You are on page 1of 14

production

economics
ELSEVIER Int. J. Production Economics 56-57 (1998) 199-212

A decision support system for supplier selection using an integrated


analytic hierarchy process and linear programming
S.H. Ghodsypour*, C. O’Brien
Department oj’Manufacturing Engineering and Operations Management, Unioersiy of Nottingham, lJniuersi& Ptrrk,
Nottingham NG7 ZRD, UK

Received 20 June 1996; accepted 16 September 1996

Abstract
Supplier selection is a multicriteria problem which includes both qualitative and quantitative factors. In order to select
the best suppliers it is necessary to make a trade off between these tangible and intangible factors some of which may
conflict. When capacity constraints exist, this problem becomes more complicated as, in these circumstances, managers
should decide about two problems: which suppliers are the best and how much should be purchased from each selected
supplier. Some authors have applied mixed integer, goal and multi-objective programming to solve this problem. As
these techniques are mathematical, they have significant problems in considering qualitative factors which are very
important in supplier selection, especially when supplier partnership is desired. In this article an integration of
a analytical hierarchy process and linear programming is proposed to consider both tangible and intangible factors in
choosing the best suppliers and placing the optimum order quantities among them such that the total value of purchasing
(TVP) becomes maximum. This model can be applied to supplier selection with and without capacity constraints.
A numerical example is presented and the model advantages are discussed. (0 1998 Elsevier Science B V. All rights
reserved.

Keywords: AHP; Linear programming; Multicriteria technique; Supplier selection

1. Introduction Several factors affect a supplier’s performance.


Stamm and Golhar [2], Ellram [3] and Roa and
In most industries the cost of raw materials and Kiser [4] identified, respectively, 13, 18, and 60
component parts constitutes the main cost of criteria for supplier selection. Hence it is a multicri-
a product, such that in some cases it can account teria problem and it is necessary to make a trade off
for up to 70% Cl]. In such circumstances the pur- between conflicting tangible and intangible factors
chasing department can play a key role in cost to find the best suppliers.
reduction, and supplier selection is one of the im- Basically there are two kinds of supplier selection
portant functions of purchasing management. problem:
(1) Supplier selection when there is no con-
straint. In other words, all suppliers can satisfy the
*Corresponding author. Present address: Industrial Engineer- buyer’s requirements of demand, quality, delivery,
ing Dept., Amirkabir University of Technology, Tehran, Iran. etc.

0925-5273/98/$-see front matter 0 1998 Elsevier Science B.V. All rights reserved
PII: SO925-5273(97)00009-l
200 S.H. Ghodsypour, C. O’Brien /lnt. J. Production Economics 56-57 (1998) 199-212

(2) Supplier selection when there are some limita- In spite of the importance of multiple sourcing
tions in suppliers’ capacity, quality, etc. In other only a few articles have addressed this problem.
words, no one supplier can satisfy the buyer’s total Weber et al. [IS], who presented a review of 74
requirements and the buyer needs to purchase articles that discussed the supplier selection prob-
some part of his/her demand from one supplier and lem since 1966, stated that only ten articles applied
the other part from another supplier to compensate mathematical programming to vendor selection.
for the shortage of capacity or low quality of the Since their review seven other articles have used
first supplier. this technique. These 17 articles applied the follow-
In the first kind of supplier selection, one supplier ing methods: Four articles used linear program-
can satisfy all the buyer’s needs (Single Sourcing) ming [lo-131 while six others proposed a mixed
and the management needs to make only one deci- integer programming approach [14-191. Two arti-
sion, which supplier is the best, whereas in the cles used goal programming techniques [20,21]
second type of supplier selection, as no supplier can and Weber and Current [22] addressed multi-ob-
satisfy all the buyer’s requirements, more than one jective programming. Two others structured
supplier has to be selected (Multiple Sourcing). In the problem as non-linear programming [23, 241.
these circumstances management needs to make Current and Weber [25] explained that vendor
two decisions: which suppliers are the best, and selection problems may be formulated within the
how much should be purchased from each selected mathematical constructs of facility layout models.
supplier? Seshadri et al. [26] developed a probabilistic model
Several methods have been proposed in the to represent the connection between multiple
literature for single sourcing supplier selection. sourcing and its consequences, such as number of
The most important of them are discussed bids, the seller’s profit and the buyer’s price.
below. These articles can be divided into the following
The Linear Weighted Point method is the widely groups:
used approach for single sourcing. This approach (a) Single objective
uses a simple scoring method which heavily de- (b) Multiple objective.
pends on human judgement Cl]. The simplest Some authors used single objective techniques such
method is the Categorical method, which assigns as linear or mixed integer programming, in which
either good ( + ), neutral (0) or unsatisfactory ( - ) one criterion, usually the cost, is considered as the
to each defined criteria for all suppliers, and then objective function and other criteria are taken into
a total rate for each supplier is calculated. Although account as constraints [lo, 141. There are two
this approach is very simple, it depends heavily on problems with these techniques: the first is that the
human judgement and also weights the criteria criteria which are considered as constraints are
equally, which rarely happens in practice. The Cost weighted equally, which rarely happens in practice,
Ratio is a more complicated method in which the and the second problem is that they have signifi-
cost of each criterion is calculated as a percentage cant problems in considering qualitative factors.
of the total purchased value, and a net adjusted cost On the other hand, some other researchers ap-
for each vendor is determined. This approach is plied goal and multi-objective linear programming
complex and needs a lot of financial information to this problem [20-223. Although these techniques
[S]. The analytical hierarchy process (AHP) is an- can assign different weights to various criteria, they
other method which uses pairwise comparison and also have problems in involving qualitative factors.
it was applied by Narasimhan [6]. This approach Therefore both single and multiple objective pro-
is more accurate than the other scoring methods gramming applied to supplier selection have prob-
C7>81 lems in including qualitative criteria which are very
The most important point which is common to important in this decision making, especially for
these methods is that they all make some kind of supplier partnership policies [27,3,5]. In this paper
trade off between tangible and intangible factors to a decision support system (DSS) is developed to
find the best supplier. include both tangible and intangible factors for
S.H. Ghodsypour, C. O’Brien/M J. Production Economics 56 ~57 (1998) 199-212 201

