation and market segmentation.
In the context of our model, service differentiation
means that customers experience different expected
delays and pay different prices, depending on the
provider that serves them. Thus, two natural mea-
sures of service differentiation are the price difference,
AP =P
1
P
2
, and the differences in expected delays,
AW = W
2
W
1
, between the two providers. Both of
these measures can be empirically estimated in the
marketplace, and each exhibits different behavior as
a function of the delay sensitivities.
Consider rst AP =P
1
P
2
. Figure 2(a) shows that
AP is an increasing function of .. The monotone
relationship between AP and . is not surprising: as
. increases, the rms both suffer from lower demand,
but S
1
also benets from higher differentiation, which
gives it greater market power. One might expect AP to
be an increasing function of d (as would indeed be the
case when . =0); however, as shown in Figure 2(b), it
is a quasiconvex. This is the net result of two opposite
effects: on the one hand, AP decreases in d because its
component V
/
(\
1
).(W
2
W
1
) is a decreasing function
of d; on the other hand, its component d(W
2
W
1
) is
an increasing function of d.
Next consider AW =W
2
W
1
. Although a decreas-
ing relationship between AW and d is not surprising
(an increase in d affects customers proportionally to
the expected delay), it is hard to predict the shape of
the relationship between AW and .. Numerical exper-
iments show that AW is a quasiconvex function of ..
As . increases, all customers are willing to pay less
for the service, so the returns on investments in excess
capacity declinethe services become more similar
and AW tends to decrease. In contrast, an increase in
. makes customers more sensitive to the difference in
delay between the two providers, which is benecial
to the high-end S. It is straightforward to prove that
for sufciently small ., AW is an increasing function
of ., and for sufciently large ., AW is a decreasing
function of ..
Empirically, this means that although AW is a
monotone increasing function of d, AP is quasicon-
vex, and though AP is a monotone increasing function
of ., AW is a quasiconcave function of d.
We next measure market segmentation by how
much the average customer in the high-value segment
loses if she switches from S
1
to S
2
. The average con-
sumer of S
1
has surplus V
/
(\
1
,2)(1 .W

) dW

Afanasyev and Mendelson: Service Provider Competition


Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS 225
Figure 2 The Effects of Increasing the Multiplicative (a) and
Additive (b) Delay Sensitivities on the Difference in
Service Fees
0 0.04 0.08 0.12 0.16
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
S
1

S
2
S
1

S
2
v
(a) Difference in service fees P
1
P
2
as a function of v for d = 1
0 5 10 15
0.42
0.43
0.44
0.45
0.46
0.47
d
(b) Difference in service fees P
1
P
2
as a function of d for v = 0.1
Notes. In these gures, t is the multiplicative delay sensitivity, J is the addi-
tive delay sensitivity, and F
1
F
2
is the difference in service fees between
the providers, which is an alternative measure of service differentiation. In
both gures, the service valuations u are uniformly distributed over the inter-
val |0, 100] and the market size is A =300 (which corresponds to l
/
(') =
100 ',3),
1
=29,
2
=31, and c =5.
if served by S

. The reduction in consumer surplus


if she switches to S
2
equals (V
/
(\
1
,2). d)(W
2
W
1
)
(P
1
P
2
). Because the marginal customer (at \
1
) is
indifferent between the two providers, (V
/
(\
1
). d)
(W
2
W
1
) (P
1
P
2
) = 0, which means that the loss
for the average customer is
6 =.

V
/

\
1
2

V
/
(\
1
)

(W
2
W
1
). (25)
Incentive compatibility means that 6 0. Indeed,
the loss from switching is commonly used in stud-
ies of incentive compatibility in queueing systems
(see Mendelson and Whang 1990, Yahalom et al.
2006), and similar differences are used in the mar-
keting context (e.g., Kalish and Nelson 1991, Werten-
broch and Skiera 2002, Chung and Rao 2003, Jedidi
et al. 2003, Wang et al. 2003). Value-based market seg-
mentation and service differentiation are not indepen-
dent, as 6 is proportional to AW. However, when
there is no interdependence between the service val-
uations and delay costs, i.e., when . = 0, we have
AW > 0 and 6 = 0. Thus, positive service differenti-
ation is a necessary but not sufcient condition for
positive market segmentation.
The level of market segmentation 6 is an increas-
ing function of .. For small ., an increase in . both
increases customers sensitivity to delay and the level
of service differentiation; hence 6 increases as well.
For large ., service differentiation starts declining
but this effect is still weaker than that of the increase
in customers sensitivity to delay (because for large .,
as . increases the number of customers served by S
1
decreases; hence the service valuation of the average
customer, \
1
,2, increases).
We next consider symmetric rms and the resulting
market structure for sufciently small values of the
delay sensitivities.
Proposition 3. When . > 0 and g
1
= g
2
= g, there
exists a set

+ in (., d) space
10
in which there are three
equilibria with both rms in the market.
(1) Mixed strategy: Each S

chooses capacity j

and
market size \

satisfying

(\

r
2
j

)j

(j

)
2

\

r
2
j

1
2j
2

=
g
\

|V
/
(\
1
\
2
). d]
V
/
(2\) \V
//
(2\)(1 .W) dW
|V
/
(2\). d]
r
2
1
2(j\)
2
=c,
(26)
where W

are given by (5).


10
The set

+ is analogous to +
1
in Proposition 2.
Afanasyev and Mendelson: Service Provider Competition
226 Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS
(2) Equilibria with value-based market segmentation:
There are two equilibria where S

serves the high-end seg-


ment and S

serves the low-end segment, where j

and \

satisfy

(\

r
2
j

)j

(j

)
2

\

r
2
j

1
2j
2

=
g
\

|V
/
(\

). d]
|V
/
(\

) \

V
//
(\

)].(W

)
|V
/
(\

) \

V
//
(\

)](1 .W

)
dW

|V
/
(\

). d]
r
2
1
2(j

)
2
=c
(27)
and

(\

r
2
j

)j

(j

)
2

\

r
2
j

1
2j
2

=
g
\

|V
/
(\

). d]
V
/
(\

) \

V
//
(\

)(1 .W

) dW

|V
/
(\

). d]
r
2
1
2(j

)
2
=c,
(28)
where W

and W

are given by (5).


