ARTICLE IN PRESS

Int. J. Production Economics 85 (2003) 233249

Production planning and order acceptance in business to business electronic commerce

T. Calossoa, M. Cantamessaa,*, Dang Vub, A. Villaa
Dipartimento di Sistemi di Produzione ed Economia dellAzienda, Politecnico di Torino, Corso Duca degli Abruzzi, 24, I-10129 Torino, Italy b Asian Institute of Technology-Center in Vietnam, 21 Le Thanh Tong, GPO Box 136, Hanoi, Viet Nam
a

Abstract Following the exponential growth of the Internet and the increasing number of software solutions enabling businessto-business electronic commerce, we study negotiation processes in a make-to-order (MTO) environment. In many industries, the value chain is fragmented both horizontally and vertically and manufacturing rms operate upon a MTO basis. In this application domain, the main decision process consists of interrm negotiation upon requests-forquotation distributed by customers and upon bids submitted by suppliers. At the same time, the negotiation process is tightly integrated with production planning and requires appropriate decision-support techniques. This paper discusses in detail the structure for a standardised negotiation process occurring in a multi-enterprise setting and presents three mixed-integer linear programming models that may be used by the different parties involved. r 2003 Elsevier B.V. All rights reserved.
Keywords: Business-to-business electronic commerce; Make-to-order operations; Order negotiation

1. Introduction It is an obvious proposition to state that the management of manufacturing operations is currently undergoing a true revolution, with the widespread diffusion of novel industrial structures and the related enabling technology. Among the former, one can list supply chains, extended and virtual enterprises (Jagdev and Thoben, 2001) and ecosystems (Moore, 1997). Enabling technology is generally centred around the Internet and includes
*Corresponding author. Tel.: +39-0115647238; fax: +39-0115647299. E-mail addresses: cantames@athena.polito.it (M. Cantamessa), dvtung@aitcv.ac.vn (T.D. Vu).

supply chain management software, electronic marketplaces and the related communication infrastructure. Most of the change process has up to now viewed the largest companies as rst adopters and, through them, it has gradually diffused to their smaller suppliers and customers. In fact, the diffusion of novel organisational forms and technological solutions among small and medium enterprise (SMEs) has been fairly slow. The European Information Technology Observatory 2000 (EITO, 2000) reports: y At the turn of the millennium the SME segment remains the biggest untapped market [for IT] in the European economies. In other words, SMEs are undergoing the risk of being kept out of the IT revolution,

0925-5273/03/$ - see front matter r 2003 Elsevier B.V. All rights reserved. doi:10.1016/S0925-5273(03)00112-9

ARTICLE IN PRESS
234 T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249

despite the fact that the economies of advanced countries are solidly based on the value produced by such rms. For example, in the European Union, SMEs with less than 50 employees provide jobs for around 56 million people with a turnover of around 11,500 Billion h. A signicant proportion of these companies belong to traditional manufacturing industries. When discussing e-business solutions for manufacturing SMEs, it must also be remembered that only 30% of European SME output goes to the consumer market. This implies that the focus should be on business-tobusiness (B2B), rather than business-to-consumer relations. If one looks at the steps being undertaken within B2B electronic commerce by the larger companies, who often act as customers to SMEs, such companies currently pay great attention to eprocurement solutions. Apart from a few signs of change that are now starting to arise, e-procurement practices have up to now focused on customer benets in terms of reduced transaction costs, and especially on price reductions. For example, reverse auctions (Engelbrecht-Wiggans, 1980)in which customers force suppliers to outbid each other by lowering pricehave become, to the eyes of many observers, the quintessence of Internet-based B2B electronic commerce. This business model certainly is not benecial to supplier SMEs, and explains much of their reluctance to join the Internet revolution (Counsell, 2001). One may therefore draw the conclusion that there is substantial space for research in e-business solutions that may become a trulyand not only rhetoricallywinwin proposition for suppliers as well as for customers, large and small alike. Such research should cover aspects related to:
*

Information and Communication Technology, so as to develop enabling systems, industrial organisation, that may help to set fair rules of the game and to foresee and manage the resulting structural transformation of industrial value chains, manufacturing management, in order to support the operational decision-making activities performed by SMEs involved in such electronic commerce networks.

In the context of the latter stream of research, this paper proposes the basis for decision support systems which may support SMEs in their electronic negotiation processes. The paper originates from a wider research project (Cantamessa et al., 2001), whose objective is the development of an electronic platform enabling B2B electronic commerce for SMEs with make-to-order (MTO) operations. As clearly presented by Hendry and Kingsman (1989), the characteristics and strategic priorities of MTO companies are signicantly different from those operating on a make-to-stock basis. Smaller MTO rms are also characterised by low product and process complexity and generally restrict themselves to performing processing and assembly operations upon standardised designs, or using order-specic designs directly provided by customers. By referring to the taxonomy by Amaro et al. (1999), the former would fall into the categories MTO-3MTO-5, while the latter would more appropriately be considered engineering-to-order companies. However, the absence of design responsibility makes the order negotiation and fulllment processes very similar in both of these cases. Many examples of such SMEs can be found in industrial clusters, in industries such as textile, furniture, ceramic tiles, etc. In operations of this kind, the critical issue generally shifts from the management of materials to that of managing manufacturing capacity, since raw materials are often provided by customers or purchased using unsophisticated approaches, while the product bill-of-materials is relatively simple. In many instances, it can therefore be stated that the commodity being traded with customers is not physical goods, but rather, machine-hours. This leads to the critical need of integrating the order negotiation and production planning processes or, from an organisational perspective, of integrating the marketing and production functions (Muda and Hendry, 2002). To date, order negotiation and production planning processes are generally performed by human decision-makers, eventually supported by industry-standard information systems, such as MRP. However, for networked SMEs using the Internet in order to trade goods and manufacturing services, the frequency with which business

