Professional Documents
Culture Documents
www.elsevier.com/locate/dsw
Abstract
Oering a variety of products is important for a ®rm to attract dierent consumer segments. However, high product
variety increases production and distribution costs. Modular product design and parts commonality are approaches
used to counter this trend in cost and still oer a variety of products. This paper develops a model to examine when
modular products should be introduced and how much modularity to oer. The model looks at a market consisting of a
high segment and a low segment. Customers choose the product that maximizes their surplus, which is de®ned as the
productÕs utility minus its price. Presence of commonality aects the utility of a product. Greater commonality de-
creases production cost but makes the products more indistinguishable from one another. This makes the product more
desirable for the low segment but less desirable for the high segment. The ®rmÕs objective is to design the products and
set the prices so as to maximize its pro®t. Ó 2000 Elsevier Science B.V. All rights reserved.
0377-2217/00/$ - see front matter Ó 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 7 - 2 2 1 7 ( 9 9 ) 0 0 2 7 1 - 4
K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621 603
Seagate) concerned about product variety. Child of-scale in production by using a common product
et al. (1991) discuss the operational problems re- module for multiple products (Rutenberg, 1969),
lated to the complexity of product variety, and reduction of inventory holding costs due to the
recognize that 80% of manufacturing costs, 50% of pooling eect against demand uncertainty (Baker
quality, 50% of order leadtime, and 80% of busi- et al., 1986), and reduction in investment in pro-
ness complexity can be in¯uenced by product and duction equipment (Cooper, 1994; Pisano and
process design. Rossi, 1994).
The use of modularity/commonality in the de- The use of commonality is not free. Excessive
sign of a product family or generations of products commonality may make products very similar
provides the ®rm with a chance to meet diverse (Ulrich and Tung, 1991). This may aect the cus-
customer needs with less cost due to economies-of- tomersÕ valuation of products and can negatively
scale in procurement, production and distribution. aect the ®rmÕs pro®ts if a product does not appeal
Hence the concept of commonality in product to the customers for whom it is designed. This will
design is widely used in many industries ranging result in product cannibalization. Note that this
from automobile to household appliance (Sanchez type of cannibalization, which is due to product
and Mahoney, 1995). Honda manufactures dier- similarity, diers from the cannibalization due to
ent versions of Accord targeting local markets the substitution eect. Cannibalization due to the
around the world based on a single design plat- substitution eect does not assume physical or
form (Business Week, 1997). Toyota Camry and functional similarity among products, but it takes
Lexus ES300, the entry-level luxury sedan, are place when dierent products exist and satisfy the
built on the same design platform (Automotive same customersÕ needs in dierent ways. Micro-
Industries, 1996). Answering machines in dierent waves will cannibalize the market for ovens not
price categories share common features such as because of physical or functional similarity to the
`memo' and `remote access' while they are dier- oven but because both can meet the same cus-
entiated through the length of recording time and tomersÕ needs in dierent ways. On the other hand,
the number of outgoing messages. cannibalization due to product similarity occurs
The issue of part commonality on cost saving in when dierent products satisfy the needs in a
inventory is ®rst addressed and studied by similar way. When high- and low-end microwaves
Rutenberg (1969) and Rutenberg and Shaftel are in the market and they share many common
(1971). They simultaneously consider the design of parts, two things can happen. Either high-end
modules and the decision of common modules to customers may ®nd the low-end product is also
be used for dierent applications and markets. In appealing, or low-end customers may also ®nd the
their model, the problem of commonality is posed high-end product is appealing if it is aordable, or
as a problem of economic balance between econ- both cases can happen simultaneously.
omies-of-scale and the disutility of refusing to In this paper, we are interested in the eects of
provide each customer segment with an item ®tting commonality on customersÕ valuation of products
its exact requirements. On the supply side of the when it is used to design products in dierent
®rm, research on modularity/commonality has classes. Common features increase similarity be-
evolved and now includes studies on understand- tween products and in¯uence customersÕ choice of
ing the impact of component commonality on products (Chernev, 1997; Tversky, 1977). Consider
stocking level (Gerchak et al., 1988; Baker et al., the example of 1997 Toyota Camry and 1997
1986), designing of product family and generations Lexus ES300. Because of the common design
(Sanderson and Uzumeri, 1997; Smith and Rein- platform, buyers may perceive Camry similar to
ertsen, 1991), and designing of products for post- Lexus ES300 and associate the quality of Camry
ponement (Lee and Billington, 1994). In many with that of Lexus. The buyers of Camry may also
cases, it is possible to achieve tangible cost saving be willing to pay a higher price for the product,
through modular product design. The reasons for knowing that an important part of the design is the
cost savings are several and include the economies- same as the high-end product, Lexus. This type of
604 K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621
valuation premium is similar to the `Veblen eect' through overall levels of quality. The model con-
in a theory of conspicuous consumption in eco- siders a monopolistic ®rm who serves a market
nomics. The Veblen eect, de®ned to be `willing- consisting of a high segment, h, and a low segment,
ness to pay a higher price for a functionally l. CustomersÕ utility linearly increases with quality
equivalent good' (Bagwell and Berheim, 1996), so that a quality q provides a utility of vh q to the
arises from the desire to signal wealth. In our high segment and a utility of vl q to the low seg-
model of product design for product classes, val- ment. The two segments dier in that the high
uation premium arises when a customer is likely to segment values the quality higher than the low
choose one product over the other. If this type of segment so that vh > vl . Cost of providing a unit of
valuation premium prevails, product similarity due quality increases at an increasing rate as the level
to common design will result in increased revenue of quality increases and is given by cq2 where c is a
from the customers of low-end product. On the constant. The size of each segment, nh and nl , is
other hand, the similarity may lower the custom- given and all consumers in each segment choose
ersÕ willingness to pay for high-end product and the product which maximizes their surplus which is
even lead high-end customers to switch to the low- de®ned as productÕs utility minus its price.
