Using Agent Technology to Overcome Project Failure
in Distributed Organizations
Holly Parsons – Hann
1
, Kecheng Liu
1
1
School of Systems Engineering, Department of Computer Science, Reading University,
Reading, UK
Abstract. As organisations become more global and interest groups more
widely distributed reaching a consensus among stakeholders when conducting a
global spanning project becomes increasingly harder to achieve. Many soft-
ware projects have failed because their requirements were poorly negotiated
among stakeholders and this problem must be solved if project failure is to de-
crease. Although many stakeholder negotiation methods have been suggested,
validated and employed in projects across the globe, project success rates are
still very low, suggesting that there is still work to be done in the distributed
organisational domain to increase the probability of project success. This paper
will highlight the current project management problems in distributed organisa-
tions and suggest a new agent based method of solving them.
1 Introduction
Due to the competitive technological industries prevalent today, it has become com-
mon practice for distributed organisations offering usually similar competencies to
federate alliances to improve their efficiency and move into a more competitive mar-
ket position. In order to remain competitive, businesses must use every technological
innovation to their advantage, thus not being out performed in efficiency or produc-
tivity.
Distributed organisations have the advantage that they are able to changing de-
mands and technological advancement, businesses respond by developing increas-
ingly flexible infrastructures in order to remain competitive in the newly defined
market boundaries [13]. However, due to the vast percentage of project failure it is
increasingly hard to remain in such an advantageous market position when just over
10% of projects are considered ‘successful’ e.g. within time, within budget and deliv-
ers what the user wants.
One of the main reasons cited for project failure are user requirements (CHAOS,
1995), 13.1% of businesses in 1995 stated that this was the main reason why their
projects had failed. This is a common problem in projects and requirements elicited
from customers can overlap or conflict in any project. Some requirements may be
ambiguous or unrealistic, while others may remain undiscovered [11]. Due to these
reasons, requirements need to be specified clearly and concisely without any margin
for ambiguity or confusion.
Parsons Hann H. and Liu K. (2005).
Using Agent Technology to Overcome Project Failure in Distributed Organizations.
In Proceedings of the 1st International Workshop on Requirements Engineering for Information Systems in Digital Economy, pages 13-23
DOI: 10.5220/0001416000130023
Copyright
c
SciTePress
The problem is further exacerbated when projects are performed in a distributed
organisation. This adds a new degree of difficulty to the project before it is even un-
derway as there is a strong likelihood that some of the stakeholders will not be lo-
cated in one geographic site. This means trying to elicit the requirements from nu-
merous different stakeholders from many different locations which can be incredibly
challenging. Project teams cope with distance in a number of different ways by using
a variety of media [6].
In this paper, we suggest a new requirements negotiation and prioritisation
method, one which takes into account the distributed nature of the stakeholders, as
well as the natural ambiguity which is usually present in such projects. By using
agent technology, we will demonstrate that requirements negotiation and prioritisa-
tion between stakeholders can be more efficient as well as less time consuming and
less ambiguous thus, increasing the probability of project success.
This paper will discuss the features of a distributed organisation, note the current
problems with requirements negotiation methods and suggest suitable agent-based
methods to tackle these problems.
2 Features of a Distributed Organisation
Distributed Organisations have emerged due to new enabling technologies and sup-
port copious numbers of e-work, e-business and ecommerce activities all over the
world. They are boundary crossing with complementary core competencies. As well
as sharing knowledge and usually consisting of a network of independent organisa-
tions, there is great geographical dispersion and many changing participants year after
year.
There is no hierarchy in a distributed organisation due to a belief of participant
equality and nearly all communication between participants is conducted electroni-
cally. A distributed organisation is formed by agreement of separate organisations to
collaborate, to share knowledge and expertise, in order to achieve a common purpose.
Although a group of legally separate entities, the organisation acts as though they
were one, thus the customer deals with what appears to be a single organisation. To
ensure that the customer sees only one organisational front, the member organisations
coordinate their activities [9].