ARTICLE IN PRESS
T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249 235

opportunities have to be evaluated and negotiated upon increases dramatically. This makes it necessary to develop tools for effectively supporting the integrated order negotiation and production planning processes. This paper discusses how this may be pursued by at rst developing the environment within which negotiations take place, including a negotiation process model and a communication protocol enabling interaction between rms (the negotiation ontology). Subsequently, by developing decision support models allowing rms to make effective decisions within the negotiation environment. The following section of the paper proceeds with an overview of the two streams of literature that study production planning and order selection on the one side and automated negotiation on the other. Section 3 will present the criteria and the hypothesis that guided the work and a formal model of a multi-tier negotiation process. Section 4 will illustrate the optimisation models developed in order to support the decision-making problems that have been identied within such process. The models have multiple objectives and may be used in order to evaluate the protability of proposed orders being offered by customers with respect to existing rm orders and to the expected value of the future inow of requests for quotation. Sensitivity analysis then allows the planner to interactively evaluate the tradeoff effects that would emerge if, during negotiation with customers, he/she should deviate from the optimally suggested plan. Section 5 presents the results from a limited-size example of the bidding and negotiation process.

2. Literature review There is without a doubt a vast amount of literature that deals with the production planning problem. Solution approaches to production planning problems generally rely on multi-period optimisation models based on linear programming (LP) or mixed-integer linear programming (MILP) formulations (Thomas and McClain, 1993). Inherent in such formulations is the splitting of the planning horizon in a nite number of time

buckets. Of general concern in production planning literature is the problem of operating hierarchically, so that there may be coherence among aggregate-level and long-term plans down to the most detailed individual product-level and short-term plans (Bitran and Tirupati, 1993). In the current problem, in which SMEs operate on an MTO basis and have to decide upon individual order proposals, it is necessary to adopt a perspective which is rather detailed concerning product granularity, since each order is different from the others, and short-term concerning the planning horizon. So, hierarchical aspects of production planning are of relatively little concern in the current problem. When dealing with lowlevel production planning, a distinction is generally made in literature among models with large and small time-buckets (Brandimarte and Villa, 1995). In the former case, it is assumed that each product launch (i.e., the time required by a resource to perform manufacturing operations on a batch of parts) lasts substantially less than the time-bucket. Setup operations occurring among consecutive time buckets will be unrelated, and the formulation of the planning problem becomes quite easy and straightforward. In the latter case, product launches and time buckets have a comparable size. Setup operations occurring in consecutive time buckets will therefore be related, which leads to a somewhat more complex formulation. For the sake of simplicity, the optimisation models presented in this paper will assume a large time bucket. Should a specic application require it, they may easily be changed in accordance. Concerning the order selection problem, there is a relatively shallow body of literature on the subject. In most of these papers (Cakanyildirim et al., 1999; Kate, 1994), order selection is generally seen as a variation to a scheduling problem. The idea of approaching the order selection problem from a scheduling perspective has the desirable feature of leading to a detailed representation of machine workload and of the progress of orders during planning time. However, in the current context of negotiation support one can wonder whether such level of detail is really required. Moreover, the negotiation process may

ARTICLE IN PRESS
236 T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249

involve multiple bidding iterations in a limited time span, which suggests that solutions must be obtained very quickly. Computational complexity therefore becomes a critical issue. Concerning order negotiation, a number of authors have applied this concept for managing shop-oor operations. In these papers, the elements in a manufacturing system (resources and/ or parts) are viewed as autonomous agents and scheduling is generally performed by distributed (or heterarchical) negotiation. Research following this approach includes the Contract Net implementation proposed by Van Dyke Parunak (1987) and the work by Shaw (1988), Dufe (1990), Lin and Solberg (1992), Dufe and Prabhu (1994). A comparison between different negotiation protocols has been made by Krothapalli and Deshmukh (1999). It must be mentioned that in all these instances, the agents undertaking negotiation are part of the same manufacturing system. So, while it is true that each of them negotiates autonomously and maximises its own utility, the overall objective of research is to devise negotiation mechanisms that, through this autonomous negotiation, may actually maximise the joint utility of the entire collection of agents. In the present study, instead, the agents undertaking negotiation are not part of the same manufacturing system, but belong to different rms. This implies that agents will only strive to maximise their own utility and that it will not be possible to impose cooperative negotiation mechanisms upon them. Authors proposing support techniques for bidding negotiation in the context of multiple rms include Moodie and Bobrowski (1999) and Qinghe et al. (2001). Research on negotiation is anyway much broader, and of course does not only deal with manufacturing applications. In particular, Holsapple et al. (1998), formulated a proposal for a rigorous denition of syntax and functions for a negotiation system. Another perspective is suggested by Beam and Segev (1997), who analysed in depth the differences between three theoretical approaches (i.e., human factors, game theory and intelligent agents) in solving both cooperative negotiation problems and non-cooperative and competitive ones. These authors show the benets of using the approach of intelligent agents in a

non-cooperative environment for supporting the eld of electronic commerce and, at the same time, they indicate the shortcomings of other approaches. The authors also state a few fundamental elements of negotiation that should be considered in an electronic commerce environment.