end product. To operationalize the notion of commonality,
We adopt the economic modeling framework we assume that the overall quality, q, is provided
developed by Moorthy and Png (1992) and extend it to customers through a modular design, qm , and a
to incorporate the issues of commonality and sim- custom design speci®c to the types of customer,
ilarity. We address the two types of cannibalization q ÿ qm . Although the quality provided through the
problems associated with the modular product de- common design will yield the same utility as the
sign for two product classes. The contribution of quality provided through the custom design, there
this paper is in providing a framework to answer is a valuation change due to the product similarity
these questions: (1) Under what condition does if the common design is used. This valuation
commonality help the ®rm to increase revenue? change factor, 0 6 b 6 1, aects the perceived
When is the common design strategy optimal quality of products. We model the valuation
compared to non-common product design strate- change such that the common design used in the
gies? (2) How does the product similarity interact low-end product will provide a valuation premium,
with cannibalization due to the substitution eect? bp , for its buyers while the high-end product will
and (3) How does the product design with com- undergo a valuation discount, bd . Thus the per-
monality dier from the one without commonality? ceived quality of the high-end product reduces to
This paper is organized as follows. Section 2 qh ÿ bd qm and the perceived quality of the low-end
introduces the model. Section 3 presents various product increases to ql bp qm . Note that this is
product design strategies the ®rm can consider. We regardless of the segment type who values the
provide a sensitivity analysis of each strategy in products. Valuation discount and premium take
Section 4. Section 5 presents the conditions where place because of the existence of two products with
each of strategies is optimal, Section 6 presents common parts, not because of the existence of
model implications, and the conclusion is in customer segments with dierent quality valua-
Section 7. All proofs are in Appendix A. tion. Therefore, if the ®rm oers a single product,
then there wouldnÕt be any valuation change. On
the supply side, cost saving of caf
qm will occur if
2. Model a common modular design is used for the design of
multiple products so that the total cost function is
In this paper, we assume that the overall dis- c
q2 ÿ a f
qm where 0 6 a < 1 is the cost saving
tinguishing features of a product can be repre- parameter and f
qm is the cost saving function
sented in a single dimension which we call `quality' that is non-decreasing in qm .
and denote its value by q. Thus, if there are mul- The ®rm wants to design two products with
tiple products in the market, they are dierentiated quality qh and ql for the two segments with a
K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621 605
ÿ
common module, qm . (In Section 3, we expand the vh
qh ÿ bd qm ÿ
vh ÿ vl ql bp qm
options of the ®rm by considering single product P Ph P vl
qh ÿ bd qm :
6
strategies.) The model for the case when the ®rm
introduces two products can be stated as follows: Given Pl , the ®rm charges the highest possible
ÿ
Max nl Pl ÿ cq2l acf
qm price for the high-end product which is given by
Pl ;Ph ;ql ;qh ;qm
ÿ the left-hand side term of (6). Note that at this
nh Ph ÿ cq2h acf
qm price the high segment is indierent in buying ei-
s.t. ther of the products as it gets the same surplus
ÿ from both products. Thus constraint (2) is binding
vl ql bp qm ÿ Pl P vl
qh ÿ bd qm ÿ Ph ;
1
(level c in Fig. 1) and can be rewritten as
ÿ
vh
qh ÿ bd qm ÿ Ph P vh ql bp qm ÿ Pl ;
2
ÿ Ph vh qh ÿ ql
vh ÿ vl ÿ qm vh bd bp
vh ÿ vl :
v l ql b p q m P P l ;
3
7
vh
qh ÿ bd qm P Ph ;
4
By considering the lower and upper bounds on Ph
qh P 0; ql P 0; qh P qm ; ql P qm ; qm P 0: in Eq. (6) we get
In Eq. (1), vl
ql bp qm (vl
qh ÿ bd qm ) is segment ÿ
qh ÿ ql ÿ bp bd qm P 0:
8
lÕs utility of the low (high) product and left (right)
hand side is the surplus derived from the low An interpretation of Eq. (8) is that the ®rm must
(high) product. Eq. (2) is similarly for the high ensure that the perceived quality of the high-end
segment. Eqs. (1) and (2) are the self-selection product is greater than or equal to that of the
constraints which are to ensure that each segment low-end product, i.e., qh ÿ bd qm P ql bp qm . We
will voluntarily buy the product designed for that assume
segment. Eqs. (3) and (4) are the participation
constraints that ensure that each segment gets non- q h ÿ b d qm > ql b p qm
9
negative surplus from the purchase of the product.
Product price and quality are decided in a manner so that the perceived quality of the high-end
so that they do not violate the consumer self- product is still higher than that of the low-end
selection and participation constraints, and the product.
®rmÕs goal is to design and price the products so as
to maximize its pro®t.
Now we analyze which constraints will be
binding. If the ®rm tries to extract all the surplus 3. Firm's product design strategies
of customers in the high segment by setting
Ph vh
qh ÿ bd qm (level a in Fig. 1), then these Substituting the values of Pl and Ph in the
customers will switch to the low-end product to get objective function with (5) and (7), we have the
a positive surplus since following objective function:
nl nh vl fbp ÿ T
bp bd gqm and Proposition 1. The existence of valuation change
nl nh acf
qm ; factors increases revenue when the degree of canni-
balization, T, is less than bp =
bp bd , while it
decreases revenue when T is greater than
respectively. The changes in revenue consist of two bp =
bp bd .
components, one from the low segment and the
other from the high segment, and given by
Proposition 2. Completely modular design is the
optimal design of commonality when the degree of
vl bp nl qm and ÿ f
vh ÿ vl bp vh bd gnh qm ; cannibalization, T, is less than bp =
bp bd .
respectively. We refer to the increase in revenue When T > bp =
bp bd , the commonality re-
from the low segment as the premium eect sults in a loss of revenue so that the optimal design
whereas the decrease in revenue from the high of commonality will be decided from the balance
segment as cannibalization due to similarity. We between the loss of revenue and cost saving due to
®rst note that when T 6 bp =
bp bd , commonal- commonality. When the cost saving is any convex
ity always increases the pro®t for any form of the function in qm , then for any value of ql , the optimal
cost saving, f
qm , which is non-decreasing in qm solution for qm lies at the extreme point, qm 0 or
so that the optimal design of commonality is qm ql . Therefore the optimal design of com-
qm ql , which we term as completely modular monality, if used, is again completely modular
design (we use the term `partially modular design' design. For other forms of cost saving functions
when 0 < qm < ql ). The value of T represents the partially modular design can be the optimal design
degree of cannibalization. The cannibalization is (see Appendix A). To obtain more insights on
severe when relative size of the high segment, w, is dierent product design strategies, we now assume
large and the valuation dierence, vhl , is large. a particular form of cost saving, f
qm q2m .
