Professional Documents
Culture Documents
Electronic Purchasing: Determining The Optimal Roll-Out Strategy
Electronic Purchasing: Determining The Optimal Roll-Out Strategy
+
'
t
t t t
t
* i + I v
*
*
*
#a( (*)
Aith the following definitions of the variables1
i
t
is the state of the /P syste# at the end of 9eriod t& This state consists of the values
of c
j
and d
in 9eriod t and it de9ends on the state at the end of the 9revious 9eriod
i
t&1
and decision #ade in 9eriod t& !urther#ore i
,
is the initial state (all c
j
and d
are
-)&
*D
*
t
is the decision what to add to the /P syste# in 9eriod t& 'f course only
co##odities or de9art#ents for which c
j
or d
'
* 1 1
* 1 1
L L L
L L L
#a(
*
*
#a(
*
*
*
t
j
j
j
t j
j
t
j j
t
t t t
d * "! # i c
c * !! # i d
* i + ())
0ote that the fi(ed costs I is just a constant subtracted fro# the 9rofit and therefore I has
no influence on the o9ti#al roll%out strategy&
4.. /sing dynamic programming
The nu#ber of 9ossible i#9le#entation se=uences is 'M #a:ing the calculation of all
values of v nearly i#9ossible tas: already for s#all values of '& However the 9roble#
(*) can be rewritten into a finite deter#inistic dyna#ic 9rogra##ing (?P) 9roble# with
bac:wards recursion& Ae define v
t
(i
t
) as the #a(i#u# total revenues that can be
obtained in state i
t
at the end of 9eriod t& v
t
(i
t
) can be calculated recursively1
( ) ( ) ( ) ( )
* * * *
#a(
*
+ + + +
+
+
t t t t t
*
t t
i v * i + i v
t
(3)
*+
.t the end of 9eriod ' there is only one 9ossible state (all c
j
and d
e=ual to *) and no
decision o9tions are left hence1
-
'
v
(4)
There is also only one initial state i
,
and with the recursive relation above v
,
(i
,
)&I will be
the #a(i#u# total 9rofit and the decisions *
t
that give rise to this #a(i#u# value
deter#ine the o9ti#al order of i#9le#entation&
The nu#ber of states at the end of 9eriod t is
,
_
t
'
and there are ' 0 t decision o9tions
left& Thus for deter#ining the #a(i#u# 9rofit
( )
,
_
'
t
t
'
t '
-
calculations have to be
#ade an e(9ression that can be si#9lified1
( )
'
'
t
'
t
'
t
'
t
t
'
t ' )
)
*
- -
,
_
,
_
(using
( )
,
_
,
_
,
_
t
'
'
t '
'
t '
t
'
t
and
'
'
t
t
'
)
-
,
_
) (,)
.bove the restriction is used that all co##odities and de9art#ents have to be
i#9le#ented& I#9le#enting certain co##odities and de9art#ents could be non%
9rofitable though hence e(cluding the# would increase the overall 9rofit& 'bviously
this would be the case when the i#9le#entation costs for a co##odity j
1
or a
de9art#ent
1
are larger than the revenues gained by adding j
1
or
1
to the /P syste#1
*5
L L
#a(
*
j
j
!! # <
or L L
#a(
*
j
j
j
"! # <
(D)
.ll non%9rofitable i#9le#entations are found at the end of the o9ti#al i#9le#entation
order& $ecause of the discount factor losses are 9ost9oned& Hence the 9rofit is
#a(i#ised when recursively all co##odities and de9art#ents are e(cluded that are
i#9le#ented fro# 9eriod t
1
until '& This 9eriod t
1
is deter#ined by starting fro# v
'
(i
'
)
and going bac:wards in the o9ti#al order until
( )
L L
t t
i v
beco#es negative& This leads to
the o9ti#al roll%out strategy without the restriction that all co##odities and de9art#ents
have to be i#9le#ented&
%. &alculating the optimal roll-out strategy
Using the ?P for#ulation of the 9revious section o9ti#al /P roll%out strategies can only
be calculated on a co#9uter within reasonable ti#e li#its for values of ' u9 to ), or so&