3. The negotiation model 3.1. Criteria and hypothesis According to Gebauer and Segev (1999), purchasing activities can be divided into three basic phases: 1. information phase, in which prospective buyers identify their needs and evaluate potential sources and suppliers, gathering information about products and services, 2. negotiation phase, in which the objective is to reach an agreement by determining prices, goods availability, delivery terms and capacity constraints, 3. settlement phase, in which the order is executed and goods and services are transferred between enterprises. These three phases must be included in a successful electronic commerce platform (Merz, 1998), and have been considered while developing the model described in this paper. The model is oriented to a fairly general industrial scenario, whose value chain is composed of three or more fragmented tiers made up of customers, suppliers and subcontractors. All these rms are hypothesised to be connected in real time to an electronic negotiation platform that operates as a message exchanging device. The platform does not possess any intelligence in the sense of being able to support the decisions made by the rms in the negotiation process. The negotiation process has been modelled following basic examples in literature (Kersten and Szpakowicz, 1999) and according to three different assumptions:
*

when negotiating with customers, suppliers do not query subcontractors but implicitly estimate

ARTICLE IN PRESS
T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249
Customer
1

237

Supplier
(k)

Last Supplier
K

M2
Offer Request

M1 M3
Bid Reply Order Settlement

Subcontract Info

M2 M3

Subcontract Info

Offer Request Bid Reply

M1

Order Settlement Modify Production Plan

M2

PP

Modify Production Plan

Fig. 1. Structure of the negotiation process.

their capacity, prices, and delivery times. This has been decided in order to avoid the explosion of queries and evaluation processes, since we assume it would be inefcient to involve too many enterprises in a negotiation process where no rm order has been settled yet, there is complete freedom on order/offer evaluation, and, most of all, on the suppliers opportunity to modify bid prices, capacities and due time. This assumption gives wider validity to the model, while tightening constraints over these parameters is always possible, during the negotiation process, rms must evaluate and negotiate on potential orders, but may not re-discuss their previous commitments. This should grant greater stability to production plans and reliability to the terms of service being negotiated. It is anyway possible to relax this assumption by enabling the cancellation of a rm order (with the possible charging of a penalty) and by re-considering the order as if it were a new one.

3.2. The negotiation model The negotiation process among suppliers and customers in a value chain is depicted in Fig. 1. Fig. 1 shows that negotiation within a multi-tiered supply chain may be studied by focusing on an elementary negotiation process that occurs in a recursive fashion among a rm at tier k and its supplier at tier k21: The process starts with an offer request, or request for quotation, in which a rm at tier k asks its suppliers at tier k21 to examine a job and to submit bids for it. The offer request must include a technical description of the job. In the context of SMEs operating upon an MTO basis, this may be hypothesised to consist of the process plan of the parts to be manufactured, including the type of resources required and related processing times. At this stage, each supplier uses a model for offer request evaluation, dubbed M1. This model allows to examine the proposed job with respect to (i) rm orders within the suppliers internal production plan, (ii) estimates of the possibility of subcontracting part of the work to be done and of the related costs and (iii) an estimate of the expected value of future order inow. This latter element is of crucial importance since, with limited resources, a supplier may decide to forgo a current

The model has been represented with the unied modelling language (UML) notation, because it represents the standard for complex system modelling and provides a basis for implementation (Quatrani, 1998; Rumbaugh et al., 1998).

ARTICLE IN PRESS
238 T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249

but scarcely protable offer request in expectation of future, more protable, orders. This task requires forecasting techniques such as those normally used in revenue management (Yeoman and Ingold, 1999), the description of which is however beyond the scope of this paper. As a result of M1, the supplier responds with a bid reply which contains a quotation (price) for the required job and, possibly, changes to the job (e.g., delivery dates and/or proposals for a partial fulllment of the job). The process is now in the hands of the customer, who gathers competing bid replies submitted by suppliers and chooses the best one(s). This is performed through a model for bid evaluation, dubbed M3. Model M3 provides, as an output, an order settlement, i.e. a statement that awards the order to the winning bidder and binds both parties contractually. The supplier must now react to the order settlement. The order settlement is not a simple proposal like the offer request was, but a rm order he is now committed to fullling. He must therefore re-formulate his production plan through a model named M2. Aside from updating the production plan, this model makes a decision upon what activities should be subcontracted out. This latter decision leads to offer requests sent to suppliers at tier k22: The whole process goes on until the last tier of the supply chain is reached. It should be noticed that models M1 and M2 are engaged in virtually the same problem. M1, upon receiving order requests, must decide which of them to full and, to do so, it must make a preliminary plan concerning the degree with which each job will require subcontracting. However, the only output of M1 is towards the customer, since the subcontracting plans are kept private and are not shared with suppliers. Model M2 is similar to M1, since the order settlements which it receives have the same structure as order requests. The difference is that these are now rm commitments and must not be decided upon any longer (i.e., the related decision variables will now be constrained). Moreover, the subcontracting plans made by M2 are not kept private any longer, but forwarded to its suppliers. In order to preserve the recursive

structure of the negotiation process, this is once more made in the form of an order request. In the negotiation model, it is assumed that information exchanges between layers and decision making within a layer are fast with respect to the frequency of incoming orders. If not, new offers may be sent to a rm while it is still waiting to know the results of its previous bidding, or is waiting for bids due from its subcontractors. Such uncertainty about ongoing negotiations of course would not allow the rm to make a sound decision. This hypothesis is supported by the fact that, within the wider research project, this process will occur on a negotiation platform which all rms are connected to in real time. Moreover, the market-makers that manage the platform may enforce rules stating that orders must be exchanged in batches either at sufciently distant intervals and, in any case, never before the previous negotiation round has terminated. The negotiation process is represented in detail using UML diagrams in the following sections. 3.3. The static view UML allows to describe complex systems by visualising, specifying and documenting different points of view. It is possible to look at a complex system by specifying its static structure and its dynamic behaviour. In the case we are analysing, the static view of the system is represented by the Class Diagram (Fig. 2), which shows classes and relationships between them. The attributes found in the class diagram are specic to the particular implementation of routines M1, M2 and M3 discussed in Section 4 of this paper. The system is composed of three types of classes: 1. classes that represent the decision-making routines used by the system (M1, M2, ModelOne, ModelThree), 2. classes that maintain memories of order and bids processed (Order and BidReplly), 3. classes that store useful information for the system (ProductionPlan and SupplierInformation). The rst type of classes carries information needed by routines (attributes of such classes) and