choosing the best supplier and assigning the optimum Level 3: At this level operational integration of
order quantity between selected suppliers. The algo- buyer and supplier is desired. Usually for full imple-
rithm of this model is described in the next section. mentation of JIT/TQM this level of integration is
selected. Therefore not only the output character-
istics of the supplier should be considered, but the
2. An integrated method for supplier selection way in which these services are provided should
also be taken into account. For example, a supplier
The model presented in this article applies the with a highly flexible production line is preferred to
AHP, which uses pairwise comparison, to make the a supplier which keep a warehouse full of finished
trade off between tangible and intangible factors products to support JIT services. At this level of
and calculate a rating of suppliers, and then by integration the suppliers’ process capability such as
applying these ratings as coefficients of an objective set up time, lot size, lead time, etc. should be con-
function in linear programming allocates order sidered in the assessment process. From a quality
quantities between the suppliers such that the total point of view, if two suppliers have the same defects
value of purchasing (TVP) becomes a maximum. In rate, the supplier who achieved this level of quality
this model real quantitative data are used in AHP by the less inspection is preferred. In other words,
to improve the system’s consistency and simplify the supplier’s process capability indexes are impor-
the calculation of the model. Therefore the main tant.
steps of the algorithm are: Level 4: The integration for this situation is
2.1. Define the criteria for supplier selection. deeper than the previous levels. The corporate
2.2. Calculate the weights of the criteria. management decide to integrate process and
2.3. Rate the alternative suppliers. products wirh their supplier. In this situation
2.4. Compute the overall score of each supplier. besides the criteria of the preceding levels, the
2.5. Build the linear model and maximise the TVP. supplier’s human resource from several points of
view such as design involvement, management
ability, culture, etc. should be involved in the evalu-
2.1. Dclfne the criteria for supplier selection ation process.
Level 5: Business Partnership is desired at this
According to the level of buyer-supplier integra- level of integration. In these circumstances, the
tion, the company’s competitive situation and its strategic directions of the suppliers become very
corporate strategies, define the important criteria important, such that the best supplier from the
for supplier selection. As the levels of buyer-sup- technological and human resource point of view
plier relationship affects the number of the criteria might be inadequate if the supplier does not con-
considered, some important buyer-supplier integ- sider the direction of the clients business.
ration strategies and their criteria are discussed as After a definition of supplier selection criteria,
follows [28, 291. based on the corporate integration strategy, the
Level 1: At this level no integration can be as- hierarchy structure of the problem can be built.
sumed. Price and quality are important criteria for
supplier selection at this level.
Level 2: At this level logistical integration exists 2.2. Calculate the weights of the criteria
between buyer and supplier and the supplier has an
important role in the buyer’s competitiveness. Once the hierarchy has been structured the
Therefore, great importance is given to the sup- weights of the criteria should be calculated. By
pliers’ logistical performance. Hence in these using pairwise comparison the preference between
circumstances, besides the quality and price, the criteria will be asked from top and purchasing
operational logistics elements such as reliability, management. Then by applying Saaty’s l-9 scales,
flexibility, supply lots, lead time and so on should which are presented in Table 1, these preferences
be considered in the supplier selection process. should be quantified and a pairwise comparison
202 S.H. Ghodsypour, C. O’BrienlInt. J. Production Economics 56-57 (1998) 199-212

Table 1 The priority of alternative i to j for negative criteria,


Saaty’s 1-9 scale for AHP preference (Source: Saaty and such as cost, is equal to wj/wi, then the pairwise
Alexander [31])
comparison matrix is
Verbal judgement of preference Numerical rate