Above observations are summarized in Proposi- Bagchi and Gutierrez (1992) studied the impact of
tions 1 and 2. number of common components on inventory
K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621 607
costs. They found not only that the introduction of decreases as the value of T increases to mitigate
more common components results in larger cost cannibalization.
reduction but also that the cost reduction increases
at an increasing rate. However, we also note that
there are situations where concave cost saving is 3.2. S2: Two modular products
more reasonable.
With the convex cost saving assumption, the In the second strategy, we set qm ql , and solve
optimal design of commonality is either qm 0 or it using the ®rst order condition to obtain the
qm ql . When qm 0, we have the ®rst strategy following equations:
(S1), two products without modular design; when vh
qh ;
qm ql , we have the second strategy (S2), two 2c ÿ ÿ
products with modular design. Two additional vl 1 bp ÿ T 1 bp bd
product design strategies available to the ®rm are, ql qm :
2cf1 ÿ w ÿ ag
a single product for both segments with modular
design (S3), and a single product for only the high The resulting pro®t will be
segment (S4) (do not serve the low segment). Note ÿ ÿ 2
that the strategy of a single product for only the v2l
nl nh 1 bp ÿ T 1 bp bd
p2
low segment is not possible since a single product 4cf1 ÿ w ÿ ag
designed for the low segment would also be at- nh v2h
tractive to the high segment. We now derive the :
4c
®rmÕs product design and consequent pro®t under
each strategy. The constraint T <
1 bp =
1 bp bd is nec-
essary to have two products with positive quality,
and
ÿ ÿ vl
3.1. S1: Two unique products 1 ÿ a > 1 bp 1 b p b d
vh
n ÿ 2 o
In the two unique products strategy, since no w 1 ÿ 1 bp bd
11
modular design is used, there is no valuation dis-
count and premium and the problem reduces to results from constraint (9). Note that when (11)
the one considered by Moorthy and Png (1992). holds and T <
1 bp =
1 bp bd , we will
With bp bd 0 we note that the objective have 1 ÿ a > w and the pro®t will be positive.
function (10) is concave in qh and ql . Using the ®rst While p2 can be positive even when 1 ÿ a 6 w but
order condition we obtain the following equations: S4 will dominate S2.
vh vl
1 ÿ T
qh ; ql : 3.3. S3: Single product for both segments
2c 2c
1 ÿ w
These products will be priced at Pl vl ql , In this case, qh ql q. As there is only one
Ph vh qh ÿ
vh ÿ vl ql and will result in a pro®t of product in the market, there is no valuation
change since a valuation change occurs due to the
v2l
nl nh
1 ÿ T 2 nh v2h existence of an alternate product with common
p1 :
4c
1 ÿ w 4c design. However, there is cost saving due to
economies-of-scale as q is provided for both seg-
To ensure that ql is positive, we need T < 1. ments and consequently, the cost function be-
Product quality for the high segment is set at its comes c
1 ÿ aq2 . In this strategy, the maximum
ecient level which is the quality that gives the price the ®rm can charge is vl q; otherwise the low
®rm maximum possible pro®t from the high segment would get negative surplus. The high
segment. Product quality for the low segment segment would get the positive surplus since
608 K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621
vh q > vl q P and both segments will buy the 4.2. S2: Two modular products
product at this price. The optimal product design,
price and pro®t are as follows: The product quality for the high segment is
again linearly increasing in vh . The product quality
vl v2l
q ; P ; for the low segment is increasing in vl , but is de-
2c
1 ÿ a 2c
1 ÿ a creasing in vh up to
nl nh v2l
p3 : ÿ
4c
1 ÿ a 1 bp 1
vhl ÿ :
1 bp bd w
3.4. S4: Single product for high segment only
Beyond this value of vhl , it is not feasible to provide
This strategy may make sense when the canni- two modular products.
balization problem is so severe that the ®rm can- As a increases, more cost saving is possible so
not aord to have two products in the market. that qm increases. Increased qm increases the val-
There is no cost saving since the product is pro- uation premium and this allows the ®rm to charge
vided only for the high segment. The product a higher price for the low-end product so that Pl
design, price and the resulting pro®t are increases. On the other hand, increased qm makes
vh v2h nh v2h the low-end product more attractive to the high
q ; P ; p4 : segment so, to prevent the high segment from
2c 2c 4c
switching to the low-end product, Ph should be
lower.
4. Sensitivity analysis As bp increases, the ®rm can increase both ql
and qm (with ql qm ) and charge a higher price for
In this section, we explore how the pro®t and the low-end product by exploiting the larger
product design for each strategy changes with amount of valuation premium from the low seg-
dierent parameters of the model. Since the central ment. On the other hand, the high segment has an
question concerns two unique products versus two incentive to switch to the low-end product and
modular products, we also compare strategies S1 enjoy the valuation premium. To prevent this the
and S2. ®rm needs to compensate the high segment with
lower Ph .