Order referring Job : int referring Resource : int requirements Value : float planning Date : date price Job : float is Accepted : int order Status : string referring Supplier : int place Order() set Order()

analyze

M1 analyze Order()

set analyze

set

set Bid Reply percentage Of Job : float replied Planning Date : date replied Price Job : float referring Job : int referring Resource : int referring Supplier : int resource Cost : float bidStatus : char analyze set Bid Reply() modifies

M2 solve Order()

Model Three resource Amount Allocated : float referring Job : int referring Resource : int referring Supplier : int planning Date : date requirements Value : matrix resource Amount Bidded : float bidding Price : float planning Horizon : float resource Reservation Coeff : float capacity Of Resource InTime : float capacity Of Resource OverTime : float unit Cost Use Resource InTime : float unit Cost Use Resource OverTime : float opportunity Cost : float setup Cost For Resource : float fixed Cost For Job : float analyze Bid Reply() modify Parameters()

Model One percentage Of Job : float resource Amount Allocated_y : float referring Job : int referring Resource : int resource Amount Subcontracted : float completion Time : date planning Date : date requirements Value : matrix penalty Tardiness : float planning Horizon : float capacity Of Resource InTime : float capacity Of Resource OverTime : float unit Cost Use Resource InTime : float unit Cost Use Resource OverTime : float unit Cost Use Resource Outsource : float opportunity Cost : float price Job : float setup Cost For Resource : float fixed Cost For Job : float modify Parameters()

T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249

ARTICLE IN PRESS

assume

modifies Production Plan resource Amount : float referring Job : int referring Resource : int referring Time : int update()

Supplier Info referring Job : int referring Resource : int referring Time : int unit Cost Resource Outsource : float referring Supplier : int

Fig. 2. Class diagram.

239

ARTICLE IN PRESS
240 T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249

information of who is performing them and for which order/bid set. M1 and M2 differ in the operations that they perform and in the different use of order and bid objects. However, as it will be explained later, they are exactly the same routine and, for this reason, they are modelled as child subclasses, inheriting all the attributes of the ModelOne class. The second types of classes represent all the sensitive objects involved in the system (orders and bids) and the information useful to represent them. As previously stated, the Offer Request is treated as an Order object with a status value indicating that it has not yet been approved. Finally, the SupplierInfo class maintains all the information that a supplier assumes about its subcontractors at the moment of taking a decision. This information is key to running the M1 module. Production Plan represents the Customer/Supplier production plans, which are affected by the results of the negotiation process and order implementation. The production plan is updated every time the ModelThree routine is performed and an order/bid implemented. Moreover, when the chain of subcontracting ends (which is the case when the last supplier does not outsource), the production plan is updated by the M2 class.

3.4. The dynamic view Following UML notation, the behaviour and the functionalities of the system are represented using different perspectives and levels of detail. The core of the system is indicated by the Main Use Case Diagram, which shows the set of use cases involved, actors and their relationships. UML Use Cases depict the processes within which users will interact with the system. Therefore, they show the functionalities and capabilities provided by the latter within its actual use. The actors involved are the Customer and the Supplier, which represent the roles that a rm interacting with the system can take on: as we have previously seen, depending on the situation, the same rm can act either as a supplier or as a customer in different negotiation processes.

As depicted in Fig. 3, the supplier uses all of the three procedures, while the customer role uses BidCreation in order to create an order, and the BidSelection to evaluate suppliers responses. OfferSelection uses BidCreation because, as previously stated, the routine is the same, but with different functional purposes. ProductionPlanUpdate is represented as a so-called (in UML jargon) extend of M2 and M3, being an optional result of these processes. For each use case represented in the Use Case Diagram, the corresponding sequence and collaboration diagrams have been modelled. Sequence and collaboration diagrams are interaction diagrams and detail object interaction and message dispatching. The sequence diagram emphasises the chronological ordering of messages, while the collaboration diagram emphasises the structural organisation of the objects that send and receive messages. Due to their isomorphic property, one can transform one diagram into the other without loss of information. An example of the diagrams developed for this negotiation model is represented in Fig. 4. The gure shows the time sequencing of the M3 routine (bid selection), with the indication of which operation is involved in each step of the process. Each operation details which parameters are used to perform requests. It is possible to notice that the M3 routine can be run more than once, modifying different internal parameters every time, in order to reach the most convenient allocation of resources and combination of variables. In these diagrams, the only perspective shown is time and the sequential relation occurring between actors and classes. A broader view and the logical processes of system functionalities is given by the activity diagram, a particular statechart diagram focused on activities performed by the system. In this diagram, illustrated in Fig. 5, the complete ow of events and activities is modelled, by which it is possible to understand the logical sequence of different operations performed during the negotiation process. Control guards and decision steps are also highlighted, in order to provide more information about the systems behaviour.

ARTICLE IN PRESS
T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249 241

customer send Bid Reply() move Next()

<< Interface >> create order << uses >>

Bid Selection Offer Selection Bid Creation

<< extends >> << extends >>

supplier islast : type = initval send Order Request() send Order()

Production Plan Update

Fig. 3. Use case diagram.