Equal importance 1
Weak importance of one over another 3
Essential or strong importance 5 ...
Demonstrated importance 7
Absolute importance 9
Intermediate values between the two 2, 4, 6, g
adjacent judgement
As this matrix is also consistent, the weight of
If activity i has one of the above numbers Reciprocal of elements are the normalised amounts of any col-
assigned to it when compared with above numbers umns, which is equal to the inverse normalised
activity j, then j has the reciprocal value amount of the alternatives:
when compared with i
Weight of ith element (for negative criteria) =
1
matrix structured. After that, the weights of criteria
and a Consistency Ratio (CR) can be calculated [30]. ”
0
-
wi

1 .
i=l wi
c C-3
2.3. Rate the alternative suppliers Hence, this step can be rewritten as:
- Are criteria either qualitative or quantitative
In this model for rating suppliers the real quant- with nonlinear preferences? If yes apply Saaty’s
itative available data can be preserved, i.e. in the l-9 scale pairwise comparison to calculate their
quantification process of pairwise comparison, any weights.
real existing data are applied as quantitative cri- - Are criteria positive? Use their normalised
teria and Saaty’s l-9 scales are used for qualitative amounts as their ratings.
criteria. Therefore, the question for pairwise com- - If they are negative apply their inverse nor-
parison for quantitative criteria is L-321: malised amounts as their weights.
“Of two element i and j, how many times i is
preferred to j”
If the values for alternative i and j are, respecti- 2.4. Compute the overall score of each supplier
vely, wi and wj, the preference of alternative i to j is
equal to wi/wY Therefore, the pairwise comparison By combining the weight of criteria and sup-
matrix is pliers’ rating, determine the final score of each
supplier. If there are no constraints choose the
W&l W,I% ... WIIW,, maximum score supplier and buy all demand from
W&i WZIWZ... WZIW,, this supplier and stop. Otherwise go to step 2.5.
...

%IW1 w,/wz . . . WJW,. 2.5. Build the linear model

As this matrix is consistent [31], the weight of each


If there are some constraints such as suppliers’
element is its relative normalised amount:
capacity, quality, etc. use the suppliers’ ratings as
wi
Weight of ith element = 7. coefficients of an objective function in linear pro-
gramming to assign order quantities to the sup-
C wi
i=l
pliers such that the total value of purchasing (TVP)
S.H. Ghodsypour, C. O’Brien jlnt. J. Production Economics 56. 57 (1998) 199-212 203

becomes maximum. The objective function and of the ith vendor, the quality constraint can be
constraints of this linear programming are as fol- shown as
lows:
~ Xiqi ~ QD-
i=l
2.5.1. Notations
Ri final ratings of ith supplier
2.5.4 Final model
Xi Order quantity for ith supplier
The final integrated linear programming model
Vi Capacity of ith supplier
can be shown as
D Demand for the period
qi Defect percent of ith supplier
Max(TVP) = i R,X,.
Q Buyer’s maximum acceptable defect rate I=,
Subject to:
2.5.2. Objective function
As Ri and Xi, respectively, denote the ratings and
i Xi = D (demand constraint),
the numbers of purchased units from the ith sup- i=l
plier and maximising the total value of purchasing
is desired, the objective function is i Xi qi < QD (aggregate quality constraint),
i=l

Max (TVP) = i R,X,. xi d vi, i = 1, 2, . . . ) n


i=l (supplier’s capacity constraints),

2.5.3. Constraints Xi30, i=l,2, ... . n


The important constraints of the problem are (nonnegativity constraint).
supplier capacity, buyer’s demand and quality [ 18, This algorithm is shown in Fig. 1.
24, 20, lo], which are formulated as follows:
Capacity constraint: As vendor i can provide up
to l/i units of the product and its order quantity 3. Numerical example
(Xi) should be equal or less than its capacity, these
constraints are: Assume that the management of a JI’T manufac-
turer decide to choose their best suppliers and
Xi< Vi, i= 1,2 . . n.
assign their optimum order quantities to maximise
the TVP. The main criteria for supplier selection
On the other hand, aggregate suppliers’ capacity
should be equal or greater than demand, there- are Cost, Quality and Service. According to the
fore, corporate strategies the Quality includes Defects
(Def.) and Process capability (Pro.C.) while Service
involves On-time delivery (Ont.), Response to cha-
i V;>D. nges (Res.) and Process flexibility (Pro.F.). Four
i=l
suppliers are included in the evaluation process and
Demand constraint: As sum of the assigned order their Cost, Quality, Ont. and Capacity are pres-
quantities to n vendors should meet the buyer’s ented in table two.
demand, it can be stated that Suppose the buyer wishes to find the best sup-
pliers and their optimum order quantities, if the
demand is 1000 units and the maximum acceptable
f: Xi = D. defect rate is 0.02.
i=l
In order to solve this problem two types of calcu-
Quality constraint: Since Q is the buyer’s max- lations should be carried out: AHP and linear pro-
imum acceptable defect rate and qi is the defect rate gramming (LP) optimisation. The first is performed
204 S.H. Ghodsypour, C. O’Brien /Ini. J. Production Economics 56-57 (1998) 199-212

Define the criteria for supplier selection.