Proposition 3 shows that when the capacity costs
are equal, there are three equilibria: two equilib-
ria with value-based market segmentation and one
mixed-strategy equilibrium. In the last equilibrium,
the rms have the same capacity and delay, and cus-
tomers are indifferent between them.
5. Effect of Market Size
In this section we study how the size of the market,
measured by the total arrival rate A, affects the mar-
ket structure and the level of competition between the
two rms. As discussed in the literature review, Chen
and Wan (2003, 2005) found that the effect of market
size on the market structure is quite surprising and
counter-intuitive. It is thus interesting to study the
effect of market size in our model vis--vis the mod-
els of Chen and Wan. In addition, we are interested
in the effects of increasing market size on service dif-
ferentiation and market segmentation.
The following proposition studies the effects of
increasing A on the market structure.
Proposition 4. For any pair (., d), there exist con-
stants 0 A
1
A
2
A
3
A
4
A
5
such that the follow-
ing applies.
(1) No S in the market: For A - A
1
, no S serves the
market.
(2) S
1
monopoly: For A |A
1
, A
2
), there exists only a
monopoly equilibrium with S
1
in the market.
(3) Two monopoly equilibria: For A |A
2
, A
3
), there
exist two monopoly equilibria with either S in the market.
(4) S
1
monopoly or duopoly equilibria and two monopoly
equilibria: For A |A
3
, A
4
), there exists a threshold value
Ag

such that (i) if g


2
g
1
> Ag

, there exists only an


equilibrium with S
1
being a monopoly; and (ii) if g
2
g
1

Ag

, there exist duopoly equilibria with both rms in the


market as well as two monopoly equilibria with either S in
the market.
(5) Duopoly equilibria and S
1
monopoly: For A
|A
4
, A
5
), there exist duopoly equilibria with both rms
in the market as well as an equilibrium with S
1
being a
monopoly.
(6) Duopoly equilibria: For A A
5
, there exist only
duopoly equilibria with both rms in the market. The
arrival rates and S capacities are given by Proposition 1
( for . =0) and 2 ( for . >0).
In Proposition 4 we nd an interesting sequence
of market equilibria, which is illustrated in Figure 3.
Potentially, there are seven regions:
Region (1): The market is so small that it cannot
support any S, so the market is not served.
Region (2): A is so small that the market can sup-
port only the lower-cost S, S
1
. Thus, the only equilib-
rium has S
1
as a monopoly.
Region (3): The market is large enough to support
either of the rms as a monopoly but too small to
allow another entrant to have nonpositive prots,
so there are two monopoly equilibria.
Region (4a): The market is large enough to support
either of the rms as a monopoly, but it is sufciently
large that when S
2
is in the market, S
1
can enter and
make positive prots. However, when S
1
enters, it
crowds out the higher-cost S and the only equilibrium
is S
1
as a monopoly. Note that this is possible only
when S
1
s capacity cost is low enough in relation to
S
2
s.
Region (4b): The market is large enough for both
rms to be in the market. However, if one of the
rms is in the market as a monopoly, the other rm
Afanasyev and Mendelson: Service Provider Competition
Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS 227
Figure 3 The Effect of Increasing the Total Market Size A for
Different Levels of the Unit Capacity Cost Differential
on the Market Structure
50 100 150 200 250 300
0
5
10
15
g
2

g
1

(6) (4b)
(5) (3)
(4a)
(1)
(2)
Notes. Region (1): The market is not served. Region (2): Only $
1
is in the
market, as a monopoly. Region (3): Two monopoly equilibria, one with $
1
in
the market and one with $
2
in the market. Region (4a): Only $
1
is in the mar-
ket, as a monopoly. Region (4b): Duopoly and two monopoly equilibria, one
with $
1
and the other with $
2
in the market. Region (5): Duopoly equilibria
and a monopoly equilibrium with $
1
in the market. Region (6): Only duopoly
equilibria. In the gure, the service valuations u are uniformly distributed
over the interval |0, 100], which corresponds to l
/
(', A) =100 (100,A)',

1
=30, c =5, t =0.3, and J =100.
cannot make positive prot. Hence, this region sup-
ports either a duopoly or two monopoly equilibria.
This region exists only if the capacity cost difference
(g
2
g
1
) is small.
Region (5): The market is large enough for the
rms to earn positive prots if both are in the mar-
ket. Only S
1
can earn positive prot as a monopoly
and deter S
2
from entering the market. However, if S
2
attempts to operate as a monopoly, S
1
enters the mar-
ket. Thus, the only equilibria are duopoly and S
1
is a
monopoly.
Region (6): The market is large enough that both
rms can make positive prots. Hence, only duopoly
equilibria exist.
Overall, market segmentation occurs when the mar-
ket is large enough and the difference between the
capacity costs of the two rms is not too large. Value-
based market segmentation is the only equilibrium
in Region (6) of Proposition 4, which corresponds to
larger values of A and smaller values of g
2
g
1
, and
it is one of the equilibria in Regions (4b) and (5),
where both A and g
2
g
1
have intermediate values.
In summary, larger market sizes and smaller differ-
ences in the rms capacity costs are conducive to
value-based market segmentation.
Next, we show that an increase in the market size
decreases both service differentiation and value-based
market segmentation, and at the limit both disap-
pear. Figure 4 shows the effects of increasing the
market size on the market share and prot share
of the high-end provider, S
1
, and on our measures
of service differentiation and market segmentation.
First, an increase in market size benets the low-
end provider, S
2
, more than it benets the high-end
provider, S
1
, in terms of both market share (Fig-
ure 4(a)) and prot share (Figure 4(b)). This is because
of scale economies in queueing systems: A facility
with few customers benets from an additional cus-
tomer more than a facility with many customers. As
S
2
serves fewer customers than S
1
, an increase in the
total market size decreases the cost of S
2
more than it
decreases the cost of S
1
. Hence, S
2
benets more from
an increase in A than S
1
does. Figure 4(c) shows that
market segmentation decreases as A increases. Ser-
vice differentiation decreases because a given increase
in A has a larger incremental effect on S
2
than on
S
1
; hence the difference between the two decreases.
Because market segmentation is increasing in our dif-
ferentiation measure AW = W
2
W
1
, our measure of
segmentation (25) decreases. At the limit of a very
large market, both service differentiation and mar-
ket segmentation disappear. This result is driven by
the fact that as the number of customers increases,
the excess capacity increases at a slower rate than
the equilibrium demand rate \

. Because the marginal


cost of excess capacity decreases, the rms reduce the
waiting times to zero at the limit. Formally, we have
as follows.
Proposition 5. lim
A~
W

(A) =0 for =1, 2.