: B idReply : supplier : custo me r

: ModelThree

: Order

: P roduction P lan

sendB idReply(percentageOfJob, repliedP lanningD ate, repliedP riceJob)

analyzeB idReply(requestedduedate, requirementsMatrix) modifyP arameters( ) update(resourceA mount) setOrder(planningD ate, priceJob, requirementsV alue)

This comes from Offer S election S equencing

Here S tarts the B id C reation S ecquencing

Fig. 4. Example of sequence diagram (bid selection).

ARTICLE IN PRESS
242 T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249

C usto m er S id e C reate O ffer R equ est

S upplier S id e

N et w ork/D B U pdate G et O ffer M odel Im plementation M1 [not ac c ept] [a c cept ] S et B id R eply N et w ork/D B U pdate G et B id R epl y M odel Im ple m e ntation M3 N et w ork/D B U pdate G et O rd er

[not ac c ept]

M odel Im plementation M2 [a c cept]

[not ac c ept]

[a c cept]

U pdate P ro ducti on P lan

S et O rd er

[islast= = yes] U pdat e P ro duction P lan [islast= = no ] C re ate O ffe r R equ est

Fig. 5. Activity diagram.

4. The decision-making models The approach has been tested by implementing routines M1, M2 and M3 as MILP models. In order to show the exibility that the approach grants to decision-makers, the models have been formulated in multi-criteria decision-making terms. Techniques such as Goal Programming may eventually be used as a solution approach, but their application goes beyond the scope of the paper. The criteria proposed in this paper are prot, resource utilisation and workload levelling. Of course, the models presented in the following are by no means unique and, as long as they t within the negotiation model discussed above, they may be modied according to the needs of specic applications. 4.1. Models M1 and M2 As a reminder, model M1 has the purpose of examining incoming order requests and to respond

with bids, at the same time making plans for work to be subcontracted (but keeping them private). Model M2 instead receives order settlements (i.e., rm orders) and must update the production plan in accordance, again making plans concerning work to be subcontracted, but using them in order to send offer requests to suppliers. Despite this different context within the negotiation process, models M1 and M2 share the same MILP formulation. The ow-charts that show the usage of the two models are provided in Fig. 2. Having run M1 and M2, sensitivity analysis may be performed upon the solution. For model M1 the decision-maker may act upon system-level parameters (e.g., available capacity, operative costs, etc.) and, mostly, upon parameters related to the bidding process. For example, one may force the acceptance or the denial of proposed jobs, change due dates, split orders in smaller batches and, most of all, change price, which will generally be the main

ARTICLE IN PRESS
T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249
Order request (RFQ) Production plan Screening and prioritizing System capacity and capability

243

Order settlement Production plan System capacity and capability

M1

M2

Sensitivity analysis

Sensitivity analysis

Change system parameters

Modify parameters?

Change negotiation parameters

Change system parameters

Modify parameters?

(a)

Bid

Order requests

(b)
Fig. 6. Interactive use of models M1 (a) and M2 (b).

issue around which negotiation rounds occur. It must be noted that, with model M1, the production plan is read but not updated since orders are only requested but not yet settled. Model M2 is activated when bidding has ended so that, at this stage, only system-level parameters may be acted upon. In this case the production plan is not only read but also re-written (Fig. 6). Sets i 1; y; m; jobs (offer requests or order settlements), j 1; y; n; resources, t 1; y; T; time buckets. Decision variables xi ; portion of job i the rm should bid upon, vi 1 if job i is accepted, =0 otherwise (binary) yijt ; amount of resource j allocated to job i in time bucket t; including regular time, overtime and outsourcing, ojt ; amount of overtime for resource j in time bucket t; sjt ; amount of resource j to be subcontracted in time bucket t; zi ; time bucket at which job i is completed, i.e., completion time (integer), git 1 if job i is being worked upon at time bucket t =0 otherwise (binary),

TDi ; tardiness of job i (integer), C maxj ; maximum usage of resource j; C minj ; minimum usage of resource j: Parameters di ; due date of job i; rij ; amount of resource j absorbed by job i; qi ; tardiness penalty for job i; per time bucket late, CAPjt ; regular-time capacity of resource j available in time bucket t; CAPjt ; maximum overtime capacity of resource j available in time bucket t; cj ; unit variable cost of utilizing resource j in regular time, c0j ; unit variable cost of utilizing resource j in overtime, c00 ; estimated unit cost of utilizing a subcontracted j resource j; pjt ; unit opportunity cost for resource j (from demand forecasts), pi ; bidding price of job i; FCi ; xed cost of accepting job i; M; a very large number. Constraints X yijt Xxi rij 8i; j;
t

1 8j; t; 2

X
i

yijt pCAPjt ojt sjt

ARTICLE IN PRESS
244 T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249

ojt pCAP0jt

8j; t;

3 4 5 6 7 8

sjt pCAP00 jt 8j; t; X yijt pMgit 8i; t;

j

zi tpM1 git 8i; t; TDi Xzi di 8i; xi pMvi 8i; X

i

C maxj X sjt C minj p sjt xi X0 8i; 8i;

yijt

8j; 8j;

9 10 11 12 13 14 15 16 17

X
i

yijt

vi Af0; 1g yijt X0

8i; j; t; 8j; t; 8i; 8i; t;

ojt ; sjt X0 TDi ; zi AI git Af0; 1g

C maxj ; C minj X0 8j: Objectives X X X x i pi TDi qi FCi vi max G1

i i i

X X
jt j;t

cj

X
i

yijt sjt ojt c00 sjt j ! 18

c0j ojt pjt

X
jt

X
jt

X
i

yijt sjt ;

