Multi-agent collaboration for B2B workflow monitoring

Dongming Xu*, Huaiqing Wang

Department of Information Systems, City University of Hong Kong, Kowloon, Hong Kong, People’s Republic of China

Received 13 June 2001; accepted 7 January 2002

Abstract

Business-to-business (B2B) applications environments are exceedingly dynamic and competitive. This dynamism is manifested in the

form of changing process requirements and time constraints. However, current workflow management technologies have difficulties to solve

the challenges problems, such as: how to deal with the dynamic nature of B2B commerce processes, how to manage the distributed

knowledge and recourses, and how to reduce the transaction risk. In this paper, a collaborative multi-agent system has been proposed.

Multiple intelligent agents in our system can work together not only to identify the workflow problems, but also to solve such problems, by

applying business rules, such as re-organizing the procurement and the transaction processes, and making necessary workflow process

changes. q 2002 Elsevier Science B.V. All rights reserved.

Keywords: Workflow-monitoring system; Intelligent agent; eCommerce; Business-to-business; Business intelligence

1. Introduction

Business is moving rapidly into the Internet age as it has

moved in the information age and industrial age before.

Business-Intelligent (BI) is a key element for organization

development, and ‘sharing knowledge is power’ becomes a

critical environment for the business development [2]. At

the early stage in the evolution of business-to-business

(B2B) commerce, there is a combination of isolated data

points and well-researched predictions. According to

Forrester Research, US B2B commerce on the Internet

will increase from $43B in 1998 to $1.3T in 2003 [1]. B2B

commerce on the Internet is generating a lot of interest and

moving quickly, many companies are developing their

business via this new tool. On the B2B commerce platform,

an open, end-to-end infrastructure of interoperable software

solutions and hosted web-based commerce services has

been built. Companies can do online trade efficiently, and

can integrate and collaborate among B2B marketplaces,

buyers, suppliers and commerce service providers. Such

globally reached B2B eCommerce platform creates econ-

omies for companies around the world.

“For B2B to become a successful channel for a business,

all of the processes within that business have to be

integrated and streamlined,” suggested Jean-Marc Faven-

nec, IBM’s Director of Software Marketing for Europe, the

Middle East and Africa. “B2B is also highlighting many

instances of inefficiency within businesses, especially in the

areas of procurement and ERP, that need to be addressed

through process redesign or process integration,” he added.

In addition, many companies are exploring B2B

commerce and make their existing business more efficient,

by improving customer service, reducing inventory,

increasing market depth and liquidity, and eliminating

geographical and temporal barriers. How to manage the

workflow processing taken in such eCommerce marketplace

is a big issue to get benefits mentioned earlier. Due to such

B2B eCommerce Marketplaces provide dynamic trade, the

traditional workflow management, that is, one to many

models could not fit this flexibility requirement. In our

study, we address intelligent multi-agents to monitor the

dynamic nature of transaction processes in B2B eCommerce

marketplace.

2. Background

2.1. Workflow and workflow management

A workflow is a composite activity consisting of tasks

involving a number of humans, databases, and specialized

applications [9]. Workflow refers to group activity auto-

mation by task sequencing and information routing [10].

0950-7051/02/$ - see front matter q 2002 Elsevier Science B.V. All rights reserved.

PII: S0 95 0 -7 05 1 (0 2) 00 0 33 -3

Knowledge-Based Systems 15 (2002) 485–491

www.elsevier.com/locate/knosys

* Corresponding author.

E-mail addresses: isxu@is.cityu.edu.hk (D. Xu),

iswang@is.cityu.edu.hk (H. Wang).

Thus, workflow is a collection of tasks organized to

accomplish some definite business processes. An activity

can be performed by one or more software systems, one or a

team of human, or a combination of them [11]. This

definition applies the workflow concept to automate

business processes. Workflow management involves the

(re)design and the (re)implementation of workflows as the

needs and the goals of an enterprise.

