Decision Support Systems 51 (2011) 620–626

Contents lists available at ScienceDirect

Decision Support Systems

j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / d s s

Cooperative planning in express carrier networks — An empirical study on the

effectiveness of a real-time Decision Support System
Sascha Dahl, Ulrich Derigs ⁎
Department of Information Systems and Operations Research, Universität zu Köln, Albertus-Magnus-Platz, D 50969 Köln, Germany

a r t i c l e i n f o a b s t r a c t

Article history: For small transportation firms cooperation in a carrier network is a proper mean to overcome the
Received 6 August 2010 inefficiencies from deadheading. To be successful, such a network has to secure two different but equally
Received in revised form 15 December 2010 important aspects: the partners have to be aware of specific consolidation potentials through order exchange
Accepted 27 February 2011
which is an optimization and communication problem, and, the partners have to experience incentives to
Available online 16 March 2011
contribute actively to the network which is the problem of finding a fair cost/profit allocation schema for
Keywords:
order exchanges. In this paper we discuss the experience with the development of a Decision Support System
Cooperative planning for a specific express carrier network. We illustrate how the consolidation potentials in such a network with
Carrier networks autonomously planning carriers can be exploited and cost effectiveness can be improved substantially
Dynamic vehicle routing problem through the use of a suitable distributed Decision Support System if the two success factors awareness and
Simulation fairness are addressed properly.
© 2011 Elsevier B.V. All rights reserved.

1. Introduction communication system supporting cooperative planning. Another

During the last years, transportation firms are faced with increasing
cost pressure and revenue erosion at the same time. Large transporta-
tion companies are able to realize a high utilization of their vehicles and
acceptable operational cost by consolidating and combining orders to
efficient roundtrips. Small carriers serving ad-hoc one-way shipping
orders only are faced with the problem of low volume of shipments with
less than truckload trips as well as dead head trips due to an imbalance
among locations. This leads to cost ineffective transportation plans and/
or may result in non-competitive prices.
Such small-sized companies may compensate their competitive
disadvantage by allying with partners to a cooperation network to
establish a more profitable portfolio of orders. In such a network each
partner plans his orders and his vehicle fleet independently with the
option to exchange orders with partners. Partners charge their own
customers based on a specific price function. The carrier operating an
order for a partner receives a monetary compensation, which has
been specified in a generally agreed upon compensation schema.
For such a network to be successful and sustainable there are a
number of critical factors: on the strategical level the choice of the right
set of partners yielding enough consolidation potential as well as mutual
trust is a cardinal point. On the operational level awareness is essential,
i.e. existing potentials have to be detected and communicated, a
task which calls for the establishment of a proper information and
⁎ Corresponding author. Tel.: + 49 2214705327; fax: + 49 2214705329.
E-mail addresses: sascha.dahl@explicatis.com (S. Dahl), ulrich.derigs@uni-koeln.de
(U. Derigs).
success factor is the establishment of a compensation schema which
puts incentives to both partners involved in exchanging orders and
which is considered to be fair by all network partners.
The issue of reducing transportation cost by collaboration and
forming alliances has been investigated in the scientific literature with
respect to different aspects like different transportation markets or
modes i.e. airfreight [7], shipping [1], trucking [8], intermodal freight
transportation [16], supply chain management [18] as well as decision
level, i.e. from evaluating strategic aspects in alliance formation [2] to
pricing/revenue management and allocation of cost benefits [4,10–12].
Most developments concern design questions and propose models
which are based on theoretical foundations stemming from game
theory, combinatorial auctions and network flow. Refs. [10,13] present
approaches for a problem where intra-enterprise in and outsourcing
decisions on bundles of logistic services i.e. bundles of transportation
requests from customers in a profit center structured forwarding
company have to be made. In both approaches the cost difference
through exchanging requests is evaluated for numerous bundles of
requests. Then the optimal distribution of the bundles among the profit
centers is determined via a combinatorial auction. Both approaches
propose mechanisms for sharing the resulting profit increase among the
centers.
Our development was driven by the necessity to control cost
effective transportation in an established collaborative network of
independent express couriers on the operational level where decisions
on the exchange of orders have to be made instantaneously. In our
problem environment the situation is highly dynamic such that at no
point in time planning of a fixed set of orders/requests is possible as it is

0167-9236/$ – see front matter © 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.dss.2011.02.018
 S. Dahl, U. Derigs / Decision Support Systems 51 (2011) 620–626
assumed in Refs. [10,12]. Hence the focus was on implementing rational
decision models which allow on-line algorithms within an effective
Decision Support System.