+
Calculate the weights of criteria

t Yes
Build the linear Programming model to find the best suppliers and
their optimum order quantities

Fig. 1. Supplier selection algorithm.

Table 2 Step 2: The priority of criteria are calculated


Suppliers’ quantitative information by using the pairwise function from the Assess-
ment menu in EC software, Fig. 3 shows the pair-
cost Quality On-time del. Capacity
wise preference and weight of cost, quality and
Supplier 1 30 0.03 0.95 400 service.
Supplier 2 40 0.05 0.98 700 Step 3: Suppliers’ rating are calculated through
Supplier 3 50 0.01 0.85 600 the three following stages:
Supplier 4 45 0.06 0.92 500 (3.1) As process capability, ease of communica-
tion, response to change and process flexibility are
qualitative factors, their priorities should be cal-
using Expert Choice (EC) software and second one culated by using the same approach as step 2.
is achieved by using Microsoft Excel Solver. After (3.2) Since on-time delivery is a positive quantit-
calculating the final score of each supplier with EC ative factor, the suppliers’ ratings are given by their
the data are exported to Microsoft Excel by using normalised performances which are calculated by
copy and paste windows and then by applying applying the Data function from the Assessment
a function called Solver from the Tools menu the menu.
optimisation process can be executed. (3.3) As cost and defects (Def.) are negative
For simplicity, the solution is described accord- quantitative factors, the suppliers’ ratings are their
ing to the algorithm steps: normalised inverse performances which are deter-
Step 1: According to the defined criteria, the mined by employing the Inversion from Data func-
hierarchy structure of the problem is as shown in tion in the Assessment menu. Fig. 4 shows the
Fig. 2. weighting of the costs for each supplier.
S.H. Ghodsypour, C. O’Brien JIM. J. Production Economics 56. 5 7 (1998) 199-212 205

Fig. 2. Supplier selection criteria.

Compare the relative PREFERENCE with respect to: GOAL

QUALITY SERVICE
COST 40 30
QUALITY 2.0

Row element is __ times more than column element unless enclosed in ()

COST ,630

QUALITY .218 -

SERVICE ,151 -

Fig. 3. Preferences and weights of cost, quality and service.

Data with respect to COST < GOAL ming should be maximised:

Max. TVP = 0.297X, + 0.252X, + 0.268X,

+ 0.183X,

Subject to:

0.03X, + 0.05X, + 0.01X, + 0.06X, 5, 20


Fig. 4. Cost weights for suppliers.
X,+ x,+ x,+ x, == 1000

X1 5: 400
Step 4: In order to calculate the final score of
each supplier the weights of criteria, subcriteria and X, <; 700
suppliers’ ratings should be combined. This is per-
formed by using the synthesis menu and the results X, $5 600
are shown in Fig. 5.
X 4 5: 500
Step 5: In order to find the best order quantities,
the TVP which is shown in the following program- Xi > 0, i = 1, 2, 3, 4
206 S.H. Ghodsypour, C. OBrienlInt. J. Production Economics 56-57 (1998) 199-212

Synthesis ofLeaf Nodes with respect to GCN


DistributiveMC&
OVER’UINCONSISTENCY INDEX = Cl.05

s.1 297
s.3 268
s.2 .252

Fig. 5. Suppliers’ final rating.