When managing running workflows, a lead engineer

might have to adjust a workflow by adding, removing, or

reordering tasks. This requires highly adaptable workflow

functionality. Therefore, the workflow meta-model must be

expressive and flexible and workflows can be defined and

managed easily [4]. However, most of today’s Workflow

Management Systems (WFMSs) rely on one-dimensional

flat process models, in which a process definition includes

every detail of the process from beginning to end. It is hard

to specify that several resources can collaborate for

executing a task [16].

Currently, most B2B workflows arise on the Internet.

How transaction is taken place effectively and efficiently in

B2B eCommerce marketplaces is a major issue of WFMSs.

However, most WFMSs lack the functionality to support the

dynamic nature of automating B2B transaction processes.

The following lists a number of important issues need to be

addressed in the future:

† Information sharing. Workflow processes and WFMSs

need to share their internal and external resources.

† Chained execution. A workflow process is divided into a

number of sequent sub-processes, which are executed

one by one.

† Process changing. A workflow process needs to be

changed automatically, when some special situation

comes.

2.2. Intelligent agents

Software agents represent a relatively new computational

technology and are not yet well defined [15]. The concept of

intelligent agents is rapidly becoming an important area of

research [3,5,6]. Informally, the work performed by

intelligent agents carry intelligent behavior of software

agents and intelligent systems. Various researches have

been conducted to apply intelligent agent-based technology

towards real world problems.

The communication between agents is considered a

sequence of communication and computation steps [14,19].

The communication capabilities of the mediator agent are

generating and sending message to other mediator agents

and the local domain agent, as well as receiving and

decoding messages from other mediator agents and the local

domain agent, incorporating beliefs, commitments and

information from all other agents in the network. However,

agents are developed from a template design that consists of

five individual layers [8]: transport (agent message trans-

port), message (agent communication), protocol (conversa-

tion policies), agent (basic agent components, such problem

solvers), and detailed agent (domain-specific or agent

specific tasks) [17,18].

Formally, the term agent is used to denote a software-

based computer system that enjoys the following properties

[12,13]:

† Autonomy. Agents operate without the direct interven-

tion of humans

† Co-operability. Agents co-operate with other agents

towards the achievement of certain objectives

† Reactivity. Agents perceive their environment and

respond in a timely fashion to changes that occur

† Pro-activity. Agents do not simply act in response to

their environment; they are able to exhibit goal-

directed behaviors by taking the initiative

† Mobility. Agents are able to travel through computer

networks. An agent in one computer may create

another agent in another computer for execution.

Agents may also transport from computer to computer

during execution and may carry forward accumulated

knowledge and data.

2.3. Knowledge level collaboration

Communication, collaboration, and co-ordination are

different layers of interaction. Communication allows

participants in the decision process to share information

(this involves networking infrastructures), collaboration

allows participants to collaboratively update some shared

set of decisions (this involves support for tele-conferencing,

etc.) and co-ordination ensures the collaborative actions of

the individuals working on a shared set of decisions are co-

ordinated to achieve the desired result efficiently [7]. Each

layer is built on the top of the next layer. To simplify,

collaboration means more than just instantaneous com-

munication, or total asset visibility or leveraging resources

and the talents of experts from different fields. Collaboration

means all of these, and synergy effects of these. Another

concept of collaboration entails a situation, where agent is a

potential equal contributor to a discussion transcript that

becomes important as a memory for the group. Ideally such

a transcript can evolve to become a knowledge base for the

collaborators and those who use the results of the

discussions [11].

Collaboration involves creativity, innovation and

decision-making. Unfortunately, it also involves looking

for files, locating and scheduling resources and populating

databases. Automated collaboration lets the human collab-

orations do what they do best think, create, and decide and

lets the computers take care of the data and resource

management. A complete collaboration environment has all

the components necessary to ensure that knowledge can be

shared and used at the moments, when it can best impact the

product. The term of ‘collaboration’ is used to a set of

D. Xu, H. Wang / Knowledge-Based Systems 15 (2002) 485–491486

participants working together to produce a product or

service. A crucial point for successful collaboration is the

manner in which individual work is related to the group as a

whole. Co-workers make autonomous decisions when

working alone, under changing and unpredictable con-

ditions, which the group cannot foresee or plan for. To

enable a separated group of coworkers to collaborate, they

need to co-ordinate themselves. The importance of co-

ordination can be seen in the need to bring the efforts of all

coworkers together in order to produce a product or service.

