Getting to Implementation: Towards a System Dynamics
Change Management Framework
B.R. Campbell and G.M. McGrath
Joint Research Centre for Advanced Systems Engineering
Division of Information and Communication Sciences,
Macquarie University,
Sydney,
Australia 2109
Ph: +61 2 9850 9107
Fax: +61 2 9850 9102
Bruce.Campbell @mq.edu.au,
mmegrath @ics.mg.edu.au
The literature tends to suggest that system dynamics has long suffered a low rate of implementation of
recommendations. This paper explores two different (but closely related) modelling exercises within
one large Australian organisation which were initiated by the same executive. Although the problems
being investigated essentially concerned the same area within the organisation, one recommendation
was acted upon whilst it was still in a draft form and the model in an early phase of validation. The
second modelling exercise was eventually abandoned, after considerably more work than the first, due
to increasing resistance to what was perceived as the likely recommendations.
The two modelling exercises are retrospectively reviewed in an attempt to elicit characteristics that
impacted on acceptance or rejection of recommendations. Consistent with previous research, the
composition of the modelling teams (although identical in both cases) was undoubtedly a factor.
However, other issues were also important, including: perceptions of the nature of the problems; the
match between recommendations and both organisation culture and the (deep structure) power source
distribution; the organisation span impacted by the recommendations; and, perhaps most significantly
(in this instance at least), the extent of change required to implement recommendations. We believe
that our findings have substantial external validity and argue that system dynamics practitioners and
researchers should adopt a much more holistic view and endeavour to take advantage of the
substantial body of change management research undertaken outside our own narrow discipline.
Introduction
The work described in this paper is part of a three year collaborative research project
jointly funded by the Australian government and a large Australian high technology
company, Gigante (a pseudonym), which is operating in the telecommunications
industry.
Gigante has adopted a product differentiation strategy (Porter 1985) in the past few
years as a response to increased competition. This strategy has meant that its product
mix is rapidly changing, making it difficult for systems and customer services
representatives (CSR’s) to keep pace. Each time a new product is introduced, or a
minor change made to an existing product, a new product code to identify the product
is generated. There are currently in excess of 40,000 product codes.
Gigante also has separate, and incompatible, systems for service provisioning and
customer billing with a number of front ends used at the user interface. Often,
different products require slightly different procedures to enter data into the
provisioning system. Each of these variations is documented and CSR’s are expected
to know how to enter orders for each product type into the provisioning system.
Incorrect entry of the product code into the provisioning system results in service
records, generated as customers use a service, dropping to an “error bucket”. CSR’s
must correct the initial product code entry before service records in the error bucket
can be re-introduced to the billing system and the customer billed for the service.
Gigante has a policy that excludes charging for services provided more than 12
months earlier. As a result of this policy, and the difficulties encountered in correcting
errors, many service records are deleted from the error bucket before errors are
corrected. This results in loss of revenue to Gigante, estimated to be in the order of
AUD$80,000 per day in 1993 (Booz, Allen & Hamilton 1993).
The problem observed by Gigante is that a new service provisioning process maintains
its integrity for about 3 months and then starts to deteriorate with ever more data entry
errors being made. The objective of this research was to find out why and to
recommend corrective action. The research was to be limited to processes within the
CSC’s as it was felt that it was in this area that most problems were occurring.
Background
The previous report of Booz Allen and Hamilton (1993) had indicated that the root
cause of the unacceptable error rate occurring during data entry within the CSC’s was
lack of training. However, a training regime put in place as a result of this report did
not affect the error rate in the long term and has since been abandoned. Training is
now primarily on-the-job with a number of employees receiving additional training to
become subject matter experts. The latter employees are co-located with other CSR’s
to assist them when needed.