As stated earlier, this LP problem is solved using 4. Sensitivity analysis


Solver from Microsoft Excel. Table 3 shows the
Answer Report of the Microsoft Excel Solver which Sensitivity analysis identifies the impact of cha-
indicate that the best suppliers and their optimum nges in the priority of criteria on the suppliers’
order quantities are respectively: Sl-400, S2-0, performance and order quantities.
S3-600, S4-0. Gradient Sensitivity Analysis (GSA) of Expert
Choice (EC), which is shown in Fig. 6 represents the
variation of suppliers’ rating to changes in cost
Table 3
Solver answer report
priority. It illustrates that if the cost priority, which
is 0.63, increases to 0.720 or decreases to 0.524, the
Microsoft Excel 5.0~ Answer Report suppliers’ rating do not change. It can be seen that
Worksheet: [VENDOR3.XLS]supplier (3) within the five following areas of GSA the suppliers’
Target Cell (Max.) ratings are not sensitive:
Cell Name Original value Final value
Area Cost priority Suppliers’ ratings
%F$lO 4.02924E-07 279.6
(1) 0.00-0.183 R, > R, > R, > R,
Adjustable Cells (2) 0.183-0.524 R, > R, > R, > R,
(3) 0.524-0.720 R, > R, > R, > R,
Cell Name Original value Final value 0.720.925 R, > R, > R, > R,
(4)
(5) 0.925-1.000 R, > R, > R, > R,
$A$9 Xl 0 400
%B$9 x2 0 0
$C$9 x3 0 600 In order to find the sensitivity of the order quant-
$D$9 x4 0 0 ities to the weighting for cost, it is necessary to
investigate how sensitive the optimal solution of
Constraints the linear programming (LP) is to changes in the
Cell Name Cell Formula Status Slack
cost priority. The sequence of changes in this prob-
value lem is different from the sequence of changes in
general sensitivity analysis because in this problem
%F$ll 18 $F$ll <%G$17 Not Binding 2 variations of cost priority affect the suppliers’ rat-
%F$9 loo0 $F$9 = $G%15 Binding 0
ings which are the coefficients of the objective func-
$B$9 X2 0 $B$9 <700 Not Binding 700
$C$9 x3 600 $C%9 6600 Binding 0
tion of LP and they can have an influence on order
$D$9 X4 0 $D$9 6500 Not Binding 500 quantities which are optimised through the solu-
$A$9 Xl 400 %A%9C 400 Binding 0 tion of LP. This sequence is shown in Fig. 7.
$A$9 Xl 400 $A$9 30 Not binding 400 This analysis therefore cannot be performed
$B$9 X2 0 $B$9 20 Binding 0 with general packages of LP such as LINDO or
$C%9 x3 600 $C$9 30 Not binding 600
$D$9 X4 0 $D$9>0 Binding 0 SOLVER which consider the variation of one coef-
ficient while the others are constant whereas in this
S.H. Ghodsypour, C. O’BrienlInt. J. Production Economics 56- 57 (1998) 199-212 207

Gradient Sensitivity w.r.t. GOAL for nodes below GOAL


AIt+
40

-_ _-___
-e-m
--- ._
---.__
----_. -- S.l
___-----
1. --_._ _--- _---
30 --___------
__
--..____
_--- ----- -.
-----_____ ---------c__-__-_________ ._---z-~____________
---
-____ !j. 2

-;- _ -. s.4
___-- -.__
20 _ _ .-- - - s.3
- -.
_ _ -- ._-----
_ -- z -
-. *
.- - -. .- --

10

O0 , I I I I 8 I * I . I I I

0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1
Priority of COST

Fig. 6. Sensitivity of supplier’s performance on cost

Variation in cost priority


+
Variation in suppliers’ ratings which are the
coefficients of the objective function of LP

Variation in order quantities which are


optimised through the solution of LP

Fig. 7. Sequence of variation in sensitivity analysis.

problem variation of cost priority changes all the x,+ x,+ x,+ x, ==1000
coefficients of the objective function.
X, :g 400
In order to solve this sensitivity analysis, para-
metric linear programming (PLP) has been applied, :$700
X2
such that the coefficients of the objective function
are considered as parameters, R,, R,, R, and R,. X3 :$ 600
Therefore the PLP programming becomes:
x, d 500
Max. TVP = R,X, + R,X, + R,X, + R,X, xi 3 0, i = 1, 2, 3, 4

This PLP problem can be solved for each area of


Subject to:
GSA by using simplex method and considering the
0.03X, + 0.05X, + 0.01X, + 0.06X_, d 20 relation between Rj’s. For instance for area three, in
208 S.H. Ghodsypour, C. O’Brien/M. J. Production Economics 56-57 (1998) 199-212

which the relation between Ri's is: R, > R, > R, Fig. 8 shows the sensitivity of order quantities to the
> R,,the optimum solution is: X, = 400, X, = 0, variation in cost priority for all five area of GSA.
X, = 600 and X, = 0 (the simplex tables are Considering the sensitivity of order quantities to
shown in appendix). the buyer maximum acceptable defect rate, assume
This means that as long as the relation be- that maximum acceptable defect rate (Q) changes
tween Ri'sis: R, > R, > R, > R, the optimal from 0.02 to 0.10. This does not affect the optimum
solution does not change. In other words, order order quantities for area three which are 400,0,600
quantities are not sensitive to the weighting of and 0, because the slack variable for the quality
cost in the interval of 0.5240.720, which is constraint, which is illustrated in Table 3, is equal
area three of the gradient sensitivity analysis to 2. But as Table 4 shows the slack variable for
(GSA). quality constraint in area 4 is zero, which means
By applying the same approach for other area of that increasing the Q changes the optimum order
GSA, 1, 2, 3,4 and 5, it can be seen that the order quantities. By using PLP for this area the best
quantities are not sensitive within each of the areas. answers respectively are: 400, 600, 0 and 0. The
The optimum order quantities for each area are as optimum order quantities for supplier selection,
follows: when the Q is increased to 0.10 units, for all areas of
GSA are as follows:
Area X1 XZ X3 X4 Suppliers’ ratings
Area X, XZ X3 X4 Suppliers’ ratings
(1) 300 100 600 0 R3 >RZ>R1 >R, (1) 0 400 600 0 R,>R,>R, >R,
(2) 400 0 600 0 R3 >R, >R, P-R, (2) 400 0 600 0 R,>R, >R,>R,
(3) 400 0 600 0 RI >R3 >R, >R4 (3) 400 0 600 0 RI >R, >R, >R,
(4) 400 50 550 0 RI >R, >R, >R, (4) 400 600 0 0 RI >R,>R, >R,
(5) 400 50 550 0 RI >R,>R,>R, (5) 400 600 0 0 RI >R2 >R,>R,