TC maxj C minj ; min G2j P t CAPjt CAPjt P P t i yijt sjt min G3j P : t CAPjt CAPjt

19

20

Constraint (1) forces to allocate sufcient resources to the set of accepted jobs. Constraint (2)

ensures that capacity allocated to accepted jobs does not exceed the limits of regular capacity, plus planned overtime, and planned subcontracting. Constraints (3) and (4), respectively require overtime and subcontracting not to exceed maximum limits. Constraint (5) ties boolean variable git to the allocation of resources on job i and time bucket t: Constraint (6) sets the completion date of job i; zi ; to a time bucket where i will no longer being worked upon, while Constraint (7) sets tardiness. It has been chosen to set tardiness as a decision variable, with a corresponding penalty, in order to grant the planner more latitude in examining offer requests and negotiating upon them. Constraint (8) ties the acceptance of a job to the fraction with which it is accepted. Again, the choice of allowing suppliers to bid upon fractions of jobs has been made in order to provide greater exibility, but may be easily removed. Constraints (9) and (10), respectively set maximum and minimum usage of resources. Eqs. (11)(17) are non-negativity and integrality constraints for decision variables. Goal (18) expresses the goal of prot. Prot is given by revenue from accepted orders and by detracting lateness penalty costs, xed costs for accepted orders, manufacturing costs with regular, overtime and subcontracted capacity, and the expected opportunity cost due to future order proposals. This latter element is very important since, as already mentioned in Section 3.2, by neglecting the expected revenue stream that may accrue from future orders, the planner would run the risk of accepting offer requests with low protability. This opportunity cost cannot be calculated from the current order portfolio through sensitivity analysis, but must be forecast with appropriate techniques, as previously mentioned. Relative uctuation in usage of resource j is a second goal, given by Eq. (19). Finally, utilisation rate of resource j is given as another goal by Eq. (20). Computational complexity is fairly low, since there are mT 1 binary variables and 2m integer ones (for both of which the value is bounded to T). This ensures fast execution of the model and, therefore, the possibility of using it interactively as shown in Fig. 2.

ARTICLE IN PRESS
T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249 245

4.2. Model M3 Model M3 is used within the negotiation process in order to examine competing bids submitted by suppliers and to select the most favourable ones. Module M3 is built around the idea that bidding rms may accept to process portions of the offer requests they receive. So, M3 must stitch together the bids received in order to create a feasible minimum-cost plan with which the rm may comply with its previous commitments. When running model M3, the production plan is rst read and then updated. Because of the previous assumptions, the entire process progresses starting downstream of the value chain (i.e., closer to the market) and moves upstream (i.e. towards suppliers) without any feedback ltering above the two tiers that are currently negotiating. When running M3, a rm is therefore allowed to change plans related to orders under negotiation, but not to change commitments (rm orders) previously made to its customers. In other words, it can neither re-issue order requests and ask suppliers to submit bids again, nor can it reverse previous decisions upon accepted jobs. Of course, this is a limitation, since the rm may discover that previously taken commitments are not feasible or that, simply put, they are not as protable as supposed beforehand. Such a mismatch can occur because rms must bid on offer requests without having a clear view of which jobs will eventually be won and can only estimate supplier prices and capacity. Changing such assumptions may be considered a further step in research. This change would require to perform rounds of negotiation backwards and forward along the entire value chain, making the emergence of stability problems very likely. Moreover, this would require a signicant structural change to the negotiation process model, with the current upstream process (with orders ltering from customers up to suppliers) set side by side to a new downstream process (with cancelled orders proceeding from suppliers down to customers). Fig. 7 shows the ow chart depicting the usage of model M3. Having run M3, sensitivity analysis may be performed upon the solution, thus providing hints for an interactive change of

parameters. In the current setting, the decisionmaker may act upon system-level parameters (e.g., available capacity, operative costs, etc.) and, as mentioned, he may not modify parameters related to the bidding process (e.g., changing the offer requests previously distributed to suppliers). Sets i 1; y; m; jobs (offer requests or order settlements), j 1; y; n; resources, k 1; y; l; bidder, t 1; y; T; time buckets. Decision variables xk 1 if bid from bidder k is to be accepted, =0 otherwise (binary), yijt ; amount of resource j allocated to job i in time bucket t; including regular time, overtime and outsourcing, C maxj ; maximum usage of resource j; C minj ; minimum usage of resource j: Parameters di ; due date of job i (integer), rij ; amount of resource j absorbed by job i; sjkt ; amount of resource j bidder k competes for at time bucket t; bk ; bidding price of bidder k; tk ; due date proposed by bidder k; bj ; fraction of resource j reserved for future jobs, CAPjt ; regular-time capacity of resource j available in time bucket t; CAPjt ; maximum overtime capacity of resource j available in time bucket t; cj ; unit cost of utilizing resource j in regular time, c0j ; unit cost of utilizing resource j in overtime, pjt ; unit opportunity cost for resource j (from demand forecasts), SCk ; xed cost of accepting bid k: Constraints X yijt Xrij 8i; j;
tpdi

21

ARTICLE IN PRESS
246 T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249

Competing bids Production plan System capacity and capability

M3

Sensitivity analysis

Change system parameters

Modify parameters?

Order settlements
Fig. 7. Interactive use of model M3.