By being part of a distributed organisation, many advantages can be reaped. Re-
sources are less tied up as the organisation is not paying enormous amount of money
for building fabrics, therefore has more money to invest in its core values. Changes in
the business environment do not bother a distributed organisation as they are flexible
enough to embrace change and not hinder their productivity, this greater flexibility
ensures a non hierarchical and dynamic business.
However, as well as there being many advantages to conducting business in a dis-
tributed fashion, there are also disadvantages to consider. The disadvantages will be
identified and analysed in the next section.
3 Requirements Negotiation
Requirements negotiation is concerned with the high level statement of requirements
elicited from stakeholders. Negotiation has become an essential part of system speci-
14
UK USA
China Australia
France
Project
Manager
Fig. 1. Distributed Stakeholders around the world
fication where stakeholders negotiate among themselves and with system engi-
neers[14] with tradeoffs happening in order to resolve conflicts.
Many software projects have failed because their requirements were poorly negoti-
ated among stakeholders. Negotiating requirements is one of the first steps in most
Software systems life cycle, yet its results have a significant impact on the system’s
value [2]. Getting project requirements right, is crucial for project planning and ulti-
mately, customer satisfaction as without customer satisfaction the project is not con-
sidered a success. Should the stakeholders not agree on the requirements of the pro-
ject, there is little chance of a successful project as one of the three criteria for project
success will not be present e.g. ‘giving the user what they want’.
Getting a general consensus between stakeholders is a crucial, yet hard task to ful-
fill in a traditional hierarchical company, without having the additional complication
of distributed stakeholders who employ different working cultures and social norms.
In spite of years of experience, many organisations still do not allow enough time
to resolve requirements conflicts. Due to the rapid technological development, many
companies now rely on other mediums of communication which are faster and
cheaper in some ways, yet at a cost. Using E-mail for negotiation takes significantly
longer than face to face meetings yet has the advantage of sending long documents to
various stakeholders in a very short time.
Figure 1 illustrates the problems that are present when trying to negotiate stake-
holder requirements in a distributed organisation. Research has been done into a
number of ‘capability barriers’ which prevent effective communication in geographi-
cally dispersed groups [16] The three identified problems included not sharing a
common first language, being separated by sixteen time zones and the difference in
typing ability when communicating via a messaging program. There are many differ-
ent working cultures in the world, and many stakeholders will experience and be part
of these work defining characteristics.
Although a minor difference in interpretation, these small misunderstandings can
grow and cause problems later in the project lifecycle if a common understanding of
all the project terms are not reached.
15
4 Current Requirement Negotiation Methods
Groupware is technology designed to facilitate the work of groups. This technology
may be used to communicate, cooperate, coordinate, solve problems, compete, or
negotiate. While traditional technologies like the telephone qualify as groupware, the
term is ordinarily used to refer to a specific class of technologies relying on modern
computer networks, such as email, newsgroups, videophones, or chat.
CSCW (Computer-Supported Cooperative Work) refers to the field of study which
examines the design, adoption, and use of groupware. It has also emerged as an iden-
tified research field during the last dozen years. It focuses on the role of the computer
to assist people working together. The development of suitable software for small and
large groups is a major focus of CSCW.
Traditional collaboration is necessarily sequential: one individual can only add
something after another one has finished. This even applies to real-time meetings or
conversations: only one person can talk at a time. In a CSCW environment, on the
other hand, people can add information in parallel. Moreover, the collaboration is not
restricted to real-time or other physical constraints depending on time or space. The
different people collaborating need not be present in the same location or even at the
same time.
In meetings assisted by various groupware technology, people who are assertive,
fluent or in a position of authority will tend to dominate the discussion, while those
who are shy or of a lower rank will find it very difficult to have their ideas accepted
or even paid attention to, however good those ideas may be. This can happen due to
both personality and the main mother tongue of the stakeholder.
The EasyWinWin process helps success-critical stakeholders to jointly discover,
elaborate and negotiate their requirements [1]. The process has been built upon the
The WinWin model [3], which provides a general framework for identifying and
resolving requirement conflicts. Easy WinWin enables and facilitates heterogeneous
stakeholder participation and collaboration.