s-1

5.2

s.4
s.3

-10 -

x,=300 x ,=400 x ,=400


- XI=100 x.=0 X2=50
X,=600 X,=600 X,=550
X,=0 x ,=o x,=0
.ooo ’ ’ ‘.I ! ’ . ’ . ’ ’ ’ ’ r . ’ ’
.1 .2 .3 .4 .7 .8 .9 1
PrioritymZfCOST.6

Fig. 8. Sensitivity of order quantities to cost priority for all area of GSA.
S.H. Ghodsypour, C. O’Brien /Int. J. Production Economics 56 57 (I 998) 199-212 209

Table 4
Solver Answer report for area four

Microsoft Excel 5.0~ Answer Report Worksheet: [VENDOR3.XLS] supplier (4)


Report Created: 6/11/95 10~52 Target Cell (Max.)

Cell Name Original value Final value

$FSlO 0 265.9500003

Adjustable Cells

Cell Name Original value Final value

$A$9 Xl 0 400
$B$9 X2 0 SO
KS9 x3 0 550
$D$9 x4 0 0

Constraints

Cell Name Cell value Formula Status Slack

SFSI I 20.00000011 $FSll < $G$17 Binding 0


$FS9 1000 $F$9 = $G$15 Binding 0
$BS9 x2 50 $B$9 < 700 Not binding 650
SCS9 x3 550 $C$9 < 600 Not binding 50
SDS9 x4 0 SDS.9 < 500 Not binding 500
$A$9 Xl 400 SAS.9 < 400 Binding 0
$A$9 Xl 400 $A$9 > 0 Not binding 400
$BS9 x2 50 SBS9 > 0 Not binding 50
SCS9 x3 550 KS9 > 0 Not binding 550
$DS9 x4 0 SD%9 2 0 Binding 0

Suppose that besides increasing the maximum de- quality, therefore the buyer selects one supplier
fect rate, the demand is decreased to 300 units, in (Single Sourcing), which is the maximum score
these circumstances the best answers for all five vendor and all demand should be purchased from
areas of the GSA are: this supplier. This result can be seen from optimum
solution of LP for all area of GSA.
Area X, Xz X3 X4 Suppliers’ ratings In the case of D = 1000 and Q = 0.10, although
300 0 R, > R2 > RI > R4 all suppliers can satisfy the quality constraint, de-
(1) 0 0
R3 > RI > R2 > R4 mand remains unsatisfied. In these circumstances
(2) 0 0 300 0
(3) 300 0 0 0 R, > R3 > R2 I=-R4 the buyer should purchase all the capacity of the
RI > R2 > R3 > R4 maximum score supplier and then the next in line
(4) 300 0 0 0
(5) 300 0 0 0 RI > R2 > R4 > R3 until the demand becomes satisfied. The optimal
solution of the model shows this result for all area
Fig. 9 shows the sensitivity of suppliers’ perfor- of GSA.
mance and order quantities to the cost priority for In the case of D = 1000 and Q = 0.02, the buyer
all areas of the GSA and also for the three discussed should purchase from the suppliers, who satisfy not
situations: D = 300 and Q = 0.10, D = 1000 and only the demand but also the quality constraints.
Q = 0.10, D = 1000 and Q = 0.02. That is the order quantity for area four changes
In the case of D = 300 and Q = 0.10, all suppliers from 400, 600, 0, 0 to 400, 50, 550, 0. This reduces
can satisfy the buyer’s requirements of demand and the delivered defect number from 42 to 20 units
210 S.H. GhodFypour, C. O’Brien JInt. J. Production Economics 56-57 (1998) 199-212

52

s-4
s.3

D -300. Q 20.10

D=lOOO,Q=O.lO

x,=300 x,=400 x,400


Xz-loo x,=50 D =I 000. Q =0.02
Xx*
x,=600 x,=600 x,=550
&=p I * I , I , x+=0 ( I , / , X4?
.m
.l .2 .3 .4 .7 .8 .9 1
Racily.: C0ST.6
Fig. 9. Sensitivity of order quantities and suppliers’ performance on the cost priority.