X
i

yijt p 1 bj CAPjt ojt X

k

min G1 22

X
k

SCk bk xk cj X
i

xk sjkt

8j; t;

X X
jt j;t

yijt

X
k

! sjkt xk ojt

ojt pCAP0jt

c0j ojt pjt X

i

8j; t; X
k

23 X
i

yijt 8j; t; 24

X
jt

yijt

X
k

! sjt xk ; 28

C maxj X

xk sjkt

C minj p

X
k

xk sjkt

X
i

TC maxj C minj min G2j P ; t CAPjt CAPjt yijt 8j; t; 25 P P P yijt k sjkt xk t P i min G3j : t CAPjt CAPjt

29

30

xk Af0; 1g

8k;

26

yijt X0

8i; j; t:

27

Objectives

Constraint (21) forces to allocate sufcient resources to accepted jobs and to do so within the agreed due dates. Constraint (22) ensures that capacity allocated to jobs does not exceed regular capacity, planned overtime and the set of accepted bids that have been submitted by subcontractors. Constraint (23) requires overtime not to exceed its

ARTICLE IN PRESS
T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249 247

5b. Decides not to bid

4. Plans production and issues offer request (M2) 6. Awards order to C4 (M3)
10
20 10 1 2

30

20

10

1 2 3 4 5 6 7 8 9 10 t 3 4 5 6 7 8 9 10 t

maximum limit. Constraints (24) and (25), respectively set maximum and minimum usage of resources. Constraints (26) and (27) are nonnegativity and integrality constraints for decision variables. The rst goal (28) expresses cost-minimisation. Total relevant costs are given by subcontracting costs, plus internal manufacturing costs, plus overtime costs, plus expected opportunity cost due to order proposals not yet arrived. As in the previous model, a second goal is concerned with relative uctuation in usage of resource j (29). Finally, a third goal is related to utilisation rate of resource j (30). Computational complexity for M3 is even lower than for models M1 and M2, since there are l binary variables and no integer ones.

5c. Responds (M1) and bids with Price = 2600 Completion date = 4 7. Plans production (M2) without further subcontracting

Regular / overtime capacity R2 30 20 10 20

C4 (R2)

30

5a. Decides not to bid

10

1 3

2 4

3 5

4 6

5 7

6 8

7 8 9 10 t 9 10 t

Regular / overtime capacity R1

C1 (R1, R2)

Negotiation platform

C3 (R1)

5. Simulation example In order to clarify the concepts discussed above and to show the performance of the negotiation system, an illustrative example of limited size is provided. In the example, two companies (C1 and C2) are connected to the negotiation platform and operate two resources (R1 and R2), while other two companies (C3 and C4) specialise in one resource only. The plan runs over a horizon that is 10 time buckets long and each company has an order portfolio made up of four to six jobs. For the sake of brevity, details of the order portfolio of each company are omitted. A rst-tier negotiation round is shown on the left-hand part of Fig. 8, with company C1 running module M2 in order to develop its production plan. As the resource workload charts show, this requires the subcontracting of work on resources 1 and 2 over the rst four time buckets. This offer request is sent to the negotiation platform and forwarded to the other three companies. Company C2 decides to bid on the offer request with its internal resources. Its threshold price for accepting the bid is 8290 and with a completion date of 8, i.e. 4 time buckets late. Company C3 too decides to bid and expects to subcontract work on the resource R2 it does not possess. It proposes a threshold price of 8310 and completion at time bucket 4. We suppose company C4 decides not to

1. Plans production and issues original offer request (M2) 3. Awards order to C3 (M3)

Regu 10 Regular / overtime capacity R1 10 1 2 3 4 5 6 7 8 9 10 t 1 2 3 4 5 6 7 8 9 10 t

2b. Responds (M1) and bids with Price = 8310 Completion date = 4

Regular / overtime capacity R2

2c. Decides not to bid

C4 (R2)

2a. Responds (M1) and bids with Price = 8290 Completion date = 8

10 Regular / overtime capacity R1 1 2 3 4 5 6 7 8 9 10 t 1 2 3 4 5 6 7 8 9 10 t

30

Regular / overtime capacity R2

C2 (R1, R2)

30

30

20

20

10

30

20

1 2 3 4 5 6 7 8 9 10 t 3 4 5 6 7 8 9 10 t

30

20

Regular / overtime capacity R1

30

20

C1 (R1, R2)

Negotiation platform

C3 (R1)

Fig. 8. Simulation of a negotiation process.

C2 (R1, R2)

30

20

lar / overtime ov ertime

10

ARTICLE IN PRESS
248 T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249

bid because it has the policy of not accepting orders for which it does not own resources. At this point company C1 runs module M3 and selects the offer by company C3 (more expensive but on time). Company C3 must now review its production plan with the addition of the rm order that company C1 has just awarded it (right-hand part of Fig. 8). Module M2 is run and gives birth to an offer request. Companies C1 and C2 decide not to bid on it, while company C3 runs model M1 to respond. The proposed price is 2600 and the due date is 4. Company C3 runs model M3 and awards the job to the sole bidder. Finally, company C4 runs module M2 to update its production plan, without the need of further subcontracting. The negotiation process just described is very quick to perform and each iteration of modules M1, M2 and M3 took under 3 seconds of processing time on a Pentium III 700 MHz PC. Of course, this is due to the limited size of the example. However, larger examples have been run in a time span that is fully compatible with the objective of providing decision support in industrial settings.

The subject may be explored by further research in a variety of different directions. The most immediate ones consist in increasing the degree of automation with which decisions covered by models M1M3 are made (see the owcharts in Figs. 6 and 7). Moreover, to study ways with which the negotiating rms may be helped to reach Pareto optimal solutions without engaging in too many time-consuming negotiation rounds. This requires to use the models M1 and M3 in order to compute the rm- and situation-specic tradeoff curves among relevant variables, such as price and due date of jobs.