Corollary 2. As A~, the measures of both ser-
vice differentiation and market segmentation tend to zero.
6. Different Unit Operating Costs
So far, we have assumed that the service providers
differ only in their unit capacity costs but not in
their unit operating costs. Do different unit operat-
ing costs lead to value-based market segmentation?
In this section we allow the unit operating costs of the
Afanasyev and Mendelson: Service Provider Competition
228 Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS
Figure 4 Effect of Increasing the Total Market Size A on the Market Share of $
1
(a), the Prot Share of $
1
(b), and Market Segmentation (c)
500 600 700 800 900 1,000
0.5258
0.5260
0.5262
0.5264
0.5266
0.5268
0.5270
0.5272
(a) Market share of S
1
as a function of
the market size
500 600 700 800 900 1,000
0.5530
0.5535
0.5540
0.5545
0.5550
0.5555
0.5560
0.5565

1
/
(

1
+

2
)

1
/
(

1
+

2
)
(b) Profit share of S
1
as a function of
the market size

500 600 700 800 900 1,000
0.034
0.036
0.038
0.040
0.042
0.044
0.046
0.048

(c) Market segmentation as a function


of the market size
Notes. In these gures, A is the total arrival rate of customers, '
1
,('
1
'
2
) is the proportion of customers served by $
1
, I
1
,(I
1
I
2
) is the proportion
of prots received by $
1
, and is our measure of market segmentation. The service valuations u are uniformly distributed over the interval [0,100] (which
corresponds to l
/
(', A) =100 (100,A)'),
1
=29,
2
=31, c =5, t =0.1, and J =1.
two rms, c
1
and c
2
, to be different. We also change
assumptions (6) and (7) so that V
/
(0) >max{c
1
, c
2
] g
and there exists a \ such that V
/
(\) -min{c
1
, c
2
] g,
respectively.
First, we show that even if the rms capacity costs
are equal, differences in unit operating costs lead to
value-based market segmentation. Then, we numeri-
cally analyze how the differences in the capacity and
operating costs affect the levels of service differentia-
tion and market segmentation.
Proposition 6. Let . > 0, g
1
= g
2
= g, and c
1
- c
2
,
and let both rms be in the market. Then in equilibrium,
S
1
serves the high-end customers and S
2
serves the low-end
customers. The capacity j

and arrival rates \

to each S

satisfy

(\
1
r
2
j
1
)j
1
(j
1
\
1
)
2

\
1
r
2
j
1
\
1
1

1
2j
2
1
=
g
\
1
|V
/
(\
1
). d]
,
|V
/
(\
1
) \
1
V
//
(\
1
)].(W
2
W
1
)
|V
/
(\
1
\
2
) \
1
V
//
(\
1
\
2
)](1 .W
2
)
dW
1
\
1
|V
/
(\
1
). d]
r
2
1
2(j
1
\
1
)
2
=c
1
(29)
Afanasyev and Mendelson: Service Provider Competition
Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS 229
and

(\
2
r
2
j
2
)j
2
(j
2
\
2
)
2

\
2
r
2
j
2
\
2
1

1
2j
2
2
=
g
\
2
|V
/
(\
1
\
2
). d]
,
V
/
(\
1
\
2
) \
2
V
//
(\
1
\
2
)(1 .W
2
) dW
2
|V
/
(\
1
\
2
). d]
r
2
1
2(j
2
\
2
)
2
=c
2
,
(30)
where W

are given by Equation (5).


Corollary 3. For . > 0, g
1
= g
2
= g, and c
1
- c
2
,
there is no duopoly equilibrium without value-based market
segmentation.
Proposition 7. For a 1-unit increase in the capac-
ity cost g

of S

( = 1, 2), to keep prot constant, the


marginal operating cost c

should decrease by j,\ units


for a monopoly and by (1, j,\) units for a duopoly.
Proposition 7 shows that a unit change in capac-
ity costs impacts each service providers prots more
than a unit change in its operating cost. From the
proof of the proposition, for a monopoly oH,oc =\,
i.e., the change in c affects the prots proportionally to
the arrival rate, whereas oH,og =j, i.e., the change
in g affects prots proportionally to capacity, which
is always larger than the arrival rate. Proposition 6
implies that other things being equal, a unit reduction
in capacity cost is worth to each S more than a unit
reduction in its operating cost.
7. Concluding Remarks
Recent years have seen an increase in the impor-
tance of service industriesand with it, an increased
focus on timeliness. As the value of customers
time increases, prompt service becomes an important
lever for gaining competitive advantage. This calls
for a more detailed examination of the relationship
between the delays and competitive strategies of ser-
vice rms. Previous research on the economics of
queues often took an additive delay cost structure for
granted. We show that when customers service val-
uations and their delay costs are interdependent, and
the service providers have different costs, the market
exhibits value-based segmentation: one of the service
providers focuses on fast service for the high-value
Table 2 How the Equilibrium Market Structures Depend on the
Multiplicative Delay Sensitivity t and Capacity Costs

j
(j =1, 2)

1
=
2

1
-
2
t =0 No service differentiation Service differentiation
No market segmentation No market segmentation
t >0 Multiple equilibria:
No service differentiation Service differentiation
No market segmentation Market segmentation
or
Service differentiation
Market segmentation
segment, and the other targets the low-value segment
with a lower price and slower service. Table 2 sum-
marizes our key results: market segmentation (value-
based) occurs only under the generalized delay cost
structure, and for service providers with asymmetric
costs this is the only possible equilibrium.
We nd an interesting progression of market struc-
tures as a function of market size. In our model, for
a small arrival rate, the market is not served; for a
moderate total arrival rate, the market supports only
the lower-cost S being a monopoly; for a larger mar-
ket size, either S can be a monopoly; and as the mar-
ket size increases further, the only equilibrium has the
lower-cost S as being a monopoly or there are the
duopoly and two monopoly equilibria. As the arrival
rate increases further, the only equilibria possible are
the duopoly equilibria and the lower-cost S being a
monopoly. Finally, for a larger market size, the only
equilibria have both rms competing in the market.
A natural extension of our model is to consider pri-
ority queues. A service provider may assign a higher
priority to customers who choose to pay a higher ser-
vice fee and thus, these customers would be served
faster than those who pay less. We expect our qualita-
tive results (in particular, the existence of value-based
market segmentation) to hold in such a setting. For
a monopoly with two priority classes, all customers
with service valuations above some threshold would
prefer the higher-priority class, and all served cus-
tomers below this threshold would opt for the lower-
priority class. Yet the study of the duopoly case raises
the issue of how the number of priority classes affects
entry and exit decisions, which, to our knowledge,
has not been addressed in the literature.
Afanasyev and Mendelson: Service Provider Competition
230 Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS
Our analytical results were derived for only two
rms with the same coefcient of variation of the ser-
vice times. The duopoly assumption allowed us to use
the properties of supermodular games, and extending
the results to the oligopoly case requires for numeri-
cal analysis. Numerical explorations suggest that with
multiple rms and different coefcients of variation
of service times, a duopoly market is still character-
ized by value-based market segmentation, with each
of the rms serving an interval of customers with a
range of service valuations.
Our results show that the delay cost structure
makes both a structural and a practical difference in
the analysis of competing congestion-prone service
providers. Markets where the delay cost and the value
of service are interdependent are substantively differ-
ent from markets with an additive delay structure,
leading to different competitive strategies, customer
behavior, and industrial structures.
Acknowledgments
The authors thank the editor, associate editor, and anony-
mous reviewers for their suggestions, which helped to
improve the substance and the exposition of the paper.
Appendix. Proofs
Proof of Proposition 1. The proof proceeds along the
following steps: (1) We derive necessary and sufcient con-
ditions for the existence of an equilibrium with both rms
in the market for sufciently small d. (2) We nd the largest
d (d
1
) such that there is a duopoly solution in the entire
interval |0, d
1
]. (3) We show that for d slightly above d
1
,
only a monopoly solution is possible, and we derive nec-
essary and sufcient conditions for the monopoly equilib-
rium. (4) We show that there exist d
2
, d
3
, and d
4
such that
in equilibrium, any of the rms may be a monopoly for d
|d
2
, d
3
) and only S
1
is a monopoly for d |d
1
, d
2
) |d
3
, d
4
)
and that for d d
4
, none of the rms can earn positive
prots.
Step 1. By (2) and (6), for d =0, which reduces our model
to a standard duopoly, both rms earn positive prots in
equilibrium. Now consider some d slightly above zero.
We rst show that if we suppress the positive prot
requirement, a duopoly equilibrium exists. We then show
that for a small enough d, each duopolist makes posi-
tive prot. As shown in 3.3, the customers are indifferent
between the rms and each rm solves the problem:
max
\