This #eans for 9roble#s with larger ' heuristics are re=uired to deter#ine a good
a99ro(i#ation of the o9ti#al strategy&
2.1 3euristics
.s an a99ro(i#ation we defined a ;greedy #%ste9; heuristic1 loo:ing m ste9s ahead&
In this heuristic at the end of each 9eriod t the decision *
t)1
1
will be ta:en that #a(i#ises
the 9rofit w
t
over m ste9s1
( ) ( ) ( )
,
_
,
_
,
_
+ + +
+ + +
+ + + + +
+ + + +
m t m t m t
*
m
*
t t t
*
t t t
*
t
* i + * i + * i + w
m t t t t
#a( &&& #a( #a( #a(
*
* )
) * ) * *
) ) *
(+)
*6
The total 9rofit v will be1
( )
+
'
t
t t t
t
* i + I v
*
L
*
*
(5)
0ote that for the greedy *%ste9 heuristic (+) and (5) can be ta:en together1
( )
+
'
t
t t t
*
t
* i + I v
t
*
*
*
#a( (6)
This greedy *%ste9 heuristic is si#9ly ta:ing the best available o9tion at every ste9
without loo:ing at its conse=uences for further ste9s
0aturally it should hold that ' m * & In .99endi( $ is 9roven that below a certain
threshold value
4
of the discount factor the greedy #%ste9 heuristic 9rovides the
o9ti#al solution&
The nu#ber of calculations needed for the heuristic in each 9eriod t is1
( ) ( ) ( ) ( )
+ + +
*
-
* ) *
m
i
i t ' m t ' t ' t ' (*-)
!or the total nu#ber of calculations it has been 9roven in .99endi( B that1
( ) ( )
+
+
+
m
i
'
t
m
i
i '
m
i t '
- *
*
-
*
*
*
* (**)
!or the *%ste9 heuristic this boils down to
( ) *
)
*
+ ' '
calculations and for the )%ste9
heuristic to
( ) ( ) * *
3
*
+ ' ' '
calculations& 0ote that for large m calculating the ?P
9roble# could be faster&
)-
The *%ste9 heuristic has an obvious flaw as in the first 9eriod no cost savings will occur
(only having either a de9art#ent or a co##odity connected to the /P syste#) hence the
co##odity or de9art#ent with the lowest i#9le#entation costs would be chosen& Here
a large i#9rove#ent can be #ade by loo:ing two ste9s ahead as in the second 9eriod
the first savings occur& Bonsidering reasonable calculation ti#e the )%ste9 heuristic can
be used for 0 even larger than *-- #a:ing it a99licable for #ost 9ractical situations& If
not all co##odities and de9art#ents have to be i#9le#ented the non%9rofitable 9arts
can be re#oved at the end of the i#9le#entation order found by the heuristic (as
described at the end of section 4) hence increasing the #a(i#u# 9rofit&
2.$ 5 numerical example
To illustrate how the #odel can be used for 9ractical calculations a s#all scale e(a#9le
is given below with seven co##odity grou9s and si( de9art#ents to be connected to the
/P syste#& The data in the e(a#9les is based on realistic values& !i(ed costs of US F-&D
"illion are assu#ed& !urther#ore the costs 9er co##odity grou9 are assu#ed to be in
the order of US F *----- and the costs 9er de9art#ent around US F 3----& These
a#ounts include discounted #aintenance costs for five years& The variable costs are
shown in Table *& !or the si( co##odity grou9s we assu#e around *,--- transactions
9er year and a s9end of US F), "illion& .lso we assu#e around US F,- can be saved
on average 9er transaction which #eans around US F-&+, "illion 9er year& Aith an
average of ),C #averic: buying that will decrease to ,C with the /P syste# and
assu#ing *-C higher 9rices with #averic: 9urchases the savings 9er year are around
US F-&, "illion& Aith discounted su##ation over five years this leads to an esti#ate of
)*
US FD "illion in total cost savings and these savings have been divided over the savings
9er co##odity grou9 9er de9art#ent in Table )& !inally a discount factor 9er year of