In this paper we describe our experience with the development and
maintenance of pool.tour, a distributed real-time internet-based
collaborative Decision Support System (DSS) for a large express courier
network, and we analyze the impact of this system on the success factors
mentioned above. In a first development we implemented a system
proposing order exchanges automatically which are profitable for both
partners involved based on the established compensation schema. This
technologically highly demanding system has been in use for about
2 years. Yet, the improvement over the formerly used approach where
dispatchers had to assume potentials from their experience and to
communicate proposals to partners over the phone was much lower
than expected. An analysis of the business i.e. the order pool and the
proposals generated showed that the system could only rarely find
consolidations i.e. insertions of orders in existing trips which were
profitable for the acquiring as well as the operating partner but was only
able to propose exchanges which result in separate trips. This
unsatisfactory behavior motivated us to analyze the contractual
compensation schema, to propose an alternative schema and to
compare the result with the existing schema. For that purpose we
performed a simulation study on the logged order pool over a significant
duration of several weeks. Our analysis showed that the rather poor
performance could be clearly attributed to the compensation schema.
Using a rather straightforward cost based compensation schema within
the model base of pool.tour network wide transportation cost could be
decreased significantly, even down to the level reachable by centralized
planning. Also the distribution of the profit showed to be much fairer.
This paper is structured as follows. In Section 2 we give a short
description of the transportation market and the specific carrier
network underlying our development and study, in the following
referred to as the Cooperative Logistic Network (CLN) or simply
network for short. In Section 3 we introduce the planning problem and
the established compensation schema. Then, in Section 4, we shortly
describe our Decision Support System pool.tour. In Section 5 we
describe the motivation and the design of our simulation study for
evaluating the effectiveness of pool.tour and in Section 6 we report
the central results of this experience.
2. The cooperative logistic network
In 2001 the market value of the courier/express/parcel segment in
Europe was about 36 billion Euro, and 30% of this volume was realized in
Germany. In Germany close to 10,000 courier firms are specialized on
this kind of general freight transport, yet, this number contains many
one-person businesses which operate as subcontractors for larger
companies only. CLN was founded in 2001 by logistics professionals for
logistics professionals to unite small and medium-sized courier
companies under a strong brand. It was created by a consortium of
independent courier companies with a more strategically oriented
organization in mind than traditional partner-based systems. All twelve
founding shareholders were professionals in international procurement
logistics, i.e. all partner companies had long term expertise in national
and international deliveries and offered procurement logistics, less than
truckloads (LTL) and complete truckloads, transportation of dangerous
goods and customs service. With the Europe-wide business partners
CLN is able to offer pickup and delivery of shipments anywhere in
Europe, with availability 24 h a day, 365 days a year and guaranteed
pickup anywhere in Germany within 60 min.
The purpose of CLN – as the purpose of every cooperation of freight
forwarding companies – is to realize a profitable equilibrium between
customer demand and available transport resources by interchanging
customer requests among partners. Being a member of CLN allows the
forwarding company to choose between two modes of fulfillment for
621
each task: To use own vehicles (self-fulfillment) or to use a partner
carrier who will then receive a compensation for the request fulfillment.
The critical success factor for a freighter is the percentage of
deadheads. According to experience this percentage had been up to
40% to 45% for the individual carriers before the cooperation, a
number resulting from the high spatial diversity of the single requests
and the inability to consolidate within the available time-frame, yet, a
number much too high to allow competitive prices and sufficient
profit. A rough analysis of the business within the first year of CLN's
operation where cost reducing interchanges had been realized
between the dispatchers occasionally via telephone conferencing
had shown only a slight reduction of deadheads. Yet, expectation was
that even with the full thrust of the cooperation, its mutual growth
and coordination potential, the bottom line of deadheads would
always remain around 30% due to the extreme short reaction time. All
partners were aware of the fact that the two major make and break
questions on the cooperation were constituted by the potential to
bring the key performance indicator on deadheads closer to this
bottom line and the discussion on how to split earnings among
partners in a fair manner. This judgement on the key issues has led to
the development of pool.tour, our DSS which supports cooperative
planning as well as communication in this highly time-sensitive
environment.