which means that although the second supplier has establish good linkage between supplier selection
the lower price and better service than the supplier and the buyer company’s policy. This model as-
three, the buyer prefers to buy from the third sup- signs order quantities to suppliers such that the
plier for its better quality. total value of purchasing (TVP) becomes maximum
using an integrated AHP and linear programming.
This model also enables the management to make
6. Summary and conclusion a trade off between several tangible and intangible
factors with different priorities.
Increasing attention toward supplier partnership Although this method of integrating AHP and
not only raises the importance of supplier selection linear programming is applied to supplier selection
but also increases the significance of considering for the first time, the approach has previously been
qualitative factors in this decision-making process. used by some authors in other areas, for instance by
Taking the suppliers’ capacity constraints and the Korpela [33] for optimising logistical structures
buyers’ aggregate quality and service limitations and by Korhonen and Wallenius [34] for formula-
into account makes the problem complicated. As ting a marketing strategy, by Gass [35] for a long-
the policy of supplier partnership varies in different Scale personnel planning model, by Olson [36] for
strategies, a flexible method which can reflect the an export planning model for a developing country
corporate strategy in the supplier selection process and by Liberatore [37] for industrial R & D project
and consider the suppliers’ and buyers’ limitations selection and resource allocation.
and involve both tangible and intangible factors is The advantages of this method for supplier selec-
required. A dynamic TVP model is suggested to tion are:
S.H.Ghodsypour,
C. O'Brien/Int.
J.Production
Economics
56 -57(1998j
199-212 211

(1) both tangible and intangible factors which are (6) analysis of several “what-if” scenarios is
very important in supplier selection can be included facilitated, for example; what happens if the
in a multiple sourcing policy, while existing models weight of some criteria changes, or another supplier
can only consider the quantitative factors. joins the system, or another criteria is added or
(2) corporate strategies can be reflected in pur- omitted.
chasing activities.
(3) using real data, the calculation is simplified
and the system’s consistency is improved. Acknowledgements
(4) using pairwise comparison reduces depend-
ency of the system on human judgement. Mr. Ghodsypour would like to acknowledge the
(5) both weight of criteria and rank of suppliers Iranian government for the financial support of his
are determined by one systematic approach. research.

Appendix

The simplex tables of PLP for area 3 (RI > R3 > R2 > R4) (Table 5).

Table 5

Basic Coefficient of Right


var. srde
X, X* X, X4 X5 X, xi X8 x, X,0

* -CR, + fa -(R, + w - (RJ + W -(R4 + M) 0 0 0 0 0 0 - 1000.M


X, 1 0 0 0 1 0 0 0 0 0 400
X, 0 I 0 0 0 1 0 0 0 0 700
XT 0 0 1 0 0 0 1 0 0 0 600
X" 0 0 0 I 0 0 0 IO 0 500
X, 0.03 0.05 0.01 0.06 0 0 0 0 1 0 20
x,0 1 I 1 1 0 0 0 0 0 I IO00

I 0 - (Rz + M) -(R3+W -(R,+M) R,+M 0 0 0 0 0 400R,~ 600M


X, 1 0 0 0 I 0 0 0 0 0 400
X, 0 I 0 0 0 IO 0 0 0 700
XT 0 0 1 0 0 0 IO 0 0 60(1
X8 0 0 0 I 0 0 0 IO 0 500
X, 0 0.05 0.01 0.06 - 0.03 0 0 0 1 0 t*
x 10 0 1 I 1 -1 0 0 0 0 1 600

z 0 RJ - RZ 0 RJ - & R,-R3 0 0 0 0 R,+M 40OR,+600R,


X, 1 0 0 0 1 0 0 0 0 0 400
X, 0 I 0 0 0 IO 0 0 0 700
X, 0 -I 0 -I I 0 I 0 0 --I 0
X8 0 0 0 1 0 0 0 I 0 0 500
X, 0 0.04 0 0.05 -0.02 0 0 0 I ---
0.01 :!
X3 0 I 1 I -I 0 0 0 0 I 600

References [2] C.L. Stamm, D.Y. Golhar, 1993. JIT purchasing: Attribute
classification and literature review. Prod. Planning Con-
[l] A. Ghobadian, A. Stainer, T. Kiss, 1993. A computerised trol 4(3), 273-282.
vendor rating system. Proc. 1st Internat. Symp. Logistics, [3] L.M. Ellram, 1990. The supplier selection decision in stra-
pp. 321-32X. tegic partnerships. J. Purchasing Mater. M,gmt. 26(4), 8-14.
212 S.H. Ghodsypour, C. O’BrienlInt. J. Production Economics 56-57 (1998) 199-212