Acknowledgements This research has been partially supported by grant MM 01274435-011 of the Italian Ministry of Instruction, University and Research.

References
Amaro, G., Hendry, L., Kingsman, B., 1999. Competitive advantage, customisation and a new taxonomy for maketo-order companies. International Journal of Operations and Production Management 19, 349371. Beam, C., Segev, A., 1997, Electronic Catalogs and Negotiations, Technical Report 96-WP-1016, University of California at Berkeley. Bitran, G.R., Tirupati, D., 1993. Hierarchical production planning. In: Graves, S.C., Rinnooy Kan, A.H.G., Zipkin, P.H. (Eds.), Logistics of Production and Inventory. NorthHolland, Amsterdam. Brandimarte, P., Villa, A., 1995. Advanced Models for Manufacturing Systems Management. CRC Press, Boca Raton, FL. Cakanyildirim, M., Chen, D., Chen, P., Freimer, M., Jackson, P.L., Melkonian, V., Roundy, V., 1999. Capacity-driven acceptance of customer orders for a multi-state batch manufacturing system: Models and algorithms, OR&IE Technical report 1233, Cornell University, Ithaca, NY. " Cantamessa, M., Nicolo, V., Pepe, A., Villa, A., 2001. Negotiation models and production planning for virtual enterprises. Proceedings of the V AITEM Conference, Bari, September 1820, 2001. Counsell, A., 2001. Small is the next big thing. Connectis, 12, June. Dufe, N.A., 1990. Synthesis of heterarchical manufacturing systems. Computers in Industry 14, 167174.

6. Conclusions The paper has discussed in detail a negotiation process model and three decision-making problems that are representative of the negotiation rounds that may be found within a multi-tiered value chain operating upon a MTO basis. In such application domain, the main decision process consists in interrm negotiation upon requests-forquotation distributed by customers and upon the related bids submitted by suppliers. The negotiation process has led to the identication of the three problems of offer request examination (when a supplier responds to a customer with a bid), bid evaluation (when a customer must select a bidder) and offer request creation (when a customer must create an offer request to be distributed to suppliers). By dealing with an MTO environment, this negotiation process is tightly integrated to production planning and thus requires specic decision-support techniques.

ARTICLE IN PRESS
T. Calosso et al. / Int. J. Production Economics 85 (2003) 233249 Dufe, N.A., Prabhu, V.V., 1994. Real-time distributed scheduling of heterarchical manufacturing systems. Journal of Manufacturing Systems 13, 94107 (real-time model in business). Engelbrecht-Wiggans, R., 1980. Auctions and bidding models: A survey. Management Science 26, 119142. European Information Technology Observatory, 2000. European Information Technology Observatory 2000. EITO, Frankfurt. Gebauer, J.A., Segev, A., 1999. Assessing Internet-based Procurement to Support the Virtual Enterprise, unclassied Working Paper, University of California at Berkeley, http:// groups.haas.berkeley.edu/citm/publications/papers/cals.pdf. Hendry, L.C., Kingsman, B.G., 1989. Production planning systems and their applicability to make-to-order companies. European Journal of Operations Research 40, 115. Holsapple, C.W., Lai, H., Whinston, A.B., 1998. A formal basis for negotiation support system research. Group Decision and Negotiation 7, 203227. Jagdev, H., Thoben, K.-D., 2001. Anatomy of enterprise collaboration. Production Planning and Control 12, 437451. Kate, H.A., 1994. Towards a better understanding of order acceptance. International Journal of Production Economics 37, 139152. Kersten, G., Szpakowicz, S., 1999. Modelling Business Negotiations for Electronic Commerce. IIASA report IR-98-015/ march, International Institute for Applied Systems Analysis, Laxemburg, Austria. Krothapalli, N.K.C., Deshmukh, A.V., 1999. Design of negotiation protocols for multi agent manufacturing systems. International Journal of Production Research 37, 16011624. Lin, G.Y., Solberg, J.J., 1992. Integrated shop oor control using autonomous agents. IIE Transactions 24, 5771. 249

Merz, M., 1998. Supporting electronic commerce transactions with contracting services. International Journal of Cooperative Information Systems 7, 125. Moodie, D.R., Bobrowski, P.M., 1999. Due date demand management: Negotiating the trade-off between price and delivery. International Journal of Production Research 37, 9971021. Moore, J.F., 1997. The Death of Competition: Leadership and Strategy in the Age of Business Ecosystems. Harper Collins, New York. Muda, S., Hendry, L., 2002. Proposing a world-class manufacturing concept for the make-to-order sector. International Journal of Production Research 2, 353373. Qinghe, H., Kumar, A., Shuang, Z., 2001. A bidding decision model in multiagent supply chain planning. International Journal of Production Research 39, 32913301. Quatrani, T., 1998. Visual Modeling with Rational Rose and UML. Addison-Wesley, New York. Rumbaugh, J., Jacobson, I., Booch, G., 1998. The Unied Modeling Language Reference Manual. Addison-Wesley, New York. Shaw, M.J., 1988. Dynamic scheduling in cellular manufacturing systems: A framework for networked decision making. Journal of Manufacturing Systems 7, 8394. Thomas, L.J., McClain, J.O., 1993. An overview of production planning. In: Graves, S.C., Rinnooy Kan, A.H.G., Zipkin, P.H. (Eds.), Logistics of Production and Inventory. NorthHolland, Amsterdam. Van Dyke Parunak, H., 1987. Manufacturing experience with the contract net. In: Huhns, M. (Ed.), Distributed Articial Intelligence. Morgan Kaufmann, London. Yeoman, I., Ingold, A., 1999. Yield management strategies for the service industries. Cassel, London, New York.