The ideal workflow management needs collaboration

among all participates. Especially, such collaboration is

critical in some exceptional situations, such as when a

workflow process faces errors or external interrupt. Such

exceptional situations are difficult to be managed by current

WFMSs. In addition, in order to deal with such situations,

related participates in the WFMS have to share relevant

information and provide necessary services each other.

3. B2B eCommerce workflow model

Electronic commerce is the movement of everything

involving business to the Internet and the World Wide Web.

Potentially, eCommerce will lead to simpler, faster and

more efficient business transactions. It also shifts business

focus from retail and physical stores to the virtual business,

which will affect both producers and consumers, as shown

in Fig. 1. Considering the workflow processes in a

collaborative marketplace for B2B eCommerce, WFMSs

need the co-operation among suppliers, buyers, Online

Payment Service providers and logistics providers, and of

course, the eCommerce marketplace providers as well.

B2B Internet-based eCommerce is the value of all goods

and services purchased over the Internet by business users

(excluding advertising revenue). In Fig. 1, we present the

B2B market model including four major parts, which are

eFulfillment, Dynamic Trade, Online Payment and Logis-

tics. Transactions are generated by Internet enabled

applications facilitating eProcurement functions. The fol-

lowing is the detailed description of the model.

Buyer and supplier portals. Both buyer and supplier will

start their electronic business from information gathering

and analyzing on such portals. This is the first step of the

process in the eProcurement, which extends across the

supply chain from B2C interfaces with consumers, through

corporate intranets supporting internal processes to external

trading on B2B hubs. B2B applications are displacing many

functions previously found in EDI and proprietary networks

and constitute a major segment of the eCommerce software

applications market. Once the decision is made, they go to

the B2B eCommerce marketplace.

Dynamic trade. In B2B eCommerce marketplaces,

auction is a very important facility to provide both supplier

and buyer dynamic trading functionality. Auction creates

value by matching buyers and suppliers. Buyers use reverse

auction function to bid the products with the lower price and

satisfied supplier, and suppliers can find the satisfied and

suitable buyer by taking forward auction.

Online payment. Once the buyer gets the right products,

the online payment would be taken. Electronic payment is

created in order to support transactions. With more and

more transactions being done through the Internet, the need

for electronic payment has become essential. Electronic

payment systems, such as smart cards, Electronic Funds

Transfer (EFT), Electronic Funds Transfer at Point of Sale

(EFTPOS), direct data entry transaction, and home banking

will be developed further as electronic commerce grows.

Logistics. Through a B2B eCommerce platform, users

and marketplaces have to access a variety of logistics

Fig. 1. B2B eCommerce workflow.

D. Xu, H. Wang / Knowledge-Based Systems 15 (2002) 485–491 487

services that can be integrated into every transaction

process. Logistics services include global visibility of

order status, exception-based performance tracking and

monitoring, delivery optimization and internet-based trans-

portation management.

eFulfillment. The eFulfillment refers not only to provide

the customers with what they ordered from B2B eCom-

merce marketplace and to do it on time, but also to provide

all related customer services either from virtual market-

places or from physical world. In addition, if the customer is

not happy with a product, an exchange or return needs to be

arranged.

Usually, traditional trade between businesses involves

overly complex processes that are constrained by infor-

mation inefficiency, geography and business hours. The

Internet can facilitate a 24-h exchange that supports

business transactions. Traditional business model is ‘one

to many’, which means one supplier relates to many buyers,

such as business of department store. In B2B eCommerce

marketplaces, the business model has changed to ‘many to

many’, which means that each participate, either supplier or

buyer, has related with others, and newcomer will be

contacted to the other previous participates in the eCom-

merce marketplace.