As bd increases, qm goes down. This reduces the
4.1. S1: Two unique products total amount of valuation discount of the high-end
product, and prevents the high segment from
Product quality for the high segment is in- switching to the low-end product. As qm reduces,
creasing in vh , and product quality for the low Pl also reduces. On the other hand, increased bd
segment is increasing in vl , but is decreasing in w may increase or decrease Ph depending on other
and vh . Recall that w is the proportion of high-end parameter values. To understand this, consider
customers. As vh increases, for given vl and w, the Eq. (7) for Ph , with ql qm .
cannibalization becomes so severe that the ®rm ÿ
has to reduce the product quality for the low Ph vh
qh ÿ bd qm ÿ
vh ÿ vl qm 1 bp :
12
segment. However, optimal pro®t, p1 , is increasing
at an increasing rate in vh and vl , respectively (see Eq. (12) shows that the price of the high-end
Proposition 3). When T P 1, the two unique product is given by the dierence between the high
products strategy does not exist since ql 6 0 when segmentÕs utility from the high-end product (the
T P 1, and S1 essentially reduces to S4. Note that ®rst term in the right-hand side) and the high
the pro®t approaches the ecient pro®t as vhl segmentÕs surplus from the low-end product (the
approaches 1. second term in the right-hand side). Substituting
K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621 609
for qh and qm in Eq. (12) and taking a derivative the sign of the ®rst term is negative, but
with respect to bd yield oPh =obd > 0; the decrease in utility from the high-
ÿ end product is less than the decrease in surplus
oPh vh vl
ÿ 1 bp from the low-end product so that Ph increases in
obd 2c
1 ÿ w ÿ a bd . When
ÿ
ÿ
vh ÿ vl 1 bp vl T ÿ
ÿ T 1 bp 2bd : 1 bp
2c
1 ÿ w ÿ a h
vl < ÿ and
1 bp 2bd w
13 ÿ
h
1 bp
1 w
When vl < ÿ ;
ÿ 1 bp bd 2w
h
1 bp
1 w
vl < ÿ ; both sign of the ®rst term in the right-hand side of
1 bp bd 2w
(13) and oPh =obd are negative; the decrease in
the derivative is less than zero, and Ph will decrease utility from the high-end product is more than the
with an increase in bd . When decrease in surplus from the low-end product so
that the high segment is more likely to switch to
ÿ
1 bp
1 w the low-end product. To prevent the switching, Ph
h
vl > ÿ ; should be lower with an increase in bd .
1 bp bd 2w
Higher values of bp and bd increase the high
an increase in bd would make Ph higher. We note segmentÕs chance of switching to the low-end
that the high segmentÕs surplus from the low-end product. Therefore, both bp and bd are sources of
product is always decreasing in bd since increased cannibalization due to product similarity. On the
bd reduces qm and makes the low-end product less other hand, higher bp helps the ®rm to improve its
appealing to the high segment. Less satis®ed with revenue from the low segment.
the low-end product, the high segment is more For a ®xed value of w, both Ph and qh linearly
likely to purchase the high-end product so the in- increase in vh . However, p2 increases at an in-
creased bd provides an incentive for the ®rm to creasing rate in vh and vl , respectively. The changes
increase Ph . While the sign of the second term in of p1 and p2 in vh and vl are summarized in
the right-hand side of (13) is always positive, the Proposition 3.
sign of oPh =obd also depends on the changes in the
high segmentÕs utility from the high-end product Proposition 3. In their respective feasible regions, p1
(the ®rst term in the right-hand side of (13)), which and p2 increase at an increasing rate in vh and vl .
is essentially the changes in the total amount of
valuation discount associated with the high-end Note that
product, bd qm . When ÿ
ÿ oqm ÿw 1 bp bd
< 0:
h
1 bp ovh 2c
1 ÿ w ÿ a
vl > ÿ ;
1 bp 2bd w
Thus, to mitigate cannibalization, qm linearly de-
the sign of the ®rst term is positive, meaning that creases in vh . Table 1 provides aditional sensitivity
the high segmentÕs utility from the high-end prod- analyses.
uct increases in bd so that Ph increases in bd . When If the ®rm serves the low segment in isolation,
ÿ the ecient quality for the low segment from the
1 bp ®rmÕs perspective is vl =2c. Without the valuation
vhl < ÿ and
1 bp 2bd w change, in the ®rst strategy, the optimal ql is al-
ÿ ways lower than vl =2c because of cannibalization
1 bp
1 w
ÿ < vhl ; whereas qh vh =2c. In the second strategy, ql in-
1 bp bd 2w creases as bp increases. The following Proposition
610 K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621
Table 1
Sensitivity analysis of a, bp , bd , vh on price, quality and p2
vl 1 Ph Pl qh ql qm p2
a " # " ± " "
bp " # ÿ " ± " "
1 bp
1 w # ± # #
bd " " if vhl > ÿ
1 bp bd 2w
# otherwise
vh " " # " # "
( !)
shows the existence of the region where the opti- b p bd
mal ql is above vl =2c and the pro®t is larger than DR 1 ÿ T vl
nl nh bp q1l ;
bp
nl v2l =4c nh v2h =4c which is the pro®t from the ef- ÿ 1 2
®cient quality for each segment. DC ca
nl nh ql ;
from the high segment. The following observations revenue from the low segment in the second
result from the pro®t comparisons. strategy compared to the ®rst strategy. Neverthe-
less pro®t in the second strategy will be higher as
Observation 1. With no valuation discount (bd 0), long as q2l > q1
l (the optimal low-segment quality
the two modular products strategy dominates the found under the strategy 1).
two unique products strategy (i.e., p2 P p1 ). !
1 a 2
1 ÿ wbp
Case 3 : ÿ 6 vhl
Observation 2. When the degree of cannibalization w a 2
1 ÿ w bp bd
is greater than bp =
bp bd and there is no cost !
saving, then the two unique products strategy dom- 1 a
1 ÿ wbp
< ÿ :
inates the two modular products strategy. w a
1 ÿ w bp bd
In comparing product designs, qh is the same As in Case 2, we have DR 6 0 and DC P 0 but,
under both strategies whereas q1 2
l and ql may not
unlike Case 2, DR DC 6 0. The optimal low-
be the same. Looking at the relative values, segment quality in strategy 2 will be higher than
! the optimal low-segment quality in strategy 1 to
1 a
1 ÿ wbp make strategy 2 attractive.
1 2 h
ql < ql if vl < ÿ
w a
1 ÿ w bp bd !
1 a
1 ÿ wbp
Case 4 : ÿ 6 vhl :
and q1 2
l P ql otherwise. We now compare the w a
1 ÿ w bp bd
pro®ts under the two strategies for dierent values
of vhl . Consider the following four cases and note As in Case 3, DR 6 0, DC P 0 and DR DC 6 0.
that DC P 0: However, unlike other cases, the cannibalization is
!
b so severe that the ®rm should lower the quality in
1 p
Case 1 : vhl < the second strategy even if there is cost saving
w bp bd
! associated with it so q2 l 6 ql
1
(where q2
l is the
optimal quality for the low-segment in S2). In
1 a 2
1 ÿ wbp
< ÿ other words, the cost saving should not result in an
w a 2
1 ÿ w bp bd
! increase of quality for the low segment since it only
1 a
1 ÿ wbp worsens the cannibalization in the second strategy.