, j

|\

c\

]
s.t. V
/
(\
1
\
2
) d W

=P

,
(31)
where W

are given by (5).


Let H

(\

, j

, \

) = \

|V
/
(\
1
\
2
) d W

c] g

be the prot function of player . From assumption (7)


and the nonnegative cross-partial derivatives of the prot
functions, game (31) is supermodular in (\
1
, j
1
, \
2
, j
2
)
and a duopoly equilibrium exists (by Theorems 4 and 5 in
Milgrom and Roberts 1990).
We next show that for a small enough d, both duopolists
make positive prots. Theorem 2 in Milgrom and Segal
(2002) shows that H

( =1, 2) are continuous functions of d.


By the continuity of H

(d), H

(0) > 0 implies that for suf-


ciently small d, H

(d) >0 as well. Summing up, if d is small


enough, there is a duopoly equilibrium with both rms
making positive prots. We denote the equilibrium arrival
rate to S

when the delay sensitivity is d by \

(d).
11
According to Theorem 6 in Milgrom and Roberts
(1990) with exogenous parameter g
1
, \
1
\
2
and j
1
j
2
.
This implies that in equilibrium, H
1
(d) H
2
(d) when
both rms are in the market. By our choice of d, both
rms are in the market. Because prot goes to ~ as
\

~, there is an equilibrium where \

and j

satisfy
the rst-order conditions (FOC) of (31). From Equa-
tion (5), oW

,o\

= (r
2
1),(2(j

)
2
) and oW

,oj

=
|(\

r
2
j

)j

,(j

)
2
\

r
2
,(j

) 1](1,(2j
2

)),
which together with the FOC of (31) give the equilibrium
condition (12) for the duopoly.
Step 2. Because of (7), there exists at least one nite d
such that \
2
(d) = 0. Let d
1
= inf{d [ \
2
(d) = 0]. Because
H
1
(d) H
2
(d), S
1
is still in the market at d =d
1
and in |0, d
1
),
we have a duopoly solution with positive prots for both
rms.
Step 3. Consider d slightly above d
1
. Then, only one S,
which we call S
m
, can survive in the market (the other S will
earn negative prots if it enters the market). Then S
m
must
set its capacity j
m
and service rate \
m
to maximize its prot:
H
A
m
(d) = max
\
m
, j
m
|\
m
P
m
c\
m
g
m
j
m
]
s.t. V
/
(\
m
) dW
m
P
m
.
(32)
By the choice of d and because prot goes to ~ as
\
m
~, there is an interior solution that satises the FOC.
From (5), oW
m
,o\
m
=(r
2
1),(2(j
m
\
m
)
2
) and oW
m
,oj
m
=
|(\
m
r
2
j
m
)j
m
,(j
m
\
m
)
2
\
m
r
2
,(j
m
\
m
) 1]
(1,(2j
2
m
)), which together with FOC give the equilibrium
condition (14) for the monopoly.
Step 4. Let H

(d, \

) ( = 1, 2) be the optimal prot


in (31) for xed d and given \

. Let d
2
be the smallest solu-
tion to H

1
(d, argmax
\
H
A
2
(\, d)) = 0. Because when both
rms are in the market, H
1
(d) H
2
(d), d
1
d
2
. Let d
3
be the
smallest solution to H
A
2
(d) = 0, and let d
4
be the smallest
solution to H
A
1
(d) = 0. Because H

1
(d, 0) = max
\
H
A
1
(\, d),
d
2
d
3
. Thus, in the interval |d
2
, d
3
) there are two equilibria,
11
If there are several equilibria, we choose the equilibrium with the
maximum number of customers for S
2
.
Afanasyev and Mendelson: Service Provider Competition
Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS 231
with either S being a monopoly, and in the interval |d
1
, d
2
)
there is only a monopoly S
1
solution. Because g
1
- g
2
, for
any value of d, H
A
1
(d) H
A
2
(d). Hence, d
3
d
4
. Because the
monopoly prot is a decreasing function of d, for d |d
3
, d
4
)
only S
1
can be a monopoly, and for d d
4
, the market is not
served.
Proof of Proposition 2. We rst show in the Prelimi-
nary Step below that when both rms are in the market,
only an equilibrium with value-based market segmentation
is possible. The proof then follows the same steps as the
proof of Proposition 1.
Preliminary Step. Consider a duopoly equilibrium and
assume, by contradiction, that it follows our mixed-strategy
structure. We show that this assumption leads to the contra-
diction that both \
1
\
2
and \
1
- \
2
hold simultaneously.
In our mixed-strategy equilibrium, each S

, =1, 2 solves
max
\

, j

|\

c\

]
s.t. V
/
(\
1
\
2
)(1 . W

) d W

=0.
(33)
Let H

(\

, j

, \

) =\

|V
/
(\
1
\
2
)(1 . W

) d W

c]
g

be the prot function of player given \

. Game (33)
is supermodular in (\
1
, j
1
, \
2
, j
2
) by Theorem 4 in
Milgrom and Roberts (1990) and the nonnegative cross-
partial derivatives of the prot functions. Next, the cross-
partial derivatives of H

with respect to the strategy


variables of S

, =1, 2, and g
1
are nonnegative. Hence, by
Theorem 6 in Milgrom and Roberts (1990), with exogenous
parameter g
1
,
\
1
\
2
. (34)
We next show that \
1
-\
2
. As shown in 4, in a mixed-
strategy equilibrium, W
1
=W
2
=W. Let j

=\

,j

. By (5),
\
1
j
1
2

j
1
r
2
1
1 j
1
1

=\
2
j
2
2

j
2
r
2
1
1 j
2
1

=W. (35)
By the FOC of S

( =1, 2) for (33), we have


(V
/
(\
1
\
2
). d)