4. Collaborative agents for workflow monitoring

The goal of this paper is to describe intelligent agents in a

collaborative eCommerce marketplace for B2B eWorkflow

monitoring. These intelligent agents operate autonomously

and co-operatively with each other to perform workflow-

monitoring tasks and to distribute knowledge, recourses and

strategies across the transaction processes.

Agent architecture, consisting an agent knowledge base,

its operational facilities and its external interface facility, is

shown in Fig. 2. The architecture specifies agent behavior

and its interactions with other agents and systems. The

external interface component manages the communication

between the agent and the outside world. The communi-

cation is message-based, and uses a simple and extensible

language for communication among agents. The operational

facility component is the central control and action part of

an agent. It holds to sub-components called Reasoning

Facility and Collaborating Facility, respectively. The

available functions are stored in the Knowledge Base

component. The Collaborating Facility sub-component is

responsible for the collaboration with other agents.

An intelligent agent is knowledge based, which includes

domain level knowledge and meta level knowledge. At the

domain level, there are two components, i.e. Domain

Knowledge and Profile. Domain Knowledge presents a

particular agent’s knowledge for the particular task, while

the history, belief, facts of this particular agent and changes

are stored in the profile time by time. At the meta level, there

are three main components, System Knowledge, Reasoning

Knowledge and Agent Goals. The System Knowledge is the

knowledge of the WFMS itself, such as goals and

configurations. The Reasoning Knowledge stores different

reasoning models.

There are a number of different kinds of intelligent

agents work together autonomously and co-operate with

each other to perform different tasks for the workflow

management. Fig. 3 shows the agent hierarchy.

Information Agent assists in information retrieval,

information filtering, or other information manipulation. It

has three sub-class agents, External Info. Searching Agent,

Internal Info. Searching Agent, and Customer Resource

Agent.

External Info. Searching Agent searches and gathers the

Fig. 2. The architecture of an agent.

D. Xu, H. Wang / Knowledge-Based Systems 15 (2002) 485–491488

relevant information and knowledge from a variety of

dispersed external resources, such a WWW resources, and

distributes the results to the other agents in the WFMS.

Internal Info. Searching Agent searches and gathers the

relevant information and knowledge from a variety of

dispersed internal resources, such as databases or file

systems, and distributes the results to the other agents in

the WFMS.

Customer Resource Agent reasons about capabilities of,

or relationship between, customers’ resources. This agent is

responsible for keeping track of the changes of Customer

Relationship Management (CRM) in the dynamic

environment.

Workflow-Monitoring Agent tracks and monitors the

status of all agents and operation of workflow processing.

For managing such WFMS running efficiently and effec-

tively, this agent tracks the knowledge flow and measures

the WFMS successfulness. When the agent identifies an

abnormal situation, it will ask the diagnostic agent to

perform further diagnosis.

Diagnostic Agent attempts to find the problem source and

to generate a set of possible solutions, after receiving a

request from the workflow-monitoring agent.

HCI Agent provides the interaction between users and the

WFMS. It can make the human–computer interface more

intuitive and encourage types of interactions that might be

difficult to evoke with a conventional interface. Users can

view the current state of the trading and monitoring

processing, convey their own opinions and arguments to

the rest of the transaction. There are two kinds of HCI

Agents: the Buyer HCI Agent and the Supplier HCI Agent.

Negotiating Agent provides supporting facilities for

negotiation between two agents. As an example, a buyer

agent may negotiate with a supplier agent about the closing

date.

Co-ordinator is an agent, that is, required to manage all

of the other agents, as well as support the system operation.

The Co-ordinator ensures that information flows smoothly

among the agents. Outputs of agents will be passed to the

Co-ordinator firstly and will be distributed to other agents

by the Co-ordinator.

In this section, we have described a number of intelligent

agents based on B2B eCommerce marketplace workflow

requirements. In Section 5, we will demonstrate how these

intelligent agents enhance the B2B eCommerce workflow

collaboratively.