3. Modeling the planning situation
Let P be the set of partners in the network. Each partner p ∈ P is
located at a specific depot depp and operates a set of vehicles Vp. Each
vehicle v is assigned to a vehicle class. Let VC be the set of vehicle
classes and vcv the vehicle class of vehicle v. All vehicles in a vehicle
class vc ∈ VC share common physical and technical transportation
capabilities as for instance capacity Qvc, average speed, cost rates for
distance and time etc. The set of vehicle classes is assumed to be
partially ordered with the semantic that a vehicle of a ‘larger’ class can
always transport orders which require a ‘smaller’ vehicle class. Each
partner p ∈ P has acquired a set Op of orders. Here, each order o is
defined by its pickup location po, its delivery location do, its capacity
requirement capo and a time window [eo, lo], which specifies the
earliest time for pickup and latest time for delivery, respectively. From
the capacity requirement one can determine vco the minimum
required vehicle class for the order. For each vehicle class vc ∈ VC
two rates apply: pricevc the transportation price per km for the
customer and compCostvc an internal cost rate per km used for
compensation. For two locations x and y let l(x, y) denote their
distance. Then the revenue rev(p, o), which a partner p obtains from
his customer after serving its order o, is calculated as follows:
h

  i
revðp; oÞ = pricevco max 0; l depp ; po −l
fix
+ lðpo ; do Þ : ð1Þ
Thus, for picking up the order (pre haul leg) only the distance
exceeding lfix is charged. Note, that each vehicle can transport more
than one order at a time, as long as the total capacity is not exceeded,
and that it is the task of the dispatching systems of the partners to
combine orders to tours.
Let us first assume the non-cooperative case, i.e. the special case of
one partner p only. This problem of combining orders to tours is
coined as Pickup and Delivery Vehicle Routing Problem with Time
Windows (PDVRPTW) and well studied in the Operations Research
literature. As a so-called rich variant of the classical Vehicle Routing
Problem it is NP-complete and thus only heuristic methods are
applicable for solving problem instances of practical size. Efficient
heuristic algorithms have been proposed by Ref. [14] and Ref. [5].
Now, a solution to a vehicle routing problem, also called a schedule,
consists of a partition of the orders into clusters/tours which are
assigned to the vehicles and an ordering/routing of the orders within a
622 S. Dahl, U. Derigs / Decision Support Systems 51 (2011) 620–626
cluster. Here, a feasible schedule for the PDVRPTW has to obey the
following properties:
• Each order o ∈ Op is assigned to exactly one vehicle/tour v ∈ Vp.
• For each order o ∈ Op location po is visited before do.
• Each order o ∈ Op is picked-up not earlier than eo.
• Each order o ∈ Op is delivered not later than lo.
• For each vehicle the load of the vehicle must not exceed its capacity
Qvcv at any time on the route.
Note that due to the relatively short lead times the number of
orders transported simultaneously on a vehicle is rather small and thus
vehicle capacity is not a critical constraint for building feasible tours.
Under this static view, one assumes that a vehicle will always return
to its depot after serving all orders of its tour. Yet, in the express carrier
environment, due to the highly dynamic business all plans have to be
adapted constantly and a vehicle may receive new orders on tour. Thus
the problem to be solved is a Dynamic PDVRPTW (DPDVRPTW).
Ref. [3] presents an overview on this class of dynamic vehicle routing
problems distinguishing different sub-classes. A basic and commonly
used solution strategy is to adapt an algorithm that solves the static
version of the problem. In one approach the static problem is solved
each time a new information, i.e. an order, is revealed. In a second
approach the current solution is only updated with heuristic methods. In
our application we solve the DPDVRPTW modifying the heuristic
PDVRPTW-solver called ROUTER (see Ref. [5]). ROUTER is a two-phase
heuristic, which first constructs good feasible solutions using a cost-
based cheapest insertion strategy and then applies a local search based
metaheuristic for improvement in the second phase. In contrast to the
assumption in Ref. [3] that generally, when time windows are present in
a pickup and delivery vehicle routing problem, they are not tight, the
extremely short lead time is a characteristic for our business. Therefore
the application of the rather time-consuming improvement phase of
ROUTER is discarded when solving the DPDVRPTW instances since its
application would be unrealistic for practical use in a real-time
environment with such narrow lead times. We have implemented the
rolling horizon paradigm, i.e. at each time an order data change occurs a
(static) instance is solved with all orders fixed which are already in
execution. Note that our problem cannot be treated as a stochastic VRP
since no information on future request is known.
Now assume the collaborative case at CLN. Here, a carrier p has two
options: to serve an order o with his own fleet or to have it served by
one of the partners. In the first case, contribution to profit ctp(p, o) is
calculated as usual, reducing revenue by the marginal imputed cost:
ctpðp; oÞ = revð p; oÞ−ΔimpCostðp; oÞ ð2Þ
where imputed cost is calculated based on the distance a vehicle has to
drive and the time a vehicle spends for driving and/or waiting. Both terms
are multiplied by the vehicle class specific cost rates for distance and time.