W
j
2

r
2
1
2(1 j

)
2

=
g

, (36)
hence for =1, 2,
\

=j
3

V
/
(\
1
\
2
). d
g

(V
/
(\
1
\
2
). d)W

r
2
1
2(1 j

)
2
. (37)
Equation (37) together with (36) leads to
W =


V
/
(\
1
\
2
). d
g

(V
/
(\
1
\
2
). d)Wj

j
2

r
2
1
4(1 j

)
2

r
2
1
1 j

. (38)
W in expression (38) is the product of three functions that
are increasing in j

. Thus, g
1
- g
2
implies j
1
- j
2
. For the
M/G/1 queue, the expected number of customers in the
system depends only on its utilization j and increases in j.
Therefore, by Littles Law, \
2
W =\
2
W
2
>\
2
W
1
=\
1
W, or
\
2
>\
1
, (39)
which contradicts (34).
Given the value-based market segmentation structure,
the rms determine j

and \

by solving
max
\
1
, j
1
|\
1
P
1
c\
1
g
1
j
1
]
s.t. |V
/
(\
1
). d](W
2
W
1
) P
1
P
2
(40)
and
max
\
2
, j
2
|\
2
P
2
c\
2
g
2
j
2
]
s.t. V
/
(\
1
\
2
)(1 .W
2
) dW
2
P
2
0,
(41)
where W

are given by Equation (5). Note that the constraint


in Equation (40) follows from Equations (19) and (20) and
the Preliminary Step, and that Equation (41) is analogous
to (10) for . >0.
We now proceed to prove the core of Proposition 2.
Step 1. By Assumptions (2) and (6), for . = d = 0 our
model is reduced to a standard duopoly with both rms
earning positive prots. Now consider some . slightly
above zero and d slightly above or equal to zero.
We will rst show that if we suppress the positive prot
requirement, a duopoly equilibrium exists and we will then
show that for a small enough d, each duopolist makes a pos-
itive prot. Each S

( =1, 2) decides on its capacity j

and
service rate \

, which maximize its prot given the strat-


egy of the other S (S

). The Preliminary Step shows that a


duopoly equilibrium must be characterized by value-based
segmentationthat is, one of the providers, S
1
or S
2
, serves
the higher-end segment and the other serves the lower-end
segment. The incentive compatibility and individual ratio-
nality constraints determine the service fees as
P

V
/
(\

).(W

) P

if W

-W

,
V
/
(\
1
\
2
)(1 .W

) dW

if W

.
(42)
From (3), the cross-partial derivatives of the payoff func-
tions in (40)(41) are nonnegative. Therefore, by Theorems 4
and 5 in Milgrom and Roberts (1990), the game is super-
modular in (\
1
, j
1
, \
2
, j
2
) and an equilibrium exists.
We now show that for . and d small enough, both
duopolists earn positive prots. Theorem 2 in Milgrom and
Segal (2002) shows that H

( = 1, 2) are continuous func-


tions of (., d). By the continuity of H

(., d), H

(0, 0) > 0
implies that for sufciently small (., d), H

(., d) > 0 as
well. Therefore, an equilibrium with both rms in the mar-
ket exists. We denote the equilibrium arrival rate to S

by
\

(., d).
According to Theorem 6 in Milgrom and Roberts (1990)
with exogenous parameter g
1
, \
1
\
2
, and j
1
j
2
. This
Afanasyev and Mendelson: Service Provider Competition
232 Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS
implies that in equilibrium, H
1
(., d) H
2
(., d) when both
rms are in the market. Indeed, if for some (., d), H
1
(., d) -
H
2
(., d), S
1
would increase its prots by moving from
(\
1
, j
1
) to (\
2
, j
2
). By our choice of (., d), both rms are
in the market. Because prot goes to ~ as \

~, there
is an equilibrium where the \

and j

satisfy the FOC with


W
1
W
2
(otherwise, P
1
P
2
and S
1
can increase its prot by
matching S
2
s offering).
To complete the proof, we show that S
2
does not deviate
to serve higher-end customers. Let \
c
1
, j
c
1
, \
c
2
, and j
c
2
be the
solution to (40) and (41), and let the corresponding delays
be W
c

( = 1, 2). If S
2
chooses its decision variables such
that W
2
W
c
1
, it serves all the customers above some value
threshold

U
2
, which determines its arrival rate

\
2
. Now, the
arrival rate

\
2
and the capacity j
2
solve
max
\
2
, j
2
|\
2
P
2
c\
2
g
2
j
2
]
s.t. |V
/
(\
2
). d](W
1
W
2
) P
2
P
1
,
(43)
where P

are given by (42) and W

are given by (5).


Because the prot function (43) is supermodular, the best
response function of S
2
serving the high-end segment,

\(\
1
, j
1
), decreases in the number of customers and capac-
ity of S
1
(Milgrom and Segal 2002, Theorem 4.2). Hence,

\
2
=

\(\
c
1
, j
c
1
)

\(\
c
2
, j
c
2
) = \
c
1
; i.e., S
2
serves fewer cus-
tomers than S
1
. Because the optimal delay decreases as the
number of customers increases (which follows from differ-
entiating the FOC for the high-end S with respect to its
arrival rate), the optimal delay of S
2
would be larger than
the optimal delay of S
1
; i.e., all high-end customers would
prefer S
1
over S
2
, a contradiction.
Step 2. Because of (7) and H
1
(., d) H
2
(., d), for every .
there is a nite d(.) such that \
2
(., d(.)) = 0. Let +
1
=
{(., d) [ \

1
(., d) > 0]. Because H
1
(., d) H
2
(., d), S
1
is still
in the market at such (., d), and in +
1
we have a duopoly
solution with positive prots for both rms.
Step 3. Consider (., d) slightly outside of +
1
. Then, only
one S, which we call S
m
, can be in the market (because the
other S will earn nonpositive prots if it enters the market).
Then, S
m
must set its capacity j
m
and probabilities \
m
to
maximize its prot:
H
A
m
(., d) = max
\
m
, j
m
|\
m
P
m
c\
m
g
m
j
m
]
s.t. V
/
(\
m
) (1 .W
m
) dW
m
P
m
.
(44)
Because we have chosen (., d) slightly outside of +
1
and
because prot goes to ~ as \
m
~, there is an interior
solution that satises the FOC. From Equation (5), oW
m
,o\
m
= (r
2
1),(2(j
m
\
m
)
2
) and oW
m
,oj
m
= |(\
m
r
2
j
m
)j
m
,
(j
m
\
m
)
2
\
m
r
2
,(j
m
\
m
) 1](1,(2j
2
m
)), which together
with the FOC gives the equilibrium condition (24).
Step 4. Let H