< ÿ :
w a
1 ÿ w bp bd In terms of overall pro®ts, strategy 1 dominates
strategy 2.
In this case DR > 0. As in Proposition 1, the degree
of cannibalization is low so that the increase in
revenue due to the premium eect is always larger 4.4. S3: Single product for both segments
than the decrease in revenue caused by the canni-
balization due to similarity from the high segment. Given w, quality and pro®t depend only on vl
In this case, the ®rm has an incentive to make and a. As the cost saving increases, quality and
q2l > q1l . The pro®t will increase even without ac- pro®t increase at an increasing rate. As vh in-
counting for cost saving due to the modular design. creases, this strategy becomes relatively unattrac-
!
bp tive because in this strategy the ®rm has to charge
1
Case 2 : 6 vhl the price based on vl .
w bp bd
!
1 a 2
1 ÿ wbp
< ÿ : 4.5. S4: Single product for high segment only
w a 2
1 ÿ w bp bd
In this case, DR < 0 since the decrease in revenue When the cannibalization is severe, the ®rm
from the high segment is larger than the increase in may serve the high market only. This is true when
612 K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621
the cost saving is not signi®cant and the gap in and w 1=3. For example, the region marked p1
quality valuation is large. In this strategy, the ®rm in Fig. 2 shows the range of values of a and vh
will increase the level of quality and increase the where strategy 1 is better than strategy 2, and p2
pro®t as vh increases. marks the range where strategy 2 is better than
strategy 1.
The iso-pro®t line (14), marked (a) in Fig. 2, is
5. Optimal product design strategies decreasing in vh at a decreasing rate. While in-
creasing a makes the modular product strategy
Given pro®ts from the four dierent strategies, more desirable, increasing vh makes the cannibal-
we make all possible pair-wise comparisons of ization more intense due to the existence of valu-
strategies to obtain regions where one strategy ation change in the second strategy. The line (b) in
dominates the others. Fig. 2 is from Eq. (11) which is the feasibility
condition of the second strategy. The feasible
p1 P p2 () 1 ÿ a P w region for the second strategy increases as vh
ÿ ÿ 2
1 bp ÿ wvhl 1 bp bd increases.
1 ÿ w 2
;
14 To compare p2 and p3 , we need to consider the
1 ÿ wvhl
relevant region where both these strategies are
feasible. Recall that in strategy S2, Eq. (11) has to
p1 P p3 () hold, which is the region above the line marked (b)
1ÿw in Fig. 3. The iso-pro®t line between p2 and p3
1 ÿ aP 2 2
;
15 (marked as (a) in Fig. 3) intersects line (b) at
1 ÿ wvhl wvhl
1 ÿ w
1 2b
2 b
vhl
p2 P p3 () 1 ÿ a P f
1 ÿ w ÿ ag= 4
1 b2 w
2 2
ff
1 bp ÿ wvhl
1 bp bd g wvhl
which satis®es
1 ÿ w ÿ ag;
16
1
1 b
6 vhl 6 :
1 2w
1 2bw
p4 P p3 () 1ÿ aP 2 ;
17
wvhl
For the lower values of vhl < vhl , the line of con-
straint lies above the iso-pro®t line and it is below
1
when vhl > vhl . For vhl < vhl , if 1 ÿ a decreases be-
p4 P p1 () vhl P ;
18
w yond line (b), S2 becomes infeasible since the
perceived quality of the low product is higher than
1 bp that of the high product so that S3 is optimal. For
p2 P p4 () vhl 6 :
19
1 bp bd
1 b
vhl < vhl < ;
By changing the above inequalities into equal-
1 2bw
ities, we get iso-pro®t lines. For brevity, we will use
the same equation numbers (14) through (19) to as 1 ÿ a decreases, it is better for the ®rm to oer
refer to these lines. Using these iso-pro®t lines and the single product for both segments rather than to
the dominating regions, we plot the optimum keep a two-product strategy since signi®cant cost
product design strategy with 1 ÿ a on the y-axis saving allows the ®rm to oer the single product
and vh on the x-axis. To obtain insights we draw with high quality and to enjoy higher pro®ts.
these lines using speci®c parameter values. Figs. 2± Fig. 4 is the comparison of S2 and S4. As noted
5 provide pair-wise comparisons of four dierent previously, S2 is bounded by its feasibility con-
strategies with bp bd b 0:1, vl 1, c 0:1 straints and dominates S4 whenever it is feasible.
K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621 613
Fig. 3. p2 vs. p3 .
The iso-pro®t line between p1 and p3 ®rst de- ®es cannibalization so that valuation change ad-
creases at a decreasing rate and then decreases at versely aects the ®rmÕs pro®t. In this case, the two
an p increasing rate when vhl is greater than
unique products strategy is better since a more
1
1 ÿ w=
4 ÿ w. The iso-pro®t line between modular design lessens pro®t (Observation 2).
p3 and p4 decreases at an increasing rate in vhl . As When a 0 it can be shown that the single prod-
vhl increases, it requires more cost saving for the uct for both segments solution is never optimal.
single product for the both segments strategy to be This result, which is also observed in Moorthy and
an optimal strategy (Fig. 5). Png (1992), is given below.