In the second case, i.e. if one of the partners, say q, serves the order, the
compensation compCS − P(q, o) is determined according to a contractually
agreed upon schema CS-P which distinguishes two cases:If the order is
served by a dedicated tour, the calculation is similar to the revenue
calculation above, with the only difference that twice the compensation
cost rate is applied:
h

  i
compCS−P ðq; oÞ = 2⋅compCostvco max 0; l depq ; po −l
fix
+ lðpo ; do Þ :
ð3Þ
If the order is inserted into an already existing tour, and thus is
combined efficiently with other orders, the load distance l(po, do) plus
the marginal distance caused by the service of o applies. This more
complicated calculation is illustrated in Fig. 1 for the case that order o
is combined with two other orders r and s, respectively. Here bold
arrows indicate the trips with orders or load which are performed by
po o do
Without p
cooperation:
pr r dr ps s ds
q
Cooperation: p
q po o do
pr dr ps ds
r s
Fig. 1. Compensation with CS-P in case of combined order execution.
the vehicle, while dashed arrows symbolize trips without load, i.e.
deadheads for instance.

compCS−P ðq; oÞ = compCostvco lðpo ; do Þ + lðpr ; po Þ + lðpo ; dr Þ
 ð4Þ
−lðpr ; dr Þ + lðps ; do Þ + lðdo ; ds Þ−lðps ; ds Þ
Note, that in both cases the cost rate of the required vehicle class is
applied, not the rate of the vehicle actually serving the order.
At first sight the problem at CLN is similar to the problem domain
treated in Refs. [10,12]. Yet, the approaches proposed are not
applicable here because of two reasons. First, the dynamic occurrence
of requests with extremely narrow time windows for pick up does not
lead to a point in time with a real planning situation where for a fixed
set of orders an optimal exchange of bundles of orders could
determined. Secondly, the established cooperation mechanism is to
exchange single orders after bilateral negotiation. Thus the autonomy
of the partners does not allow a central authority which is able and
legitimate to reallocate orders.
4. Pool.tour — a Decision Support System
In general a Decision Support System (DSS) is an interactive computer-
based system which supports decision makers in solving semi-structured
problems [17]. According to the classical concept a DSS consists of three
modules: database, modelbase and dialog component [17]. Yet, to be able to
support planning by the dispatchers as described above this architecture
has to be enhanced since several aspects are not addressed by the basic
DSS-concept which is focusing on supporting a single decision maker only.
First, a number of decision makers coming from different partner firms
and thus following different individual objectives have to work together in
a cooperative manner. A DSS for such a scenario is called a Group DSS [15].
In our planning environment support of communication between
dispatchers is an essential requirement. Also, the problem requires to
control the operation of many mobile units (vehicles) which themselves
send messages on their geographical position via fleet telematics. Ref. [15]
classifies a DSS with such a functionality as Communications-Driven DSS.
The dynamic of the business and the extremely short lead time between
order entry and pickup time require the installation of a Real-Time DSS
which is capable to react to the changing environment instantly. Since the
users of the DSS are geographically spread over Europe a Distributed DSS
[9] has to be developed.
Based on these requirements we have designed an interactive
concept where the DSS integrates the decentralized databases of the
partners' dispatching systems and uses a simple but fast and powerful
on-line heuristic which is based on the current information on orders,
planned tours and vehicle positions. The heuristic permanently
generates proposals for inserting new orders into tours which are
then communicated to the dispatchers in a pro-active manner. This
concept is realized with an internet-based client–server-architecture
where data, application logic and generation of the user interface are
centralized while presentation and interaction with the dispatchers is
performed through a web interface on the dispatchers' clients. Ref.
[15] has characterized such a DSS as Web-Based DSS.
 S. Dahl, U. Derigs / Decision Support Systems 51 (2011) 620–626
Fig. 2 shows the architecture of our DSS. The central DSS-database
stores and manages all relevant data from orders, vehicles and vehicle
routings from all partners as well as the current vehicle positions. This
data is imported automatically and in real-time from the local
dispatching systems of the partners as well as the database of the fleet
telematics provider using appropriate interfaces. The web interface
allows the partners to search and filter the data with respect to different
criteria. From this exposition the planning paradigm becomes apparent:
decentralized planning of orders at each partner site with a central
supporting system, the so-called proposal generator which searches and
proposes options of profitable exchanges in real-time based on data
from the individual dispatching systems and fleet telematics.
The proposal generator is the core component ensuring awareness of
consolidation potentials. It is activated automatically and in real-time in
case a new order is inserted into the database or an order data is changed.
Moreover the proposal generator can be called manually by the partners
for every own order. After insertion of an order, ō say, in a first step, the
generator uses the cheapest insertion logic from ROUTER [5] to check all
feasible positions for pickup and delivery in routings of vehicles of that
partner, p say, who according to the current situation has to fulfill the
order. In case of a new order this is the partner who has acquired the order
and in the case of an order change this is the acquiring partner or the
partner who accepted to overtake the order following an earlier proposal.
Analogously, all feasible positions for pickup and delivery in vehicle
routings of partners are identified and the position which results in the
lowest compensation cost is selected. In these calculations the currently
available information on vehicle positions is used.