-&5 was ta:en& .ssu#ing 9rojects of 3 #onths this leads to N -&64D&
Table 3 shows the o9ti#al roll%out strategy& The last two 9eriods are 9ut between
brac:ets as they are non%9rofitable and therefore should be e(cluded fro# the
i#9le#entation& Table 4 gives a co#9arison of five roll%out strategies regarding1 the
9rofit i#9le#entation order and the total i#9le#entation costs& They vary between the
strategies as the costs are distributed differently over the total i#9le#entation 9eriod&
The worst case was found by #ini#ising instead of #a(i#ising the total 9rofit& In this
case the #a(i#u# 9rofit is 46C higher than the #ini#u# 9rofit& !urther#ore the
greedy heuristics a99ro(i#ate the o9ti#al solution =uite well& 'ne can see the value of
loo:ing two ste9s ahead instead of one by the 9rofit difference between the )%ste9 and *%
ste9 heuristics& The 3%ste9 heuristic already 9rovides the o9ti#al solution here&
'. &onclusions
The advantages of /P see# undis9uted regarding costs savings and ad#inistrative
9rocess auto#ation& .s i#9le#entation costs are lowering /P is e(9ected to be widely
ado9ted by co#9anies& .t the #o#ent /P is still in its infancy though& Pilot 9rojects are
under way but co#9anies still hesitate with the full roll%out of /P into their
))
organisation because the technology is new and not #uch e(9erience e(ists yet about
i#9le#entation strategy&
The #athe#atical /P roll%out #odel is a fra#ewor: for 9roviding a good roll%out
strategy based on e(9ected costs and cost savings& It deter#ines the o9ti#al order of
i#9le#entation for co##odity grou9s and de9art#ents together with total cost savings
and how these savings will occur over ti#e& !or larger 9ractical cases greedy heuristics
can be used to calculate (near) o9ti#al strategies& Using s#all scale e(a#9les with
realistic values of the costs and cost savings the #odel shows that there can be a large
difference in the total savings between the o9ti#al roll%out strategy and other (rando#)
strategies& . good roll%out strategy is therefore an i#9ortant factor in the successful
i#9le#entation of /P&
.lthough #any as9ects already have been incor9orated in the #odel it is 9ossible to
#a:e so#e e(tensions& ?ifferent functionality levels of /P syste#s different lengths of
i#9le#entation 9eriods can be incor9orated easily& 'ne could also thin: of learning
curves for i#9le#enting /P syste#s& Stochasticity could also be included as in9ut
values #ay not be easy to esti#ate in 9ractice&
The a99licability of the #odel with 9ossible e(tensions still has to be verified with
e#9irical evidence in the near future& !or 9ractical 9ur9oses it is good to e#9hasise that
only financial as9ects of /P i#9le#entations are o9ti#ised in the #odel& To i#9le#ent
/P successfully also other organisational factors #ay need to be considered& These
)3
factors could influence the 9referred i#9le#entation se=uence& However with the #odel
at least the financial conse=uences of other roll%out strategies can be calculated&
(eferences
.berdeen grou9 )--*& e%Procure#ent1 finally ready for 9ri#e ti#e .berdeen grou9
*4())&
.ngeles <& )---& <evisiting the role of Internet%/?I in the current electronic
co##erce scene >ogistics Infor#ation "anage#ent *3(*) 4,%,+&
.ttaran "& )--*& The co#ing age of online 9rocure#ent Industrial "anage#ent O
?ata Syste#s *-*(4) *++%*5-&
$ergeron !& and <ay#ond >& *66+& "anaging /?I for cor9orate advantage1 .