Since the earlier report, Gigante has undergone a number of re-organisations and, like
many other businesses, has downsized to reduce costs, and reduced the number of
levels of management. This has left little opportunity for advancement for CSR’s who
are among the most poorly paid employees of Gigante. In spite of this, the turnover
rate of staff has actually dropped from the 200% reported earlier (Booz, Allen &
Hamilton 1993). This is most likely due to: the less favourable economic conditions
which are makes finding other employment problematic; a staff freeze which has been
in force for some time and prevents staff transferring to other sections within Gigante;
and the introduction of personnel policies that make it more attractive to employees to
remain for a given period before leaving. However, discussions with CSR’s indicated
that morale and motivation is low and that staff are very dissatisfied with the situation.
The above indicates that the situation being investigated fell into the category of a
problem defined by Vennix (1996) as “messy”. These problems are characterised by
complexity, uncertainty, interrelated sub-problems, recursive dependencies and
multiple interpretations of the problem’s essence (Vennix 1992, McGrath et al.
forthcoming). Because of this it was decided to develop a system dynamics model
using a group modelling approach. Three people from the university were involved,
one of whom had extensive knowledge of Gigante, together with another three people
from Gigante. The latter were all from the same section and included the executive
who had originally requested the investigation. Access to other members of Gigante
was available if required, and additional people from Gigante were involved in the
development of the various sections of the model.
During early modelling sessions the causal loop diagram at Figure | showing the
provisioning process within a CSC was developed (Campbell 1998).
na:
* ~~.
Lost Revenue Customer
’ ae ~e
Unbilled
CSR nea “ee ~ Enquiries
~ ee “
Capability of CSC” ee Corrective
Action
Knowledge ae +
Gap ~~ Work Load
- +
Knowledge + Training 4g -—~ x
+ +
— = Hours
ty Target
ie Productivity
New Products + ey
pee
Training
Figure 1. A high level causal loop diagram showing factors affecting the provisioning system
within a customer service centre within Gigante.
The causal loop diagram (CLD) at Fig. 1 indicates the complexity of inter-related
variables within a CSC affecting the provisioning process.
Once the CLD had been accepted by all persons involved in the modelling process a
quantitative SD model was commenced. As will be explained, this was not completed,
but sufficient work was completed to show why the training program instigated at
Gigante had not been successful. The high rate of staff turnover, combined with an
inability to replace staff rapidly, meant that Gigante was continually training new
staff. In effect, as soon as staff were trained, they were walking out of the CSC doors.
The findings were similar to those reported by Senge (1992) when he was discussing
the problems of Hanover Insurance.
A concurrent SD model also explained, in part, the observation that the provisioning
process for a new product maintained its integrity for a number of months then slowly
deteriorated. The model indicated that the elapsed time between product introductions
was critical, as was the availability of sufficient numbers of trained staff. Gigante is
walking a tightrope in both areas.
A Related Problem
Part way through the above modelling exercise our client at Gigante presented us with
a related problem. There was sufficient overlap with the main thrust of the research to
allow much of the existing modelling of the system to be utilised. Most of Gigante’s
information services are provided via electronic voice and other telecommunications
facilities. The organisation’s policy (for a certain class of product) had been to not
charge for connection times less than 6 seconds. Gigante’s accounting department had
analysed this product and found that many calls lasted between 2 and 5 seconds. A
proposal made by a marketing manager was that Gigante start billing for these calls. It
appeared, on the surface, that income could be greatly increased with little cost to
Gigante (McGrath et al. forthcoming).
However, our client, who had been given a copy of the proposal for comment, was
concerned that the proposal could increase billing enquiries and so increase the
workload of CSR’s. This could then have an impact on the existing provisioning
processes as the same CSR’s were responsible for both billing enquiries and
provisioning. This became known colloquially as the “Six Second Problem”.