[4] C.P. Roa, G.E. Kiser, 1980. Educational buyers’ percep- [23] W.C. Benton, 1991. Quantity discount decision under
tions of vendor attributes. J. Purchasing Mater. Mgmt. conditions of multiple items, multiple suppliers and re-
16,25-30. source limitation. Int. J. Prod. Res. 29(10), 1953-1961.
[S] E. Timmerman, 1986. An approach to vendor performance [24] J.D. Hong, J.C. Hayya, 1992. Just-in-time purchasing:
evaluation. J. Purchasing Mater. Mgmt. Winter, 2-8. Single or multiple sourcing? Int. J. Prod. Econom. 27,
[6] R. Narasimhan, 1983. An analytical approach to supplier 1755181.
selection. J. Purchasing Mater. Mgmt. Winter, 27-32. [25] J. Current, C. Weber, 1994. Application of facility location
[7] S.J.E. Smith, 1994. Productivity measurement and capital modelling constructs to vendor selection. Eur. J. Oper.
investment appraisal in electronics design. Unpublished Res. 76(3), 387-392.
Ph.D. Thesis, Nottingham Univ. [26] S. Seshadri, K. Chatterjee, G.L. Lilien, 1991. Multiple
[S] K. Swann, 1988. Investment in AMT a review. Prod. Eng. source procurement competitions. Marketing Sci. 10(3),
October, 53-57. 246-253.
[9] C.A. Weber, J.R. Current, W.C. Benton, 1991. Vendor [27] D.Y. Golhar, C.L. Stamm, S. Banerjee, 1993. Just-in-time
selection criteria and methods. Eur. J. Oper. Res. 50, purchasing: An empirical investigation, Prod. Planning
2-18. Control 4(4), 392-398.
[IO] A.C. Pan, 1989. Allocation of order quantity among sup- [28] C. Masella, A. Rangone, 1995. Managing supplier/cus-
pliers. J. Purchasing Mater. Mgmt. Fall, 3639. tomer relationships by performance measurement systems.
[ll] B.G. Kingsman, 1986. Purchasing raw materials with Proc. 2nd Internat. Symp. Logis., pp. 955102.
uncertain fluctuating prices. Eur. J. Oper. Res. 25, [29] G. Merli, 1991. Co-makership, The New Supply Strategy
358-372. for Manufacturers, Chapter 6. Productivity Press,
[12] T.F. Anthony, F.P. Buffa, 1977. Strategic purchasing Cambridge.
scheduling, J. Purchasing Mater. Mgmt. Fall, 27-31. [30] D.R. Anderson, D.J. Sweeney, T.A. Williams, 1994. An
[13] D.L. Moore, H.E. Fearon, 1973. Computer-assisted deci- Introduction To Management Science Quantitative Ap-
sion-making in purchasing. J. Purchasing (9), 5-25. proaches to Decision Making, 7th Ed. West Publishing
[14] A.A. Gaballa, 1974. Minimum cost allocation of tenders, Company, USA.
Oper. Res. Quart. 25(3), 389-398. [31] T.L. Saaty, J.M. Alexander, 1981. Thinking With Models:
[15] P.S. Bender, R.W. Brown, H. Isaac, J.F. Shapiro, 1985. Mathematical Models in the Physical, Biological and
Improving purchasing productivity at IBM with a Social Sciences, Chapter 8, Pergamon Press, London.
normative decision support system. Interfaces 15, [32] F.S. Weber, 1993. A modified analytic hierarchy process
106-l 15. for automated manufacturing decisions. Interfaces 23(4),
[16] I. Turner, 1988. An independent system for the evaluation 75-84.
of contract tenders. Oper. Res. Sot. 39(6), 551-561. [33] J. Korpela, 1994. An analytic approach to distribution
Cl73 R. Narasimhan, K. Stoynoff, 1986. Optimizing aggregate logistics strategic management, Published Ph.D. Thesis
procurement allocation decisions. J. Purchasing Mater. Lappeenranta University of Technology Lappeenranta,
Mgmt. Spring, 23-30. Finland.
[18] S.S. Chaudhry, F.G. Forst, J.L. Zydiak, 1993. Vendor [34] P. Korhonen, J. Wallenius, 1990. Using quantitative data
selection with price breaks. Eur. J. Oper. Res. 70(l), in multiple objective programming. Eur. J. Oper. Res. 48,
52-66. 81-87.
Cl93 E.C. Rosenthal, J.L. Zydiak, S.S. Chaudhry, 1995. Vendor [35] S.I. Gass, 1986. A process for determining priorities and
selection with bundling. Decision Sci. 26(l), 3548. weights for large-scale linear goal programming. J. Oper.
[20] F.P. Buffa, W.M. Jackson, 1983. A goal programming Res. Sot. 37(8), 779-785.
model for purchase planning. J. Purchasing Mater. Mgmt. [36] D.L. Olson, M. Venkataramanan, J.L. Mote, 1986. A tech-
Fall, 27-34. nique using analytical hierarchy process in multiobjective
[21] D. Sharma, W.C. Benton, R. Srivastava, 1989. Competitive planning model. Socio-Econom. Planning Sot. 20(6),
strategy and purchasing decision. Proc 1989 Ann. Conf. of 361-368.
the Decision Sciences Institute, pp. 1088~1090. [37] M.J. Liberatore, 1987. An extension of the analytic hier-
[22] C.A. Weber, J.R. Current, 1993. A multiobjective approach archy process For industrial R&D project selection and
to vendor selection. Eur. J. Oper. Res. 68, 1733184. resource allocation. IEEE Trans. Eng. Mgmt. EM-34. (1).

You might also like