As Boehm et al. point out, the Win Win models’ primary distinguishing character-
istic is the use of stakeholder win-win relationship as the success criterion and organ-
ising principle for the software and system definition. Our research incorporates vari-
ous specific techniques from the Win Win model, however the use of intelligent agent
technology will help make requirements negotiation a faster and less time consuming
process. Briggs and Gruenbacher [4] acknowledge the potential problem of checking
every requirement by hand by stating ‘Intelligent agents may be able to conduct ex-
haustive pair-wise comparisons amongst thousands of win conditions, a task that
would be impossible for individual humans’
5 Agent Technology
The basic dictionary definition for an agent is one who acts. However, when develop-
ing Information Technology systems this definition is far too vague and therefore
needs to be defined in terms of attributes and properties that an IT agent possesses.
The term ‘agent’ has become increasingly common within the IT industry and is
usually used to describe computational entities such as a ‘multi agent system’. When
referring to agents within the computing industry, agents are defined as
16
‘A computer system capable of flexible autonomous action in a dynamic, unpre-
dictable and open environment.’ [10]
It is imperative to point out than an agent can be a person, a piece of software or a
variety of other things. During the 1980’s, research into the Artificial Intelligence
paradigm was prevalent and various new concepts arose. One of the new branches
was designated as Distributed Artificial Intelligence (DAI) This branch of AI,
strongly linked with social sciences is mainly concerned with the study of multi agent
systems [7].
Sample applications of agent technology to date include data filtering and analysis,
brokering, process monitoring and alarm generation, business process and workflow
control, data/document retrieval and storage management, personal digital assistants
[8], Computer Supported Co-operative working (CSCW) simulation modelling and
gaming also contain various agent technology. These mechanisms are now being
realised by agent technologies, which are already providing copious and sustained
benefits in several business and industry domains.
There are many properties which agents can possess in various combinations. Al-
though there is no industry standard for the definition ‘Agent’ (Object Management,
2000) most experts agree that agents which are bound for IT systems are not useful
unless they exhibit at least three specific attributes. The first attribute is autonomy,
therefore the agent is capable of acting without direct external intervention. Compo-
nents should be enabled so that they can respond dynamically to changing circum-
stances and act based on the agents own experiences. Agents should be able to per-
form the majority of their problem solving tasks without the direct intervention and
they should have a degree of control over their own actions and their own internal
state [5].
The development of intelligent adaptive agents has been rapidly evolving in many
fields of science. Such systems should have the capability of dynamically adapting
their parameters, improve their knowledge-base or method of operation in order to
accomplish a set of tasks.
An information agent, on the other hand, is a computational software entity (an in-
telligent agent) that can access one or multiple, distributed, and heterogeneous infor-
mation sources available. It can pro-actively acquire, mediate, and maintain relevant
information on behalf of its user(s) or other agents preferably just-in-time. The bene-
fit of employing information agents is that they are able to cope with the difficulties
associated with the information overload of the user and output the information in a
detailed and orderly way.
An information agent can communicate with its environment, other agents or hu-
man users depending on who has the information that is needed. If two agents are
communicating with each other, it is vital they can ‘speak’ the same language, there-
fore the use of a commonly agreed agent communication language (ACL) such as
FIPA ACL and KQML has to be considered.
6 Our Research
Assuming the stakeholders have been identified already, the first action to take is to
‘rank’ the stakeholders from rank 1 to rank 4. In previous literature Boehm et al.
state that ‘optimistically expecting a result that is mutually satisfactory to all stake-
17
holders is nearly absurd.’ [12] Therefore by identifying all stakeholders and classify-
ing which are success critical and which are not, various levels of involvement can be
obtained based on which rank a stakeholder has been given.
Therefore the question that must be asked is How can we judge which stakeholders
are the most important and what can we base this judgement on? Sadly, there is no
easy answer and comparatively little research has been done into prioritising the
stakeholders themselves. One method which is widely used however, has been devel-
oped by business strategy theorists Gerry Johnson and Kevan Scholes. It was initially
developed as a tool to help organisations implement strategic developments, but is
equally applicable to project management [17].