5. An example

In order to illustrate our concepts and techniques, we will

demonstrate a workflow taken place in B2B eCommerce

marketplace, which is similar as www.ariba.com, with

multiple agent collaborative support. Our example delivers

an Ariba-like B2B eCommerce platform for managing

eProcurement, supply and exchange-based commerce

process. There are several services in this marketplace,

such as Transaction Routing Service, Dynamic Trade

Service, Online Payment Service and Logistics Service,

etc. (shown in Fig. 4). We will focus on how the multi-

agents collaborative support such B2B eCommerce work-

flow process, and how to enhance the risk-monitoring tasks.

When a transaction starts in our example, assume that the

Workflow-Monitoring Agent finds a problem. For example,

it finds an unrecognized delay of logistics and online

payment during the processing (step 1) and reports to the

Co-ordinator (step 2). The Co-ordinator asks both External

and Internal Information Agents to gather related infor-

mation (step 3(a) and (b)), and sends the searching results to

the Diagnostic Agent to find the reason of the problem (step

4). In order to reason the problem, the Customer Resource

Agent may be required to supply the information about the

relationship among the users and divisions involved in this

particular transaction to see if any changes may cause the

problem (step 5). Then, the Diagnostic Agent reanalyzes the

information and reports the diagnostic result to the Co-

ordinator eventually (step 6). The Co-ordinator asks the

Negotiating Agent for the solution based on the Diagnostic

Fig. 3. The intelligent agent hierarchy.

D. Xu, H. Wang / Knowledge-Based Systems 15 (2002) 485–491 489

Agent’s results (step 7). The Negotiating Agent will discuss

with both Buyer HIC Agent and Supplier HIC Agent for a

compromised deal (step 8(a) and (b)) and send the deal back

to the Co-ordinator (step 9). Finally, the Co-ordinator will

distribute the deal to relevant parties for necessary actions

and changes (step 10(a)–(c)).

From this example, it is clear that intelligent agent’s

functionality, such as reactivity, and pro-activity are very

useful for solving B2B workflow problems. Multiple agents

perceive the B2B eCommerce marketplace environment

and respond in a timely fashion to changes that occur.

In this example, multi-agents do not simply act in

response to their environment; they are able to exhibit goal-

directed behaviors by taking the initiative. Intelligent agents

are able not only to reactive the workflow environment

change, but also predict the change caused by external

environment. Assume that a supplier has two B2B

transactions with same goods in different B2B eCommerce

platforms. Assume that the supplier’s inventory is not

enough for two buyers. In this case, the External

Information Agent can find the supplier’s inventory control

problem from other WWW sources, and reports the findings

to the Co-ordinator. The Co-ordinator will try to confirm the

information’s accuracy by asking the Internal Information

Agent, who will check if there is any same circumvent

happened before, and by asking Customer Resource Agent,

which will check the profile of supplier and the relationship

between the supplier and buyer. After such information

gathering, all the related information will be sent to the

Diagnostic Agent for further diagnosis. The Diagnostic

Agent will do deep diagnosis and find solutions to prevent a

potential problem before the problem occurs. Such solutions

may involve dynamic changes of relevant workflow

processes.

6. Conclusions

Workflow management in the B2B eCommerce Market-

places has gained great attention. Workflows are the

structured activities or tasks that take place for a transaction

processing in the B2B eCommerce marketplace. These

activities frequently involve several database systems, user

interfaces, and application programs. Unfortunately, most

current WFMSs lack the functionality to support the

dynamic nature of automating B2B transaction processes,

such as information sharing, chained execution, and process

changing.

Multi-agent collaboration for workflow monitoring is

supplement to existing workflow management techniques

and has a number of advanced functionalities, such as

flexibility and prediction. The major contribution in this

paper is that we have designed a novel architecture for the

workflow monitoring of the B2B eCommerce transaction. A

number of different kinds of agents have been defined. Such

different agents can collaborate together for the workflow-

monitoring tasks. Through agent collaboration, when a

problem occurs, our system is able to predict its possible

consequences and try to reorganize workflow processes to

reduce possible harms. Further more, through agent

collaboration, a workflow process can receive relevant

information from other processes or outside.