Thus a proposal for an order exchange is generated if
revðp; oÞ−compCS−P ðvi ; oÞ N ctpðv; oÞ ð5Þ
which is equivalent to
compCS−P ðvi ; oÞ b ΔimpCostðv; oÞ ð6Þ
where v is the vehicle of partner p and vi is the vehicle of the partner pi
for which the cheapest insert position was found. Now, the proposal is
acceptable for the network partner pi if
compCS−P ðpi ; oÞ N ΔimpCostðpi ; oÞ ð7Þ
i.e. if the partner takes an advantage from serving o. The proposal
together with the compensation amount to be paid is then communi-
Fig. 2. DSS architecture.
623
cated to partners p and pi via e-mail. Note that through this service the
DSS also reduces communication effort and cost significantly.
5. Design of a simulation study for evaluating the effectiveness
of pool.tour
The system described in Section 4 has been implemented in CLN
and has been in use for about 2 years, i.e. during their daily planning
the dispatchers received proposals for order exchanges. Yet, the
experience with the use of the system was not satisfactory, i.e. the
reduction in deadheads and the increase in contribution to profit that
materialized were far below expectation. When monitoring the
system we could observe that the system was constantly detecting a
significant number of proposals. Yet, a closer analysis revealed three
problems: only a small fraction of the proposals concerned consolida-
tions, i.e. the insertion of orders in existing tours, the majority of the
proposals were not implemented by the dispatchers, and, a large
number of obviously promising exchanges were not proposed.
According to the comment of the partners the reason for the
second failure is the fact that for exchanges which would result in a
separate deadhead tour for the receiving partner the compensation to
be paid by the acquiring partner is too high, i.e. all partners are aware
of carriers outside CLN who demand less compensation fee for such an
order. Now, our hypothesis was that the specific compensation
schema CS-P which is implemented in CLN and hence in the proposal
generator of pool.tour is causing the other deficits, too. We argued
that the compensation schema CS-P disguises apparently cost-
efficient exchange options by either setting the compensation fee
too high for the acquiring partner or too low for the receiving partner.
Thus the schema is unfair and hence irrational.
Therefore we decided to propose and evaluate an alternative
compensation schema which observes the specific situation of the
business at all partners explicitly and whose fairness is comprehensible.
Of course, such a substantial contractual change cannot be implemented
without prior proof of concept, i.e. evidence of improvement. Therefore
we decided to perform an extensive simulation study. The purpose of
the study was not only to evaluate an alternative schema but also to
analyze to which level the coordination potential is exploited. For that
purpose we have simulated different organizational settings and
scenarios. In all settings and scenarios we assume that the proposals
generated by pool.tour are accepted and implemented by the partners.
624
5.1. Simulation data
For the simulation study we have used real data from our
European-wide operating cooperative logistic network CLN consisting
of about 50 carriers at the time of the study. We stored all orders over
twelve randomly chosen weeks in 2008. By purpose we have chosen a
period before the economic crisis with regular demand. At that time
the network had a capacity of 8905 vehicles assigned to five vehicle
classes. This high number of vehicles reflects the fact that in this
courier network each carrier has access to a virtually unlimited
number of subcontractors. This carrier and vehicle data has been
considered as static, while the order data is highly dynamic. To gather
real order data we have timestamped and logged every change of
order data for all carriers. Then we have eliminated all data changes
reflecting order exchanges between carriers because such decisions
are subjected to the planning approach in the simulation.
5.2. Alternative planning scenarios and methods
We have analyzed the impact of alternative organizational
settings: the collaborative planning approach supported by our DSS,
a non-collaborative planning scenario where each carrier plans and
operates his own orders only and a centralized approach imitating
the situation that all resources and orders belong to one carrier.
The non-collaborative scenario requires to solve a set of Dynamic
Pickup and Delivery Vehicle Routing Problems with Time Windows
(DPDVRPTW), one for each partner, while the centralized approach
requires to solve a single Dynamic Pickup and Delivery Vehicle
Routing Problem with Time Windows and Multiple Depots
(DPDVRPTWMD) (see Ref. [14]). As already mentioned we have
solved the non-collaborative scenario problems modifying the
heuristic PDVRPTW-solver ROUTER (see Ref. [5]). In our work, we
found that the real-world problem instances resulting from the
centralized planning approach can be solved by this modification of
the ROUTER-insertion heuristic considering all orders already planned
to be fixed and continuously evaluating the cheapest insertion
position for each incoming or updated order. This approach mimics
a realistic planning procedure in the dispatch and thus the solutions
can be seen as the results achievable by a realistic practical solution
method respecting the high dynamic for the centralized organiza-
tional setting. In that sense they build the reference for our simulation
results. Based on these solutions as bottom line, plans resulting from
DSS-simulation can be evaluated to examine the effects on total cost,
distribution of orders, etc.