Once all the stakeholders have been identified, a matrix can then be plotted which
classify the stakeholders according to interest and influence. Interest and Influence
are defined in the Oxford English Dictionary as:
Figure 2. shows the Matrix , along with an example of the type of stakeholders who
will be in each category. It is now possible to assign each stakeholder a ‘rank’ which
will help the project team to assess which requirements should be given more consid-
eration than others.
Interest: The state of wanting to know about something
or someone
Influence: The capacity to have an effect on the charac-
ter or behaviour of someone or something
18
Influence
High
High
Low
Low
Interest
• End Users
• Community Groups
• Small funders
• General Public
• Large Public sector
funder
• Government
• Directors
• Management
Unlike the WinWin method which identifies and focuses on just the success criti-
cal stakeholders, we believe that even non-critical stakeholders are important. The
stakeholders who are rank 2, e.g. the high influence and low interest stakeholders
may not be success critical, however they must still be given time to discuss any is-
sues they have with the project. If the project changes direction or the individuals
involved in the project change frequently, the group may have an increased interest in
the project.
Therefore it is very important that rank 2 stakeholders are kept as informed as they
wish. Rank 3 stakeholders (low influence and high interest) and Rank 4 stakeholders
(low influence and interest) should also be treated as part of the project, albeit in a
smaller role. Useful information can be learned from these stakeholders therefore
they must not be ignored.
Due to its open methodology DOORS (Dynamic Object Oriented Requirements
System
) can be flexibly used in any industry for virtually any product. It is based on
Fig. 2. The stakeholder classification Matrix
Stakeholder ID:
Agent ID:
Requirement ID:
Heading: [ ]
Short text
[ ]
Long text
[ ]
Attributes
[Validity –
Priority - M
Cost
Created by –
Dependencies]
Fig. 3. Requirements specification for each stakeholder
19
an integrated OO (Object Orientated) data store and can be applied to either a one
main or whole company project. In this method, each requirement is known as a ‘Re-
quirement Object’ hence the Object Orientated perspective. It provides good visuali-
sation of such documents as hierarchies, and its extension language enables a wide
range of supporting tools to be built, and many are provided as menu commands and
examples.
Figure 3. illustrates how, using the DOORS specification framework, we have
adapted it to suit the stakeholders needs when asking them to enter their require-
ments. The stakeholder ID and agent ID are assigned by the project team whereas the
requirement ID is assigned by the agent and is created by combing various parts of
the stakeholder ID and Agent ID into a unique requirements reference.
Throughout the process of entering each requirement, the agent will be able to
monitor each word to check for spelling errors and also be able to detect ‘keywords’
for example, if the stakeholder entered ‘The system must be able to operate 24 hours
a day’ The agent might ask the stakeholder to define the word ‘operate’. All identified
keywords will then be saved in an online glossary which the project team can access
and check for ambiguities between stakeholders, e.g. if ‘operate’ is a culturally de-
pendant word and different stakeholders interpret the word in a different way.
Once all requirements have been entered, spelling checked and keywords high-
lighted. The stakeholder is then asked to prioritise their requirements – 1 being the
most important, 2 the next important and so on, until they all have a priority associ-
ated with them. When all requirements have been entered, the agents will therefore
enter the negotiation process without any assistance from the project team.
An email does not fall neatly into the domain of text classification. There is much
extraneous information that is redundant when classifying the ‘type’ of email it is
(e.g. needed or spam mail) All the methods serve to simplify the problem and reduce
the amount of noise inherent in the problem Applying these methods to each email
message transforms the message into an acceptable input document to a text classi-
fier. Each message can then be viewed as a collection of words which can then be
examined for frequency of occurrence.
The classification problem can be simplified by removing the common pronouns,
verbs and adverbs such as “the”, “it”, “here”. These words can be put into a list called
a ‘stoplist’ and filtered out when trying to classify emails. Since a majority of these
words appear in almost all documents, they can be ignored without losing any infor-
mation. A stoplist will be implemented in the agents in order to reduce the ‘noise’ and
allow focus to be on the less frequent text.