longitudinal study Infor#ation O "anage#ent 3* 3*6%333&
$oer >& de Harin: J&H&.& and Heijboer G&J& )--*& . conce9tual #odel for assessing
the i#9act of electronic 9rocure#ent acce9ted for 9ublication in the /uro9ean Journal
of Purchasing O Su99ly "anage#ent&
Broo# S&<& )---& The i#9act of web%based 9rocure#ent on the #anage#ent of
o9erating resources su99ly Journal of Su99ly Bhain "anage#ent Ainter 4%*3&
?obler ?&A& and $urt ?&0& *66D& Purchasing and su99ly #anage#ent1 te(t and cases
si(th edition "cGraw%Hill&
Harin: J&H&.& *666& /(celling with e%9rocure#ent PriceAaterhouseBoo9ers
0etherlands&
)4
@aefer !& and $endoly /& )---& The ado9tion of electronic data interchange1 a #odel
and 9ractical tool for #anagers ?ecision Su99ort Syste#s 3- )3%3)&
@i# S&H& Jang ?&H& >ee ?&H& and Bho S&H& )---& . #ethodology of constructing a
decision 9ath for IT invest#ent Journal of Strategic Infor#ation Syste#s 6 *+%35&
@olchin "&G& and Trent <&J& *666& <educing the transactions costs of 9urchasing low%
value goods and services B.PS focus study (htt91EEwww&ca9sresearch&org)&
@raljic P& *653& Purchasing #ust beco#e su99ly #anage#ent& Harvard $usiness
<eview Se9te#ber%'ctober *-6%**+&
>eenders "&<& and !earon H&/& *663& Purchasing and #aterials #anage#ent Tenth
edition Irwin $oston&
>ong $&G& )--*& Ten /%Bo##erce Guestions .ll Purchasing "anagers "ust .nswer
9roceedings of the 5Dth .nnual International Purchasing Bonference&
>ysons @& *66D& Purchasing !ourth edition Pit#an Publishing >ondon&
"cIvor <& Hu#9hreys P& and Huang G& )---& /lectronic co##erce1 re%engineering
the buyer%su99lier interface $usiness Process "anage#ent Journal D()) *))%*35&
"in H& and Galle A&P& *666& /lectronic co##erce usage in business%to%business
9urchasing International Journal of '9erations O Production "anage#ent *6(6) 6-6%
6)*&
Pad#anabhan S& )--*& /lectronic catalog #anage#ent Purchasing Today June 9& D-&
Segev .& Porra J& and <oldan "& *66+& Internet%based /?I strategy ?ecision
Su99ort Syste#s )* *,+%*+-&
),
Sherrill !& .nthony T& ?anish S& and Bhrist @& )--*& eProcure#ent1 the 9ro#ise
and reality Peo9leSoft white 9a9er series (htt91EEwww&9eo9lesoft&co#)&
Telgen J& *66+& <evolution through electronic 9urchasing& Aor:ing Pa9er University
of Twente 0etherlands&
Appendi) A
The notation used 1
Indices1
j 1 denoting the co##odity grou9 (*&&&jmax)
1 denoting the de9art#ent (*&&&max)
t 1 denoting the 9eriod (*&&&')
Here jmax and max is the res9ectively the nu#ber of co##odity grou9s and the
nu#ber of de9art#ents to be added to the /P syste#& ' is the nu#ber of
i#9le#entation 9eriods and therefore e=ual to jmax ) max.
In9ut data1
I 1 Bosts of the uncusto#ised /P syste# itself&
!!
j
1 I#9le#entation costs of /P for co##odity grou9 j ( -)&
"!
fro# -
to *)&
+
t
1 ?irect 9rofit in 9eriod t&
v
t
(i
t
) 1 "a(i#u# 9rofit in 9eriod t for state i
t
&
Appendi) *
Theorem
$elow a certain threshold value
4
of the discount factor the greedy #%ste9 heuristic
gives the o9ti#al solution&
Proof