(., d, \

) ( = 1, 2) be the optimal prot


determined as a solution to FOC for xed (., d) and \

. Let
+
2
= {(., d): H

1
(., d) = argmax
\
H
A
2
(\, ., d) 0]. Because
then both rms are in the market, H
1
(., d) H
2
(., d),
+
1
+
2
.
Let +
3
= {(., d): H
A
2
(., d) 0]. Because H

1
(., d, 0) =
max
\
H
A
1
(\, ., d), +
2
+
3
. Thus, in the set +
3
+
2
there are
two equilibria with either S being a monopoly. Because
g
1
-g
2
, for any value of (., d),
H
A
1
(., d) = max
\, j
|\(V
/
(\)(1 .W) dW) c\g
1
j]
max
\, j
|\(V
/
(\)(1 .W) dW) c\g
2
j]
= H
A
2
(., d). (45)
Hence, +
3
+
4
. Because the monopoly prot is a decreas-
ing function of . and d, in the set +
4
+
3
, only S
1
can be a
monopoly, and for (., d) ;+
4
, the market is not served.
Proof of Corollary 1. Corollary 1 follows from the Pre-
liminary Step in the proof of Proposition 2.
Proof of Proposition 3. We rst show that there is
a symmetric mixed-strategy equilibrium. Then taking the
limits of the results in Proposition 2 shows that there
are two asymmetric equilibria with value-based market
segmentation.
To prove the existence of a mixed-strategy equilibrium,
it is sufcient to prove the existence of an equilibrium for
the game
max
\

, j

|\

c\

gj

]
s.t. V
/
(\
1
\
2
)(1 .W

) dW

=0, =1, 2,
(46)
where W

are given by (5).


By Theorems 4 and 5 in Milgrom and Roberts (1990)
and the nonnegative cross-partial derivatives of the payoff
functions, (46) is supermodular in (\
1
, j
1
, \
2
, j
2
) and
an equilibrium for the game (46) exists. By the choice
of (., d) such that both rms are in the market and by
assumption (7), there is an interior solution that satises the
FOC. Because in a mixed-strategy equilibrium customers
are indifferent between the rms, W
1
= W
2
and P
1
= P
2
.
Because in (38) with g
1
= g
2
= g, the right-hand side is a
strictly increasing function of j

, we must have j
1
=j
2
. This,
together with W
1
=W
2
, leads to \
1
=\
2
in the equilibrium.
Differentiating FOC w.r.t. . (d, respectively), by the symme-
try of the equilibrium and assumption (3), \

decreases in .
(d, respectively). This, together with assumption (7), implies
that the set

+ ={(., d) [ \

2
(., d) >0] is bounded.
Because prot is a continuous function of the capacity
costs, the set

+ of values (., d) where the segmented equi-
libria exist is determined as a limiting case of Proposition 2
with g
2
g
1
. Both

+ and

+ are connected and contain
(0, 0). Hence, the set

+ =

+

+ contains both mixed-
strategy and value-based market segmented equilibria.
Proof of Proposition 4. In Propositions 1 and 2
we have shown that the possible equilibria are as fol-
lows: duopoly, S
1
being a monopoly, and S
2
being a
Afanasyev and Mendelson: Service Provider Competition
Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS 233
monopoly. We now examine how the structure of the equi-
libria depends on A.
Let H

(A, \

) ( = 1, 2) be the optimal prot of S

for
xed A and \

, H
D

(A) ( =1, 2) be the equilibrium prot


of S

12
and H
A

(A) ( =1, 2) be the optimal prot in (44).


Let A
1
= max{A [ H
A
1
(A) 0]. By the envelope theorem,
H
A
1
(A) is an increasing function of A and by the cost advan-
tage of S
1
, we must have H
A
2
(A) H
A
1
(A) for AA
1
. Hence,
for AA
1
, either S obtains nonpositive prots and the mar-
ket is not served.
Let A
2
=max{A[ H
A
2
(A) 0]. By the cost advantage of S
1
,
A
1
A
2
and in A |A
1
, A
2
) the only viable equilibrium has
S
1
as a monopoly, with \
1
and j
1
given by solution to (23).
Let

A = min{A [ H

1
(A, argmax
\
H
A
2
(\, A)) 0],

A =
max{A [ H
D
2
(A) - 0], A
3
= min{

A,

A], and A
4
= max{

A,

A].
Because of the cost of advantage of S
1
, A
2


A. By the enve-
lope theorem, H
D
2
(A) and H
A
2
(A) are increasing functions
of A. Furthermore, because H
A
2
(A) H
D
2
(A) for any A, A
2

A. Therefore, A
2
A
3
. In the interval A |A
2
, A
3
), there are
no duopoly equilibria (by the choice of

A) and the entry
barrier is high enough to prevent entry if any of the rms
is a monopoly (by the choice of

A). Hence, for A |A
2
, A
3
),
there are only two equilibria, with either S, S
m
(m = 1, 2),
as a monopoly, where \
m
and j
m
are given by the solution
to (24).
If

A

A, in the interval A |A
3
, A
4
) S
1
enters the mar-
ket if S
2
operates as a monopoly, but S
2
cannot enter the
market if S
1
is a monopoly (by the choice of

A) and there
is no duopoly solution (by the choice of

A). Hence, in A
|A
3
, A
4
) only an equilibrium with S
1
being a monopoly, with
\
1
and j
1
given by the solution to (23), exists.
If

A>

A, in the interval A |A
3
, A
4
), the entry barrier is
high enough that either S can be a monopoly (by the choice
of

A) and a duopoly equilibrium is possible (by the choice
of

A). In these equilibria, \

and j

, = 1, 2, are given by
the solution to (24) for the monopoly and to (21)(22) for
the duopoly.
By the envelope theorem, H

1
(A, argmax
\
H
A
2
(\, A)) is an
increasing function of g
2
for xed A and g
1
(see Mas-Colell
et al. 1995, Theorem M.L.1.), and by Theorem 6 in Milgrom
and Roberts (1990), H
D
2
(A) is a decreasing function of g
2
.
Thus,

A decreases and

A increases in A. Hence, there is a
Ag

at which

A =

A. If (g
2
g
1
) > Ag

, we can have only


an equilibrium with S
1
being a monopoly for A |A
3
, A
4
).
If (g
2
g
1
) Ag

, we have the duopoly and two monopoly


equilibria in A |A
3
, A
4
). In each case, the rms arrival
rates and capacities are given by the solution to (24) for the
monopoly and to (21)(22) for the duopoly.
Let A
5
=min{A[ H

2
(A, argmax
\
H
A
1
(\, A)) 0]. Because
of the cost advantage of S
1
, H

1
(A, argmax
\
H
A
2
(\, A))
H

2
(A, argmax
\
H
A
1
(\, A)). Because the prots of both rms
12
If there are several duopoly equilibria, we consider one with S
1
serving the high-end segment.
increase in A, A
5