The Porter Stemming Algorithm was developed by Martin Porter in 1980 and is
another methods for refining the text even further. It has been very widely imple-
mented and coded in various programming languages since its publication with re-
search still being conducted into making the algorithm more efficient [18]. ‘Stem-
ming’ has the purpose of treating words the same root as being equivalent. For exam-
ple, see, seen, sees and saw would all be transformed into their root ‘see’, thus elimi-
nating the irrelevant differences between ‘saw’ and ‘seen’.
By using both methods and comparing the similarity of the text left behind it
should be easy for the agents to judge whether the two requirements are the same or
different. If the agents decide they are the same requirement, the agent with the low-
est ranking stakeholder will ‘freeze’ that requirement and not use in the negotiation
process. This means that the requirement will still be prioritised, but no more than
once.
20
Once all duplicate requirements have been identified, agents will then attempt to
negotiate their stakeholder requirements with other agents. Before entering into nego-
tiation with another agent, the agent identifies the stakeholders rank and identifies the
stakeholder multiplication factor.
Stakeholder
rank
Multiple
factor
1 x 1
2 x 3
3 x 5
4 x 7
Table 1. shows the stakeholder ranks 1-4 and also shows a ‘multiplication factor’
This is used to multiply each requirement priority with the corresponding multiple
factor for the correct stakeholder rank. For example, a rank 1 stakeholder would have
their requirement priorities all multiplied by ‘1’,thus the priorities 1,2,3,4 would re-
main the same.
On the other hand, a rank 2 stakeholder would have their requirement priorities
multiplied by ‘3’, thus 1,2,3,4 would in fact be turned into 3,6,9,12 and so forth. Fig-
ure 5 shows what will happen to a rank 2 stakeholders’ prioritised requirements.
The stakeholders will not know the rank they are assigned as this could be seen as
unprofessional and risks embarrassing the stakeholders should they find out they are
Table 1. The stakeholder multiplication
Stakeholder Rank 2: Multiple factor = x 3
Requirements
1. [ <Requirement> ] x1 = 1. [ ]
2. [ <Requirement> ] x1 = 2. [ ]
3. [ <Requirement> ] x1 = 3. [ ]
4. [ <Requirement> ] x1 = 4. [ ]
Fig. 4. Applying the stakeholder multiple factor to a rank 1 stakeholders requirements.
Stakeholder Rank 1: Multiple factor = x 1
Requirements
1. [ ] x3 = 3. [ ]
2. [ ] x3 = 6. [ ]
3. [ ] x3 = 9. [ ]
4. [ ] x3 = 12. [ ]
Fig. 5. Applying the stakeholder multiple factor to a rank 2 stakeholders requirements.
21
rank 4. Therefore the information is undisclosed to all but the project team and the
agents.
It is worth noting that although this method should prove to be quicker and more
efficient than other negotiation methods presented, it is not supposed to be used as a
standalone project technique. As soon as the agents have finished the negotiation and
prioritisation process, we cannot assume that the requirements will remain static and
at the same priority throughout the entire project. Once the project progresses and
stakeholders gain a better understanding of what is needed their requirements will, in
turn mature with the project.
Therefore other methods should be used to compliment the agent negotiation tech-
nique rather than using it in a standalone fashion. When the agents produce a re-
quirements negotiated, prioritised list, the project team will have a clearer idea of
what is needed to fulfil the requirements and a project plan can be created. Stake-
holders can then be contacted and their requirements discussed and refined. This
helps lessen any ambiguity that may still be present and strengthen the project teams’
understanding of each stakeholders’ needs and wants.
Conclusion
By using intelligent agents to perform much of the requirements negotiation and
prioritisation, less effort is needed by either the project team or the stakeholders. The
method addresses the problems commonly found in distributed organisations such as
culture and time zone differences, as well as identifying and hopefully decreasing a
lot of the initial ambiguity of the requirements. Taking into account all stakeholders,
rather than just the success critical ones should enable a richer and more solid founda-
tion for a project to unfold and thus decrease the chance of project failure.