Su99ose the greedy #%ste9 heuristic does not give the o9ti#al solution then for at least
one 9eriod t
1
it holds that the o9ti#al decision *
t)1-opt
gives a lower 9rofit in the ne(t m
ste9s than the decision based on the heuristic *
t)1-gr
1
)+
( ) ( )
+
<
*
*
*
*
L
L
L
L
L
L
m t
t t
gr t t
t t
m t
t t
opt t t
t t
* i + * i + ($&*)
while still the value of v
t
(i
t
) is higher than when the greedy decision is chosen1
( ) ( ) >
L L L L
t gr t t opt t
i v i v
( ) ( ) ( ) ( ) + > +
+ +
+
+ +
+
m t gr m t
m
m t
t t
gr t t
t t
m t opt m t
m
m t
t t
opt t t
t t
i v * i + i v * i +
L L
L
L
L
L L
L
L
L
*
*
*
*
( ) ( ) ( ) ( ) ( ) ( )
m t gr t m t opt m t
m t
t t
opt t t gr t t
t t
m
i v i v * i + * i +
+ + + +
+
+ +
<
L L L L
L
L
L
*
* *
*
*
($&))
.s can be chosen arbitrarily close to - choosing below a certain threshold value
4
will lead to a contradiction&
Appendi) &
Theorem
!or 0 N *)3&&& and # N * ) &&& 0 it holds that
( ) ( )
+
+
+
m
i
'
t
m
i
i '
m
i t '
- *
*
-
*
*
*
* (B&*)
Proof
<ewriting the indices of the l&h&s& leads to1
( ) ( ) ( )
+ +
'
t
m
i
'
t
m
i
'
t
m
i
i t i t i t '
*
*
-
*
-
*
- *
*
-
* *
(B&))
0ow we 9rove with induction that1
)5
( ) ( )
+
+
m
i
'
t
m
i
i '
m
i t
- *
*
-
*
*
*
(B&3)
!or 0 N * (B&3) holds as * N *&
Su99osing (B&3) holds that it re#ains to be shown that1
( ) ( )
+
+
m
i
'
t
m
i
i '
m
i t
-
*
*
*
-
)
*
*
(B&4)
Starting with the l&h&s& we get1
( ) ( ) ( ) + +
*
- *
*
-
*
*
*
-
*
m
i
'
t
m
i
'
t
m
i
i ' i t i t
( ) ( ) ( ) +
,
_
+
+
+
+ + +
+
*
-
*
- -
* *
*
*
* *
*
*
m
i
m
i
m
i
i '
m
m '
i ' i '
m
( ) ( )
+
+
+
,
_
+
+
m
i
m
i
i '
m
i '
m
'
- *
)
*
*
)
*
)
(B&,)
)6
Table 1: Implementation costs of the commodity groups an departments +in 1,,,
-. dollars/
&ommodity
+0/
CC
j
1epartment
+2/
DC
k
B* *5, ?* ),
B) *-- ?) 6,
B3 6, ?3 )-
B4 *4, ?4 3D
B, 6) ?, *)
BD D- ?D ++
B+ *4,
3-
Table : E)pected cost sa!ings R
jk
per commodity per department +in 1,,, -.
dollars/
1epartment +2/
&ommodity +0/
11 1 1" 1# 1% 1'
&1 3-- ,DD )4- *6- *,- 6
& )-- *5- )3, *3- *6- D
&" +- *), *-- ,)3 **- 3
&# ),- 3D, 3), *)) *4- ,
&% 3*- *+, D- *)- *,, *4
&' **, 4- 3) +- 3D, 5
&3 *, 34 *3 )+ ), ,-
3*
Table ": 4ptimal roll-out strategy together with the re!enues associated with each
implementation period +in 1,,, -. 1ollars/
Period 4ptimal
solution
1irect profit 1irect profit
+discounted/
&umulati!e
profit
* ?) %D6, %D6, %D6,
) B* 35* 3D- %33,
3 ?* )+, )4D %56
4 B4 4+- 365 3-6
, ?3 ,4, 43D +4,
D B) ,*, 36- **3,
+ ?4 4-D )6* *4)D
5 B3 +)3 46- *6*D
6 ?, ,+5 3+* ))5+
*- B, +)5 44) )+)6
** BD ,D) 3)3 3-,)
(*)) (B+) (%3*) (%*+) (3-3,)
(*3) (?D) (*5) (6) (3-44)
3)
Table #: (esults regarding the e)pected profit +in 1,,, -. dollars/ using !arious
roll-out strategies
$ethod Profit Total
in!estment
Implementation order
'9ti#al 3-,) *)6, ?)B*?*B4?3B)?4B3?,B,BD
Greedy *%ste9 )6)- *)3) ?,BDB)?*B,?3B4?)B*?4B3
Greedy )%ste9 3-4* *)5) ?4B3?)B*B4?3?*B)?,B,BD
Greedy 3%ste9 3-,) *)6, ?)B*?*B4?3B)?4B3?,B,BD
Aorst case )-4) *45* B*B4B+B)B3B,BD?D?3?,?4?*?)
33