A. Because S
1
serves more customers as a
monopoly than as a duopoly, if H

2
(A, argmax
\
H
A
1
(\, A))
0, then H

2
0. Hence, A
5


A. Thus, A
5
A
4
and for any
AA
5
, there are no monopoly equilibria.
Proof of Proposition 5. Proposition 4 showed the exis-
tence and structure of the segmented equilibria for any
nite A. Let \

(A) be the expected equilibrium arrival rate to


S

as a function of A. We show that W

(A) 0 as A~: by
duopoly FOC and the relationship between () and V
/
(),
as A~,
lim
A~
oW

oj

=0 =1, 2. (47)
By (5),
oW

oj

(\

r
2
j

)j

\
2

r
j

1
2j
2

,
which together with (47) implies that
lim
A~

r
2
j

(j

)
2
j

=0.
Hence, W

0 as A~.
Proof of Corollary 2. Because W

0 as A ~, by
denition both AW and 6 go to zero as A~(see (25)).
Proof of Proposition 6.
Step 1. Similar to Proposition 2, we rst show that only
a value-based market segmentation equilibrium is possible.
The proof follows a similar logic to the Preliminary Step in
the proof of Proposition 2. Specically, the prot maximiza-
tion problem is formulated here with the decision variables
\

and j

=\

,j

as
max
\

, j

g
\

s.t. V
/
(\
1
\
2
)(1 . W

) d W

=0.
(48)
The cross-partial derivatives are nonnegative, and follow-
ing Theorem 4 in Milgrom and Roberts (1990), the game is
supermodular in (\
1
, j
1
, \
2
, j
2
). Equation (34) becomes
j
2
j
1
. (49)
Equation (36) becomes
V
/
(\
1
\
2
) \

V
//
(\
1
\
2
)(1 .W) (V
/
(\
1
\
2
). d)

W
j
2

r
2
1
2(1 j

)
2

=c

=1, 2. (50)
Equation (35) becomes
\

=
j

2W

r
2
j

1
1 j

=1, 2, (51)
which together with (50) leads to
V
/
(\
1
\
2
)
j

2W

r
2
j

1
1j

V
//
(\
1
\
2
)(1.W)
W(V
/
(\
1
\
2
).d)

1j

1
r
2
j

2j

r
2
1
(1j

=c

. (52)
Afanasyev and Mendelson: Service Provider Competition
234 Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS
We now examine the left-hand side (LHS) of (52) as a func-
tion of the variables (\
1
\
2
) and j

. Because W
1
=W
2
=W, if
the LHS increases in j

, then c
1
-c
2
implies j
1
-j
2
. Because
the functions (x,2)((xr
2
1),(1 x) 1) V
//
(\
1
\
2
) and
(1x(1,(xr
2
2 x))((r
2
1),(1 x))) are decreasing in x,
in the equilibrium,
j
1
-j
2
, (53)
which contradicts (49).
Step 2. Furthermore, similar to Step 1 in the proof of
Proposition 2, Equations (40) and (41) become
max
\
1
, j
1
|\
1
P
1
c
1
\
1
gj
1
]
s.t. |V
/
(\
1
). d](W
2
W
1
) P
1
P
2
(54)
and
max
\
2
, j
2
|\
2
P
2
c
2
\
2
gj
2
]
s.t. V
/
(\
1
\
2
)(1 .W
2
) dW
2
P
2
0.
(55)
The proof now follows the outline of that of Proposition 2
with the above substitutions.
Proof of Proposition 7. By the implicit function theo-
rem, for the monopoly problem (44), (oH
A
,oc) (oc,og)
oH
A
,og = 0. Because oH
A
,oc = \ and oH
A
,og =j,
we obtain oc,og = j,\. For a duopoly, the implicit func-
tion theorem leads to (oH