100%
98%
96%
Network-wide total costs
94%
92%
90%
88%
86%
84%
82%
80%
Mean A B
Isolated planning
Fig. 3. Relative cost, scenario without outsourcing, 100% = cost with isolated planning.
S. Dahl, U. Derigs / Decision Support Systems 51 (2011) 620–626
5.3. Alternative compensation schema
The compensation schema determines whether a specific collab-
oration through an order exchange is of advantage for both partners
involved. Hence, the design of the compensation schema with respect
to incentiveness and fairness is a key factor for the overall
performance of the network. An analysis of the effect of alternative
compensation schemata and a comparison with alternative planning
approaches as reference will certainly help to improve the potentials
of the collaboration and to establish mutual trust. While in Ref. [12]
profit is shared based on the exchange of single bundles, Ref. [13]
uses the Shapley value as a mechanism to share the dividend of
cooperation among the participating couriers. Ref. [10] proposes a
flexible mechanism to distribute total cost savings obtained from the
exchange of bundles after solving the combinatorial auction problem.
Here the trade-off between cost saving through outsourcing and cost
increase through insourcing can be influenced by adjusting the ratio
between in and outsourcing.
Our approach is based on the evaluation of exchanging single
requests and here we have analyzed the impact of two different com-
pensation concepts: the one presently used by CLN and implemented
in pool.tour (CS-P) and another simple marginal cost-based schema
(CS-CB):
1
compCS−CB ðq; oÞ = ½ΔimpCostðp; oÞ + ΔimpCostðq; oÞ ð8Þ
2
i.e. we calculate the marginal imputed cost obtained from cheapest
insertion of o into the transportation plans of p and q respectively.
Schema compCS − CB is incentive compatible in the sense that
whenever there is the potential for decreasing the total network-
wide cost through an exchange then there is also a positive gain from
it for each of the involved carriers. Note that schema CS-CB can easily
be parameterized to implement alternative distributions of the
revenue between the acquiring and operating partner.
6. Simulation results
In a first scenario (scenario without outsourcing) the simulation and
the optimization procedures are based on the assumption that all
transport requests must be fulfilled within CLN, i.e. it is not possible to
have transportation requests executed by eventually less costly
subcontractors from the spot market. Fig. 3 shows the network-
wide total imputed cost per week which occurs when simulating the
C D E F G H I J K L
Simulation period
Collaborative planning CS-P Collaborative planning CS-CB Centralized planning

four different settings introduced above: collaborative planning with
the compensation schemata CS-P and CS-CB, respectively, as well as
isolated planning and centralized planning. Moreover the mean value
over all simulation periods is given. Note that we have normalized the
results, i.e. all cost values are stated as percentage of the cost arising
from the isolated planning setting.
Collaborative planning with the compensation schema CS-P leads
to a mean cost level of 95.77%, i.e. a cost decrease of 4.23% compared
to isolated planning. Using the cost-based compensation schema CS-
CB cost can be reduced significantly to a cost level of 86.15%, i.e. the
cost decrease is 13.85%. Even in the case of centralized planning we
can see that with the use of realistic planning methods the additional
cost reduction to a level of 85.93% is only marginal, i.e. the maximal
cost decrease achievable by cooperating is 14.07%. Note that over the
sample data all these values have only small variances and their
relative order and difference are stable.
These results are rather significant. They support the estimation
and experience of the partners that cooperation with their compen-
sation schema CS-P and traditional manual planning and communi-
cation by phone could only reduce cost marginally. They manifest that
there is a substantial bottom line of cost resulting from inefficiencies
caused by the imbalance of orders with respect to location and time
which cannot be avoided even under centralized cooperative
planning. Yet, most important, the results clearly demonstrate that
the choice of the compensation schema is crucial for the economical
success of the cooperation network. Collaborative planning with the
simple, well understandable and easily applicable compensation
schema CS-CB leads to a significant decrease of network-wide total
cost compared to CS-P. Yet, what is even more important is the result
that by simply applying the cost-based compensation schema CS-CB
the CLN network can exhaust nearly all cost saving potentials and
reach the cost level of centralized planning.
Fig. 4 displays the corresponding values for a second scenario
(scenario with outsourcing) where we assume that transportation
requests may be executed by subcontractors from the spot market. To
allow a comparison of both scenarios all values are again normalized
relative to the cost resulting from isolated planning without
outsourcing.
As expected, in the outsourcing scenario cost are systematically
lower than the respective cost without outsourcing for all settings.