Outcomes
Considerable effort was spent over some months developing an SD quantitative model
of the provisioning system and in model validation (Barlas 1996; Forrester 1961;
Forrester & Senge 1980). This reached a point where, although the model was not
complete nor completely validated, the modelling team was comfortable with the
behaviour of the model. Simulation of the model indicated that Gigante should
address some of its personnel policies as well as re-thinking its strategy of rapid
product introduction. It became obvious over time, however, that our client was
reluctant to compose a formal report and present it to management. Consequently, no
decision was ever taken regarding the provisioning system problems.
In contrast, the model of the Six Second Problem was constructed in a matter of days
and made some gross assumptions of suspect validity. For example, due to lack of any
other data that could be collected within the time given to comment on the proposal, it
was assumed that the only additional cost was that of having to employ more people
to handle the additional billing enquiries. Sensitivity analysis was used to model
various assumptions on the number of additional enquiries that could have been
expected. This was done as the true effect was not known. The model structure was
never validated. However, the model indicated that although billing for the additional
few seconds could increase income and profits in the short term, a longer term effect
was an overall loss of profits due to the increase in billing enquiries. This could have
been up to USD$18 million over a 12 month period. The model did not take into
account the effect of these additional enquiries on the provisioning system.
Due to the time constraints placed on the Six Second Problem, a very early draft
report that included model output, as well as numerous spelling and grammatical
errors, was given to our client. This was presented to the marketing manager, who had
the final decision, without our knowledge. A decision was made immediately not to
proceed with the proposal.
Evaluation
These two modelling exercises were conducted by the same modelling team, were
carried out within the same functional area of Gigante, and were championed by the
same executive within Gigante. They had entirely different outcomes. It is therefore
useful to evaluate the two problems to identify differences. This is particularly
relevant considering the poor implementation rate of SD modelling exercises that has
been a concern of SD practitioners and reported by other writers (Forrester 1994). For
SD to be effective in the next millennium we must understand what is affecting the
implementation of recommendations and address that problem.
The two modelling exercises previously described (The CSC Problem, and the Six
Second Problem) are compared in Table 1, below.
The CSC Problem
The Six Second Problem
Decision could have impacted a number
of functional areas.
Decision did not impact other functional
areas
Many decision makers as a result of
number of functional areas involved
A single decision maker
High risk to decision makers, as
recommendation was contrary to long
held managerial beliefs and policies
Little risk to decision maker
Problem was conceptually difficult
Problem was conceptually simple
Champion did not have credence with the
decision makers
Champion had credence with the decision
maker
Champion did not have strong vested
interest in the decision. In fact, the
champion would have been at risk
presenting the recommendations
Champion had strong vested interest in
the decision. If the proposal had gone
ahead it would have resulted in a large
increase in the champion’s workload and
that of his staff
A decision would have been to create
change within a number of functional
areas within Gigante
The decision did not create any change
within Gigante
Decision would have been expensive to
implement, although saving money in the
long term.
Decision did not cost anything
Table 1. A comparison of the attributes of the two problems modelled at Gigante.
The SD literature is replete with exhortations to include soft social factors into our
models (see, for example, Forrester 1961). However, as practitioners we seem unable
to consider these factors in our own work, as distinct from our models. Table 1 would
indicate that the majority of differences in attributes between the two problems are
primarily social in origin. They deal with risk to both decision makers and to those
affected by any change which may occur due to the implementation of the
recommendations. It would appear that to improve our implementation rate we must
learn to manage change and risk. Although both of these subjects have attracted
considerable research in other areas, there appears to be little in the SD literature. One
of the most critical activities in any change management initiative is dealing with
politically-motivated resistance (Pfeffer 1981 and 1992). We now address this issue in
more detail.
Change, Resistance, Power and Politics
Perhaps we could do a lot worse than revisit Machiavelli’s highly-perceptive insight
(Machiavelli 1993):
There is nothing more difficult to plan, more doubtful of success, nor more
dangerous to manage than the creation of a new system. For the initiator
has the enmity of all who would profit by the preservation of the old
institution and merely lukewarm defenders in those who would gain by the
new one.