When a 0 and vhl is small (< 1=2w), the
completely modular product solution is optimal Proposition 5. When a 0, the single product for
since, given the small dierence of quality valua- both segments strategy is never optimal.
tions, the ®rm should take advantage of the pre-
mium eect (Proposition 1). However, as vhl We now compare all four strategies to ®nd
increases beyond 1=2w, valuation change intensi- the optimum product strategy. Overlaying the
614 K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621
Fig. 4. p2 vs. p4 .
pair-wise comparisons discussed above, Fig. 6 is For mid range values of vh , p2 increases with
obtained. In drawing this ®gure, as before, we cost saving parameter a (see line (b) in Fig. 6). The
make a and vh changing variables and set increased economies-of-scale allows the ®rm to
bp bd b 0:1, vl 1, c 0:1 and w 1=3. In design the same modular product with less costs so
Fig. 6, ®rst consider the case where the dierence as to increase the pro®t (as a increases qm also
in the quality valuation is large. In this region, increases). However, as a further increases, the
upper part of line (a) in Fig. 6, cannibalization is value of qm approaches qh so that two products are
so severe that the best strategy for the ®rm is to less and less distinguishable from each other,
oer a single product for the high segment only. A which eventually leads to the adoption of single
single product for both segments is the optimal product for both segments strategy. The two un-
strategy when the cost saving is signi®cant. At ique products strategy becomes an optimal strat-
these extreme values of vh , a two-product solution egy in the small region at the northwest corner of
is not feasible. Fig. 6 (see Proposition 6).
K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621 615
Given an appropriate value of a, for low values Proposition 6. To adopt the modular design strate-
of vh (see line (c) in Fig. 6) at ®rst the ®rm can take gy, a minimum degree of valuation difference is re-
advantage of cost saving and low valuation dif- quired, which is equal to
ference by adopting S3. As vh increases, the ®rm ÿ ÿ
can improve the pro®t by dierentiating products 1 bp 1 bp bd
h
vl;min n ÿ 2 o :
using the modular design strategy while enjoying
1 ÿ a ÿ w 1 ÿ 1 bp bd
cost savings. As vh further increases, the ®rm is
better o by adopting the two unique products
strategy not to suer from the cannibalization due
To study the eect of the relative size of the two
to product similarity. As vh increases, the quality
segments, Figs. 7 and 8 are drawn with w 0:5
for the low segment decreases in S1. When vh in-
and w 0:75, respectively, keeping other param-
creases beyond 1=w, the two unique products
eters the same as in Fig. 6. We observe that the
strategy is not feasible since ql is reduced to zero.
region where S2 is the optimal strategy decreases
An interesting observation from the pro®t
with w. For the modular product design strategy
comparison is that there has to be some degree of
to be viable we have the constraint that
valuation dierence between the segments for the
ÿ
®rm to use the modular design strategy. This is 1 bp
because the ®rm must design high and low end vhl < ÿ :
1 bp bd w
products so that the perceived quality of products
should be in the correct order after the valuation Since the minimum value of vhl is 1, this implies that
change takes place. When there is a cost saving ÿ
and the relative size of the low segment is large, the 1 bp
optimal ql is higher (oql =oa > 0, oql =ow < 0 if Eq. w<ÿ
1 bp bd
(11) holds), which in turn increases the perceived
quality of the low end product, given b > 0. The should hold for the modular product design
following proposition derived from Eq. (11) indi- strategy to be feasible. If the relative size of
cates the necessary amount of valuation dierence the high segment is larger than this value, then
for the ®rm to use the modular design. we wouldnÕt have the modular product design
616 K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621
segmentÕs low willingness to pay for quality. Buyers cost saving that depend on the size of the low-seg-
of the higher-end products seek quality that cannot ment wouldnÕt justify the loss of a product attribute
be found in the lower-end products. However, that can be used for product dierentiation, and the
commonality reduces the perceived quality dier- subsequent loss of pro®t from the high-segment.
ence, and despite of its cost saving, it will further As cost saving parameter (a) increases, the
adversely aect the pro®t when there is a higher amount of common quality increases when other
expectation for quality dierence. In this regard, parameter values are ®xed (Table 1). Higher cost
commonality would only be used for a group of saving certainly motivates the ®rm to adopt more
products in similar price ranges, and the products commonality between products. In the product
in vastly dierent price ranges should be distin- development stage, Nissan Motor Company
guished by adopting a design that has unique at- identi®es ways to increase the commonality of
tributes. Recent use of memory chips in answering parts across variations and models to achieve the
machines is a case in point. The memory chips are desired cost level (Cooper, 1994). Commonality
used for message recording in machines with higher with higher-end product helps the ®rm in increas-
prices whereas cassette tapes are used in lower-price ing the price of the low-end products. However,
machines. If the chips were used in the low-end the increased similarity may weaken the justi®ca-
products as well it would have blurred merits of the tion for any price dierence between products so
machines with higher prices. that the ®rm has to reduce the price of the high-
Although we havenÕt considered multiple attri- end product. For instance, incorporating the in-
butes in this paper, the notion of valuation change dexed gear shifting system into lower-end bicycle
carries some implications for the selection of com- may make the high-end bicycle less attractive due
mon attributes when products possess multiple at- to valuation discount so that the price of the high-
tributes. If the cost saving is similar across the end bicycle has to be lowered.
attributes, a common attribute should be such that Our analysis has shown that commonality in
either it does not in¯uence the valuation of prod- design strategy can be optimal even if there is no
ucts, or it is viewed favorably by the low segment cost savings; the valuation change due to com-
but less in¯uential in the judgment of the high-end monality itself can be pro®table to the ®rm. This
product (Observation 1). For example, if there is an happens when the target segmentsÕ quality valua-
attribute that no longer characterizes the high-end tions are such that the premium eect dominates
product, then incorporating the attribute into the the valuation discount with high-end products
low-end products would not be critical in the (Proposition 1), or when the valuation discount
evaluation of high-end product while low segments doesnÕt exist so that there is no negative eect of
might bene®t from this feature addition. The in- commonality (Observation 1). In addition to judi-
dexed gear shifting mechanism that was used only cious selection of target segments and of common
in expensive bicycles now can be found in less ex- attributes as discussed before, ®rms can be proac-
pensive bicycles as well. The high-segment now tive in avoiding the negative eects of commonal-
considers the attribute less unique so that the ity. In practice, this can be done in several ways
manufacturers oer the attribute in lower-end bi- including enhancing unique attributes on high-end
cycles to enhance the product value while enjoying products to deviate buyersÕ attention from com-
the cost saving through economies-of-scale. mon attributes (Chernev, 1997), and displaying
In using commonality, the ®rm should take ad- products in separate places (e.g., Camry and Lexus
vantage of the premium eect (vl bp qm nl ) associated are displayed and sold at separate dealer shops).
with the low-end product, and the premium eect is
greater when the size of the low-segment is larger.