Every solution of the scenario without outsourcing remains feasible,
yet, the option of outsourcing extends the solution space. For isolated
planning as well as collaborative planning with compensation schema
CS-P allowing outsourcing reduces cost significantly by 5.8% and 4.6%,
100%
98%
96%
Network-wide total costs
94%
92%
90%
88%
86%
84%
82%
80%
Mean
Isolated planning
Fig. 4. Relative cost, scenario with outsourcing, 100% = cost with isolated planning in the scenario without outsourcing.
S. Dahl, U. Derigs / Decision Support Systems 51 (2011) 620–626 625
respectively. In the settings of collaborative planning with compen-
sation schema CS-CB and centralized planning cost are only slightly
reducible through outsourcing, i.e. the reduction is only 0.64% and
0.55%, respectively.
This different behavior of the four settings can be explained as
follows: in the first setting of isolated planning outsourcing is the only
possibility to reduce the number of unprofitable transportation. Now,
applying compensation schema CS-P similar reductions which exist
within the network remain invisible. In contrast, applying the cost-
based compensation schema CS-CB such cost savings existing within
the network are made explicit and allow the more efficient allocation
of transport orders within the network to an extent, that the cost
levels with and without outsourcing differ only slightly. Thus, when
applying CS-CB outsourcing does not pay any more and the network
offers to the partners the entire market potential for collaboration.
And this is what triggered the motivation and strategy of forming the
network.
So far, our simulations have shown that for the network as a whole
collaborating using a proper compensation schema like CS-CB pays.
Now, a second aspect is the fairness of the compensation schema. This
aspect is the focus of the following analysis where we have measured
the profit gains for the individual network partners resulting from
their participation in the network. For that purpose we have
calculated for every partner p and for every week an indicator Δgp
which is a proxy for the gain with respect to contribution to profit. Δgp
measures the difference between cIPp the cost occurring in the
isolated planning setting and cCPp the cost occurring in the
cooperative planning setting. Now this value is corrected by adding
the compensation payments p received from network partners and
subtracting the compensation payments of p to other network
partners. These compensation payments are obtained from the results
of the simulation, i.e. the cooperation proposals/transactions com-
puted. Let R denote the set of all compensation payments and rp, q the
compensation which partner p has to pay partner q. Then the indicator
is calculated as follows:
Δgp = cIPp −cCPp + ∑ rq;p − ∑ rp;q : ð9Þ
rq;p ∈R: rp;q ∈R:
p = p;q≠p p = p;q≠p
Fig. 5 shows the distribution of the cooperation profit among the
individual network partners for the scenario without outsourcing.
Fig. 5(a) gives the distribution using collaborative planning with the
compensation schema CS-P, whereas Fig. 5(b) gives the distribution
A B C D E F G H I J K L
Simulation period
Collaborative planning CS-P Collaborative planning CS-CB Centralized planning
626 S. Dahl, U. Derigs / Decision Support Systems 51 (2011) 620–626
a b
Fig. 5. (a) + (b): Profit distribution, collaborative planning without outsourcing, mean
value over all simulation periods.
using CS-CB. In the case of CS-P only a rather small number of network
partners receive the bulk of the (relatively small) cooperation profit.
On the other hand, using CS-CB the (relatively larger) profit is
distributed much more equally and thus more fairly among all
network partners: The number of partners which gain a positive profit
is significantly larger and the profit is distributed more evenly. Thus,
with respect to the equality principle from the theory of distributive
fairness (see Refs. [6,19]) the profit allocation resulting from CS-CB is
more fair than the one resulting from CS-P. Furthermore, the absolute
profit each partner receives is larger than in CS-P, since the total
amount to divide is larger.
7. Conclusion
In this paper we have presented an empirical analysis of the
effectiveness of a Collaborative Decision Support System in an express
carrier network. Due to the highly time sensitive planning environ-
ment supporting awareness of consolidation potentials has been
identified as a critical success factor and has led to the development
and implementation of pool.tour, a technically complex Decision
Support System whose core is a real-time proposal generator. The
experience of 2 years of operation and the results of our simulation
analysis have demonstrated the dominant importance of the choice of
an adequate compensation schema. It is another critical success factor
for the cooperation network, for the potential of materializing the
overall cost reduction as well as for the fairness of profit distribution
within the network. Our simulation reveals that with the support of a
real-time DSS based on an adequate compensation schema the
network is able to work close to the level obtainable by centralized
planning. In a sense this is the strongest possible empirical result.
Thus, the compensation schema has to be regarded as the primal
design parameter of decision support with respect to economic
success and stability of the network. It can make or break the network.
References
[1] R. Agarwal, O. Ergun, L. Houghtalen, O.O. Ozener, Collaboration in cargo
transportation, in: A. Chaovalitwongse, F. Roberts (Eds.), Optimization and
Logistics Challenges in the Enterprise, Springer, 2006.