By applying multiple agent collaboration for B2B

eCommerce marketplace, our system has the following

novel features:

† Storing organizational and individual knowledge and

giving guidance timely.

† Achieving individual and shared goals of the

participants.

Fig. 4. Multi-agent collaboration.

D. Xu, H. Wang / Knowledge-Based Systems 15 (2002) 485–491490

† Collecting and monitoring relevant information

automatically.

† Making necessary workflow process changes during

execution.

Acknowledgments

This research is supported by research grants (7001141

and 7001014) from the City University of Hong Kong.

References

[1] S. Newell, H. Scarbrough, Intranets and knowledge management:

complex process and ironic outcomes, Proceedings of the 32nd

Annual Hawaii International Conference on Systems Sciences 5–8

January (1999) 10.

[2] J. Liebowitz, Knowledge management and its link to artificial

intelligence, Expert Systems with Applications 20 (2001) 1–6.

[3] H.K. Bhargava, W.C. Branley Jr., Simulating belief systems of

autonomous agents, Decision Support Systems 14 (1995) 329–348.

[4] C. Bussler, Enterprise-wide workflow management, IEEE Concur-

rency July–September (1999) 32–43.

[5] O. Etzioni, D. Weld, A Softbot-based interface to internet,

Communications of the ACM 37 (7) (1994) 72–76.

[6] C.M. Khoong, Decision support system: an extended research agenda,

Omega 23 (2) (1995) 221–229.

[7] M. Klein, Coordination science: challenges and directions, Coordi-

nation Technology for Collaborative Applications (1998) 161–176.

[8] C.M. Pancerella, N.M. Berry, Adding intelligent agents to existing EI

frameworks, IEEE Manufacturing September/October (1999) 60–61.

[9] M. Hubns, M. Singh, Workflow agents, IEEE Computing July–

August (1998) 94–96.

[10] K. Takeda, M. Inaba, K. Sugiara, User interface and agent prototyping

for flexible working, IEEE Multimedia 3 (1996) 40–50.

[11] M. Turoff, S.R. Hiltz, M. Bieber, J. Fjermestad, A. Rana,

Collaborative discourse structures in computer mediated group

communications, Proceedings of the 32nd Annual Hawaii Inter-

national Conference on Systems Sciences 5–8 January (1999) 9.

[12] H. Wang, Intelligent agent assisted decision support systems:

integration of knowledge discovery, knowledge analysis, and group

decision support, Expert Systems with Applications 12 (3) (1997)

323–335.

[13] M. Wooldridge, N. Jennings, Intelligent agents: theory and practice,

The Knowledge Engineering Review 10 (2) (1995) 115–152.

[14] M. Luck, From definition to deployment: what next for agent-based

systems? The Knowledge Engineering Review 14 (2) (1999)

119–124.

[15] S.E. Lander, Issues in multi-agent design systems, IEEE Expert

March–April (1997) 18–26.

[16] W. van der Aalst, Loosely coupled interorganizationl workflow:

modeling and analyzing workflows crossing organizational bound-

aries, Information and Management 37 (2) (2000) 67–75.

[17] T.-P. Liang, Guest Editor, Introduction to the special issue: intelligent

agent for electronic commerce, International Journal of Electronic

Commerce 4(3) (2000) 3–5.

[18] A. Moukas, G. Zacharia, Agent-mediated electronic commerce: an

MIT media laboratory perspective, International Journal of Electronic

Commerce 4 (3) (2000) 5–23. Spring.

[19] M. Barbuceanu, M.S. Fox, Coordinating multiple agents in the supply

chain, Proceedings of the 5th Workshop on Enabling Technologies

19–21 June (1996) 134–141.

D. Xu, H. Wang / Knowledge-Based Systems 15 (2002) 485–491 491