In the case of bicycles discussed before, if there isnÕt 7. Concluding remarks
a sizable low-segment who might bene®t from the
addition of common features, the ®rm should deter We have assumed in the model that cost saving
from it. This is because the premium eect and total due to the modular design increases at an
618 K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621
increasing rate. This convex cost saving assump- linear in perceived quality, it is useful to under-
tion makes the optimal value of qm either 0 or ql stand the pattern of changes in perceived quality
and allows us to compare two main strategies. This and utility with the magnitude of commonality.
reduces mathematical complexity without obscur- Incorporating multiple attributes in the model will
ing, in our view, the main idea of the paper. Dif- also be a useful extension for the theory of mod-
ferent cost saving functions, of course, will yield ularity and commonality in product design. A
dierent design strategies (e.g., 0 < qm < ql ) as multiple-attribute model of product line design
shown in Appendix A. In case of concave cost may help in answering questions such as: Which
saving function, when the condition in Proposition attributes should be made common? Which set of
2 is not met, we conjecture that as the cost saving products/customers must get common attributes?
increases qm will approach ql , and further increase
in cost saving will make qm ql qh . However,
qualitative insights obtained here will still hold Appendix A
since the eect of valuation change depends on the
existence of common quality regardless of the The purpose of this appendix is to show
form of cost saving which leads to dierent designs through numerical examples that the partially
of common quality. modular design can be the optimal design of
Our model assumes a 0±1 market share with commonality when the cost saving is concave in
known size of each segment, nh and nl , so cost qm . We consider a speci®c form of cost saving,
savings due to volume is known for sure when a p
ac qm , thus the total cost function is
customer segment favors a particular product. If p
c
q2 ÿ a qm . The product design problem, (10),
the customer choice process is probabilistic then becomes as follows:
the cost savings will reduce and its computation
may become more involved. However, examining Max nh vh qh ÿ nh cq2h
nl nh vl
1 ÿ T ql
qh ;ql ;qm
other types of market share model, such as a logit ÿ
model, will be a useful extension. In the logit ÿ nl cq2l
nl nh vl bp ÿ T bp bd qm
model, proportional market share of a product is p
nl nh ac qm
given by a ratio between utility from the product
s:t:
and the sum of the all utility from products in the
market. As commonality increases the perceived qh ÿ bd qm P ql bp qm ;
similarity between products, it will lead to con- q h P qm ; ql P qm ; qh P 0; ql P 0; qm P 0:
vergence of utility and thereby market shares of
the products. Table 2 is obtained using GAMS, a standard
In this model, we exogenously assume the ex- optimization software, with nh 1, nl 2, vl 1,
istence of bp and bd . Examining the existence of bp c 0:1, bp bd 0:1 and a 0:05. It shows how
and bd , and factors in¯uencing the magnitude of the values of qh , ql and qm change as vh varies. First
them are interesting subjects of our on-going re- note that above problem formulation covers
search. Furthermore as we assumed that com- strategies 1, 2 and 4, but not strategy 3. By letting
monality linearly in¯uences the perceived quality bp bd 0 and qh ql qm q, we obtain a
of products, and customersÕ utility functions are formulation for the single product for both
Table 2
Numerical examples
vh 1.4 1.7 2.0 2.3 2.6 2.9
qh 7.0 8.5 10.0 11.5 13.0 14.5
ql 4.06 3.25 2.50 1.75 1.00 0.25
qm 4.06 0.035 0.006 0.002 0.001 0.0007
Pro®t 8.21 9.34 11.25 13.87 17.10 21.04
K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621 619
segments strategy (S3), which yields the optimal Using this and the feasibility condition (11), we get
solutions of q 5.006 and pro®t of 7.5336 re- the RHS and LHS, respectively, of the following
gardless of the values of vh . Note that the pro®t of inequalities:
S3 is dominated by the pro®t obtained in Table 2. ÿ ÿ
Note that 1 bp 1 bp bd h
! ÿ 2 < vl
1 ÿ w ÿ a w 1 bp bd
bp 1
1:5: bp w a
bp bd w <ÿ :
A:1
1 bp bd w
Table 2 shows that, when vh 1:4 < 1:5, the
completely modular design is the optimal design of To show that there is positive amount of region for
commonality (Proposition 2), and that as vh in- vhl in (A.1), letÕs assume a 0. We need to show (1)
creases beyond 1.5, the partially modular design is RHS of (A.1) is greater than 1 since vhl > 1, and (2)
the optimal design of commonality. RHS > LHS in (A.1). First, RHS of (A.1) is greater
than 1 when
Proof of Proposition 3. We make a proof only for
vh and omit a proof for vl as it is similar to one for bp
w<ÿ :
A:2
vh . We take derivatives of p1 and p2 w.r.t. vh to bp bd
obtain the following:
op1 2
vh ÿ vl nh
nh nl We now show (2), which can be reduced to
>0 and
ovh nl 4c b
w<ÿ ÿ p :
A:3
o2 p1 2nh
nh nl bp bd bp bd 2
> 0:
ov2h nl 4c
Note that when (A.3) holds, (A.2) also holds. Thus
op2 2nh vh ÿ ÿ when (A.3) holds there is a positive amount of
1 ÿ a ÿ 1 bp 1 region for vhl in (A.1) where ql > vl =2c.
ovh 4cf1 ÿ w ÿ ag
vl n ÿ 2 o ÿ Next, to show that the pro®t is larger than
bp bd ÿ w 1 ÿ 1 bp bd >0 nl v2l nh v2h =4c in this region, we need to show
vh
ÿ
since
1 ÿ w ÿ a
1 ÿ w < 1 bp
ÿ 2
ÿ ÿ vl ÿ T 1 bp bd :
A:4
1 ÿ a P 1 bp 1 bp bd
vh
n ÿ 2 o Again assuming a 0, (A.4) holds when
w 1 ÿ 1 bp bd
bp w
is a feasibility condition in strategy 2. vhl < ÿ ;
1 bp bd w
o2 p 2 2nh h n ÿ
2
1 ÿ a ÿ w 1 ÿ 1 bp which is the RHS of (A.1). Therefore when (A.3)
ovh 4cf1 ÿ w ÿ ag
2 oi holds, the proposition is true.
bd > 0:
Numerical example. Let c 0:1, vl 1,
vh 1:3, bp bd 0:1, nl 50, nh 10 and
Proof of Proposition 4. Let a 0. Then w 1=6 < 1=4:4 so (A.3) is satis®ed.