[2] J. Antes, U. Derigs, M. Zils, Strategic airline alliances portfolio analysis, Avmark
Aviation Economist 16 (1999).
[3] G. Berbeglia, J.F. Cordeaub, G. Laporte, Dynamic pickup and delivery problems,
European Journal of Operational Research 202 (2010) 8–15.
[4] A. Boyd, Airline alliances, OR/MS Today 25 (1998) 28–31.
[5] U. Derigs, T. Döhmer, Indirect search for the vehicle routing problem with pickup
and delivery and time windows, OR Spectrum 30 (2008) 149–165.
[6] M. Deutsch, Justice and conflict, in: M. Deutsch, P.T. Coleman, E.C. Marcus (Eds.), The
Handbook of Conflict Resolution: Theory and Practice, Jossey-Bass, San Francisco,
2006, pp. 43–68.
[7] O. Ergun, L. Houghtalen, J. Sokol, Designing mechanisms for the management of
carrier alliances, to appear in Transportation Science.
[8] O. Ergun, G. Kuyzu, M. Savelsbergh, Reducing truckload transportation costs
through collaboration, Transportation Science 41 (2007) 206–221.
[9] A. Gachet, P. Haettenschwiler, A decentralized approach to distributed decision
support systems, Journal of Decision Support Systems 12 (2003) 141–158.
[10] O. Gujo, M. Schwind, J. Vykoukal, The design of incentives in a combinatorial
exchange for intra-enterprise logistic services, IEEE Joint Conference on E-Commerce
Technology (CEC'07) and Enterprise Computing, E-Commerce and E-Services
(EEE'07), Tokyo, Japan, 2007.
[11] L. Houghtalen, Designing allocations mechanisms for carrier airlines, Doctoral
Thesis, Georgia Institute of Technology, 2007.
[12] M.A. Krajewska, H. Kopfer, Collaborating freight forwarding enterprises. Request
allocation and profit sharing, OR Spectrum 28 (2006) 301–317.
[13] M.A. Krajewska, H. Kopfer, G. Laporte, S. Ropke, G. Zaccour, Horizontal
cooperation among freight carriers: request allocation and profit sharing, Journal
of the Operational Research Society 59 (2008) 1483–1491.
[14] D. Pisinger, S. Ropke, A general heuristic for vehicle routing problems, Computers
& Operations Research 34 (2007) 2403–2435.
[15] D.J. Power, Decision Support Systems: Concepts and Resources for Managers,
Quorum Books/Greenwood Publishing, Westport, 2002.
[16] C. Püttmann, Collaborative planning in intermodal freight transportation, in: R.
Koschke, O. Herzog, K.H. Rödiger, M. Ronthaler (Eds.), INFORMATIK 2007. Informatik
trifft Logistik. Band 1, GI Edition — Lecture Notes in Informatics (LNI), Köllen, Bonn,
2007, pp. 62–65.
[17] R.H. Sprague, E.D. Carlson, Building Effective Decision Support Systems, Prentice-
Hall, Englewood Cliffs, 1982.
[18] H. Stadtler, A framework for collaborative planning and state-of-the-art, OR
Spectrum 31 (2009) 5–30.
[19] N.A. Welsh, Perceptions of fairness, in: A.K. Schneider, C. Honeyman (Eds.), The
Negotiator's Fieldbook: The Desk Reference for the Experienced Negotiator,
American Bar Association, Chicago, 2007, pp. 165–174.
Ulrich Derigs is professor and director of the Information Systems and Operations
Research Department at the University of Cologne, Germany. His research interests are in
the areas of network optimization, combinatorial optimization and mathematical
programming with emphasis on efficient algorithms and data-structures as well as on
decision and data modeling and includes the development of planning methods and
decision support systems especially in the area of transportation and logistics as well as
banking and finance. His research was funded by the Deutsche Forschungsgemeinschaft
(DFG), Deutscher Akademischer Austauschdienst (DAAD), several private companies and
the government. Ulrich Derigs ranks 20th in the 2009 Handelsblatt-Ranking of 2100
professors for Business Administration and Management in Germany, Austria and the
German-speaking part of Switzerland regarding life-time research performance.
Sascha Dahl studied Information Systems at the University of Cologne, Germany, from
1999 to 2004 and specialized in Operations Research, Decision Support Systems, Supply
Chain Management and Production. From 2004 to 2009 he worked as a research
assistant at the Information Systems and Operations Research Department at the
University of Cologne. Sascha Dahl received his doctoral degree in 2009. His research
interests are operational planning and control of inter-organizational cooperations,
decentralized optimization, mechanism design, auction and game theory as well as
logistic problems, especially dynamic vehicle routing problems.