Future work will include refining the agent negotiation protocols as well as start-
ing to build a prototype and actually implement the methods described in this paper.
References
1. Boehm, B., Grünbacher, P., Briggs, B. 2001. Developing Groupware for Requirements
Negotiation: Lessons Learned, IEEE Software, pp. 46-55
2. Boehm, Barry., Egyed, Alexander., 1998. Software Requirements Negotiation: Some Les-
sons Learned, Proceedings of the 20th International Conference on Software Engineering,
Kyoto, Japan
3. Boehm, B., Bose, P., Horowtiz, E., and Lee, M., 1995. Software Requirements Negotiation
and Renegotiation Aids: A Theory- W Based Spiral Approach, In 17th International Con-
ference on Software Engineering, Seattle: IEEE Computer Society Press.
4. Briggs, Robert., O.; Paul Gruenbacher, 2002. Easy WinWin: Managing Complexity in
Requirements Negotiation with GSS, In Proceedings of the 35
th
Hawii International Con-
ference on systems sciences
5. Castelfranchi, C., 1995. Guarantees for autonomy in cognitive agent architecture. In Intelli-
gent Agents - Theories, Architectures, and Languages, (eds. M. Wooldridge and N. R.
Jennings). Springer-Verlag Lecture Notes in AI Volume 890. pp 56-70.
22
6. Damian, Daneila., E., Didar, Zowghi,., 2003. The impact of stakeholders’ geographical
distribution on managing requirements in a multi-site organization, In IEEE Joint Interna-
tional Conference on Requirements Engineering.
7. Filipe, Joaquim, 2000. Normative Organisational modelling using Intelligent Multi-Agent
Systems, Ph.D thesis
8. Graham, Ian., 1998. Requirements Engineering and Rapaid Development – An object Orien-
tated Approach, Addison- Wesley, ACM press, ISBN 0-201-36047-0
9. Lethbridge, N., 2001. An I-Based Taxonomy of Virtual Organisations and the Implications
for Effective Management, http://inform.nu/Articles/Vol4/v4n1p017-024.pdf (Accessed
23/02/2005)
10. Luck, Michael., McBurney, Peter., Preist, Chris., 2003. Agent Technology: Enabling Next
Generation Computing, A Roadmap for Agent-Based Computing, AgentLink II report,
ISBN 0854 327886
11. Maciaszek, Leszek., 2001. Requirements Analysis and System design – developing infor-
mation systems with UML, Addison-Wesley, IBSN 0-201-70944-9
12. Park, Jun-wong.; Daniel Port.; Barry Boehm., 1999. Supporting Distributed Collaborative
Prioritization forWinWin Requirements Capture and Negotiations, In Proceedings of the In-
ternational 3rd World Multiconference on Systemics, Cybernetics and Informatics , Vol. 2,
pp.578-584, IIIS,
13. Parsons-Hann, Holly.; Kecheng Liu., 2005. Measuring Requirements Complexity to In-
crease the Probability of Project Success, In IEEE 7
th
International Conference on Enter-
prise Information Systems.
14. Robinson, W., N., 1990. Negotiation Behaviour During Requirement Specification, In
Proceedings of the . 12th International Conference of. Software Engineering IEEE Com-
puter Soc. Press, Los Alamitos, Calif, pp. 268–276.
15. Standish Group, 1995. CHAOS report.
16. Toomey, L., Smoliar, S., Adams, L., 1998. Trans-Pacific Meetings in a Virtual Space, FX
Palo Alto Labs Technical Reports
17. Vogwell, Deborah.; (2002) Game Theory, Building Magazine
http://www.davislangdon.com/europe_middleeast/uk/images/Game%20Theory.pdf (Ac-
cessed 24/02/2005)
18. Yamout, Fadi.; Rana Demachkieh; Ghalia Hamdam; Reem Sabra; 2004. Further Enhance-
ment to the Porter’s Stemming Algorithm, In Machine Learning and Interaction for Text-
based Information Retrieval
23