,oc

)(oc

,og

) (oH

,o\

)
((o\),og

) (oH

,oj

)(oj

,og

) oH

,og

= 0 for
S

( = 1, 2). Because (oH

,o\

)((o\),og

) (oH

,oj

)
(oj

,og

) - 0, we have (oH

,oc

)(oc

,og

) oH

,og

> 0.
From oH

,oc

=\

and oH,og

=j

, we obtain oc

,og

-
j

,\

. Because j

>\

, oc

,og

-1.
References
Adler, P., A. Mandelbaum, V. Nguyen, E. Schwerer. 1995. From
project to process management: An empirically-based frame-
work for analyzing product development time. Management
Sci. 41(3) 458484.
Afanasyev, M., H. Mendelson. 2009. Production and distribution
delays in the motion picture industry. Working paper, Stanford
University, Stanford, CA.
Afeche, P., H. Mendelson. 2004. Pricing and priority auctions in
queueing systems with a generalized delay cost structure. Man-
agement Sci. 50(7) 869882.
Allon, G., A. Federgruen. 2006. Competition in service industries
with segmented markets. Working paper, Columbia University,
New York.
Allon, G., A. Federgruen. 2007. Competition in service industries.
Oper. Res. 55(1) 3755.
Armony, M., M. Haviv. 2003. Price and delay competition between
two service providers. Eur. J. Oper. Res. 147(1) 3250.
Cachon, G., P. Harker. 2002. Competition and outsourcing with
scale economies. Management Sci. 48(10) 13141333.
Chen, H., Y.-W. Wan. 2003. Price competition of make-to-order
rms. IIE Transactions 35(9) 817832.
Chen, H., Y.-W. Wan. 2005. Capacity competition of make-to order
rms. Oper. Res. Lett. 33 187194.
Chung, J., V. Rao. 2003. A general choice model for bundles with
multiple category products: Application to market segmenta-
tion and pricing of bundles. J. Marketing Res. 40(2) 115130.
Clark, K., W. Chew, T. Fujimoto, J. Meyer, F. Scherer. 1987. Product
development in the world auto industry. Brookings Papers on
Econom. Activity 3(Special Issue on Microeconomics) 729781.
Currim, I. 1982. Predictive testing of consumer choice models not
subject to independence of irrelevant alternatives. J. Marketing
Res. 19(2) 208222.
Deneckere, R., K. Peck. 1995. Competition over price and service
rate when demand is stochastic: A strategic analysis. RAND J.
Econom. 26(1) 148162.
Dewan, S., H. Mendelson. 1990. User delay costs and internal pric-
ing for service facility. Management Sci. 36(12) 15021517.
Dewan, S., H. Mendelson. 1998. Information technology and time-
based competition in nancial markets. Management Sci. 44(5)
595609.
Dewan, S., H. Mendelson. 2001. Information technology and trader
competition in nancial markets: Endogenous liquidity. Man-
agement Sci. 47(12) 15811587.
Feldman, S. 2007. Distributor spotlight: Apollo: Focus, exibil-
ity, fundamentals goal is to be better, but not necessarily
bigger. FloorBiz.com (October 25), http://www.oorbiz.com/
BizNews/NPViewArticle.asp?cmd=view&articleid=2620.
Fradera, I. 1986. Perfect competition with product differentiation.
Internat. Econom. Rev. 27(3) 529538.
Gibbens, R., R. Mason, R. Steinberg. 2000. Internet service
classes under competition. IEEE J. Selected Areas Comm. 18(12)
24902498.
Grover, R., V. Srinivasan. 1987. A simultaneous approach to market
segmentation and market structuring. J. Marketing Res. 24(2)
139153.
Gupta, A., D. Wilemon. 1990. Accelerating the development of
technology-based new products. California Management Rev.
32(2) 2444.
Guttentag, J. 2001. How much do mortgage brokers make?
mtgprofessor.com, (April 9) http://www.mtgprofessor.com/A%
20-%20Mortgage%20Brokers/how_much_do_mortgage_brokers
_make.htm.
Hamilton, B. 2007. Laminate update 2007. Floor Focus Maga-
zine (August/September), http://www.oordaily.net/Search/
SearchItem.aspx?article=11722.
Hassin, R., M. Haviv. 2002. To Queue or Not to Queue: Equilibrium
Behavior in Queueing Systems. Springer, New York.
Hess, C., C. Kremer. 1994. Computerized loan origination systems:
An industry case study of the electronic markets hypothesis.
MIS Quart. 18(3) 251275.
Jedidi, K., S. Jagpal, P. Manchanda. 2003. Measuring heteroge-
neous reservation price for product bundles. Marketing Sci.
22(1) 107130.
Kalish, S., P. Nelson. 1991. A comparison of ranking, rating and
reservation price measurement in conjoint analysis. Marketing
Lett. 2(4) 327335.
Kalvenes, J., N. Keon. 2007. Revenue management in mobile com-
munication networks. Working paper, University of Texas
at Dallas.
Kamakura, W., G. Russell. 1989. Probabilistic choice model for mar-
ket segmentation. J. Marketing Res. 26 379390.
Afanasyev and Mendelson: Service Provider Competition
Manufacturing & Service Operations Management 12(2), pp. 213235, 2010 INFORMS 235
Kamakura, W., R. Srivastava. 1984. Predicting choice shares under
conditions of brand interdependence. J. Marketing Res. 21
420434.
Katta, A., J. Sethuraman. 2005. Pricing strategies and service differ-
entiation in queuesA prot maximization perspective. Work-
ing paper, Columbia University, New York.
Lederer, P., L. Li. 1997. Pricing, production, scheduling, and deliv-
ery time competition. Oper. Res. 45(3) 407420.
Levhari, D., I. Luski. 1978. Duopoly pricing and waiting lines. Eur.
Econom. Rev. 11(1) 1735.
Loch, C. 1991. Pricing in markets sensitive to delays. Ph.D. thesis,
Stanford University, Stanford, CA.
Luski, I. 1976. On a partial equilibrium in a queueing system with
two servers. Rev. Econom. Stud. 43(3) 519525.
Mabert, V., J. Muth, R. Schmenner. 1992. Collapsing new product
development times: Six case studies. J. Product Innovation Man-
agement 9 200212.
Martin, R., E. Malykhina. 2007. Data latency playing an ever
increasing role in effective trading. Wall Street Tech. (May 25),
http://www.wallstreetandtech.com/showArticle.jhtml?articleID
=199702208.
Mas-Colell, A., M. Whinston, J. Green. 1995. Microeconomic Theory.
Oxford University Press, New York.
Mendelson, H. 1985. Pricing computer services: Queueing effects.
Comm. Assoc. Comput. Machinery 28 312321.
Mendelson, H., S. Whang. 1990. Optimal incentive-compatible pri-
ority pricing for the m/m/1 queue. Oper. Res. 38(5) 870883.
Milgrom, P., J. Roberts. 1990. Rationalizability, learning, and
equilibrium in games with strategic complementarities.
Econometrica 58(6) 12551277.
Milgrom, P., I. Segal. 2002. Envelope theorems for arbitrary choice
sets. Econometrica 70(2) 583601.
Millson, M., S. Raj, D. Wilemon. 1992. A survey of major approaches
for accelerating new product development. J. Product Innova-
tion Management 9 5369.
Pekgun, P., P. Grifn, P. Keskinocak. 2008. Centralized vs. decen-
tralized competition for price and lead-time sensitive demand.
Working paper, Georgia Institute of Technology, Atlanta.
Reitman, D. 1991. Endogenous quality differentiation in congested
markets. J. Indust. Econom. 39(6) 621647.
Smith, P., D. Reinertsen. 1998. Developing Products in Half the Time:
New Rules, New Tools. John Wiley & Sons, New York.
Smith, W. 1956. Product differentiation and market segmenta-
tion as alternative marketing strategies. J. Marketing 21(1)
38.
So, K. 2000. Price and time competition for service delivery. Manu-
facturing Service Oper. Management 2(4) 392409.
Stalk, G. 1988. TimeThe next source of competitive advantage.
Harvard Business Rev. 66(4) 4151.
Stalk, G., T. Hout. 1990. Competing Against Time: How Time-Based
Competition Is Reshaping Global Markets. Free Press, New York.
Sun, J., X. Sun, T. Xu. 2004. Competitive advantage based on
innovation. Technical report, Fourth Asia Academy of Man-
agement Conference, http://faculty.washington.edu/karyiu/
confer/beijing03/papers/sun&sun&xu.pdf.
Takeuchi, H., I. Nonaka. 1986. The new product development game.
Harvard Business Rev. 64(1) 137146.
Taylor, H., S. Karlin. 1998. An Introduction to Stochastic Modeling,
3rd ed. Academic Press Limited, London.
Topkis, D. 1978. Minimizing a submodular function on a lattice.
Oper. Res. 26 305321.
Viton, P. 1981. On competition and product differentiation in urban
transportation: The San Francisco Bay area. Bell J. Econom. 12(2)
362379.
Wang, T., R. Venkatesh, R. Chatterjee. 2003. Reservation price as a
range: An incentive compatible measurement approach. J. Mar-
keting Res. 40(2) 115130.
Weller, I., E. Wai, S. Jaglal, H. Kreder. 2005. The effect of hospi-
tal type and surgical delay on mortality after surgery for hip
fracture. J. Bone Joint Surgery 87-B(3) 361366.
Wertenbroch, K., B. Skiera. 2002. Measuring consumer willingness
to pay at the point of purchase. J. Marketing Res. 39(2) 228241.
Yahalom, T., M. Harrison, S. Kumar. 2006. Designing and pricing
incentive compatible grades of service in queueing systems.
Working paper, Stanford University, Stanford, CA.

Potrebbero piacerti anche