ÿ ÿ Furthermore ql 5:04 > vl =2c 5 and the re-
vl 1 bp ÿ T 1 bp bd vl sulting pro®t is 169.26 which is higher than 167.25,
ql > :
2cf1 ÿ w ÿ ag 2c the pro®t from the ecient quality.
620 K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621
Proof of Observation 1. Using the optimal pro®t Proof of Proposition 7. If T < bp =
bp bd , then
expressions derived in Sections 3.1 and 3.2 (with valuation change becomes the revenue increasing
bd 0) we get factor in S2 (see Proposition 1) but vhl > 1 thus
ÿ 2 w < bp =
bp bÿ d should
hold. Substituting a 0
v2l
nl nh
1 ÿ T 1 bp and vhl
bp = bp bd
1=w in Eq. (11) we ob-
p 2 ÿ p1
4c
1 ÿ w ÿ a tain w < bp =
1 bd
bp bd . Both w < bp =
bp
2 bd and w < bp =
1 bd
bp bd yield w < bp =
v2l
nl nh
1 ÿ T
ÿ
1 bd
bp bd :
4c
1 ÿ w
( ÿ 2
2
v2l
nl nh
1 ÿ T 1 bp
4c
1 ÿ w ÿ a References
)
1
ÿ > 0: Automotive Industries. 1996. Lexus ES300, October, p. 107.
1ÿw Bagwell, L.S., Berheim, B.D., 1996. Veblen eects in a theory of
conspicuous consumption. The American Economic Review
Note that 1 ÿ w ÿ a > 0 for S2 to be feasible. 86 (3), 349±373.
Baker, K.R., Magazine, M.J., Nuttle, H.L.W., 1986. The eect
of commonality on safety stocks in a simple inventory
Proof of Observation 2. With a 0; model. Management Science 32 (8), 982±988.
Bagchi, U., Gutierrez, G., 1992. Eect of increasing compo-
v2l
nl nh n 2 nents commonality on service level and holding cost. Naval
p1 ÿ p2
1 ÿ T
4c
1 ÿ w Research Logistics 39 (6), 815±832.
ÿ 2 o Business Week, 1997. Can Honda build a world car?
ÿ
1 ÿ T bp ÿ T bp bd : September 8, 100±108.
Chernev, A., 1997. The eect of common features on brand
Now suppose T > bp =
bp bd . Substituting choice: Moderating role of attribute importance. Journal of
Consumer Research 23, 304±311.
bp =
bp bd for T in the last term and simplifying
Child, P., Diederichs, R., Sanders, F., Wisniowski, S., 1991.
terms we get The management of complexity. Sloan Management Review
33 (1), 73±80.
v2l
nl nh n 2 Cooper, R. 1994. Nissan motor company, Ltd.: Target costing
p 1 ÿ p2 >
1 ÿ T ÿ
1 ÿ T
4c
1 ÿ w system. HBS case.
2 o Gerchak, Y., Magazine, M.J., Gamble, A.B., 1988. Component
b p ÿ bp 0: commonality with service level requirements. Management
Science 34 (6), 753±760.
Lee, H.L., Billington, C., 1994. Designing products and
processes for postponement. In: Dasu, S., Eastman, C.
Proof of Proposition 5. When a 0;
(Eds.), Management of Design, Kluwer Academic Publish-
2 ers, Boston, pp. 105±122.
nl nh v2l 2 vh Martin, M., Hausman, W., Ishii, K., 1998. Design for variety.
p 1 ÿ p3
1 ÿ T T
1 ÿ w
4cnl vl In: Ho, T.-H., Tang, C., (Eds.), Product Variety Manage-
ment, Research Advances, Kluwer Academic Publishers,
ÿ
1 ÿ w Boston, pp. 103±122.
Moorthy, K.S., Png, I.P.L., 1992. Market segmentation,
( 2 ) cannibalization, and the timing of product introductions.
2
nl nh v2l vh Management Science 38 (3), 345±359.
w ÿ1 >0
4cnl vl Pisano, G., Rossi, S., 1994. ITT automotive global manufac-
turing strategy. HBS Case.
Pollack, A., 1992. Japan eases 57 varieties marketing. The New
but p1 is valid only when T < 1. When T P 1 with
York Times, 15 October.
a 0, from (17) we get p4 P p3 () vhl P 1=T . Rutenberg, D.P., 1969. Design commonality to reduce multi-
Therefore when a 0, S3 is not a dominant item inventory: Optimal depth of a product line. Operations
strategy. Research 19 (2), 491±509.
K. Kim, D. Chhajed / European Journal of Operational Research 125 (2000) 602±621 621
Rutenberg, D.P., Shaftel, T.L., 1971. Product design: Subas- Smith, P.G., Reinertsen, D.G., 1991. Developing Products in
semblies for multiple markets. Management Science 18 (4), Half the Time. Van Nostrand Reinhold, New York.
220±231. Stalk, G. Jr., Webber, A.M., 1993. JapanÕs dark side of time.
Sanchez, R., Mahoney, J.T., 1995. Modularity, ¯exibility, Harvard Business Review, July±August, pp. 93±102.
and knowledge management in product and organization Tversky, A., 1977. Features of similarity. Psychological Review
design. IEEE Engineering Management Review 25 (4), 84 (4), 327±352.
50. Ulrich, K.T., Tung, K., 1991. Fundamentals of product
Sanderson, S.W., Uzumeri, V., 1997. Managing Product modularity. Issues in design manufacturing/integration,
Family. Irwin, Homewood, IL. ASME DE-39, 73±79.