Machiavelli’s observations are as pertinent today as when they were first penned in
1513. Machiavelli’s name is, of course, synonymous with power and politics. New
systems must not only satisfy economic and technical criteria for success but be
politically feasible as well. Much political activity is concerned with the development
and protection of power and new systems change existing sources of power (Markus
1983; Markus and Bjorn-Andersen 1987; Pfeffer 1981 and 1992). In essence, new
systems change the organisation. Changing responsibilities, or more pejoratively,
changing power and influence, is the hard core of an activity with strong political
overtones.
We believe that to successfully implement any new system, a thorough analysis is
needed of where power lies in the organisation and what effect the new systems
environment will have on redistributing that power. If this is done, those who will
gain and lose power can be identified so that support and resistance based on political
considerations alone can be anticipated. Except in the most liberal, fraternal and
egalitarian organisation, resistance must be expected because some who need power
will lose power. As change agents, SD practitioners should consider the results of
power source redistribution analysis when challenged by adverse technical or
economic argument. Such resistance may really derive from a threat to a power base.
Power source redistribution analysis is also essential for organisation restructuring
which often accompanies SD activity.
The aim of power source analysis is to systematically identify change to existing
power in the organisation caused by new systems. By this process, potential sources of
resistance can be recognised early. Depending on the culture of the organisation the
consequences can then be openly dealt with rather than becoming obscured in the mire
of organisational politics.
As exhaustive analysis using manual means is very expensive, a systematic focused
analysis method supported by an automated tool is more desirable. For corporate SD-
based planning exercises, this means developing a model of the implementation
domain and automating this as an advisory expert system. In an earlier paper
(McGrath, Dampney and More 1995), a power source distribution model "MP/L1"
(Model of Power in First-Order Logic) was presented. MP/LI describes power
sources, their distribution and their relationships with organisation parties and
processes. Given details on policy implementation activities, MP/L1 can then be
employed to predict likely areas of resistance resulting from changes to
responsibilities, changes to authorities and challenges to well-established beliefs,
values and organisational rules. Automated as an advisory expert system, MP/L1 has
been used successfully, in the field, to assist an information systems planning team to
implement their strategy.
Political influence on systems activities may appear as discordant to some IS
managers as political influence on the application of economics was to traditionalists
when political economy was first argued in the early 1970s (Galbraith 1972). Bowman
and Asch (1987), however, have argued that no clear distinction should be made
between rational and irrational strategic planning decisions. Decisions should be
assessed only as more rational or less rational. Moreover, the degree of uncertainty is a
major determinant of decision rationality and uncertainty leads to political activity
(Pfeffer 1981). Decision making in SD work is made uncertain by the flimsy scientific
base on which the immature discipline rests. Power and politics is therefore inevitable
in SD work. MP/LI is a strategic management tool that helps manage the total SD
process by recognising likely resistance to change caused by threats to power sources.
Conclusion
The foregoing presented our experiences with two SD modelling exercises that
involved the same people, but which had very different results. This presented a
situation that was ideal for a comparative evaluation. The evaluation indicated that
risk, change management and power politics may have a substantial influence on the
implementation rate of SD modelling recommendations. The remainder of the paper
introduced the concept of change management and power politics, as well as briefly
describing an automated tool which has been used successfully in the past to identify
resistance to change. However, at this stage we have not specifically incorporated
change management practices within an SD modelling exercise even though this
seems inevitable if we wish to improve our SD implementation rates.
Research is required to establish the effectiveness, or otherwise, of incorporating
power political models, and change management practices, within an SD modelling
exercise. SD is used in numerous ways. It may be used by an employee investigating
a problem within his own organisation. It is also used by consultants who have no
other dealings with their clients. However, it is felt that even in the latter situation
inclusion of power political models and change management practices is still feasible.
At the very least, consultants should be making their clients aware of the effect of
these issues on any recommendations that may be made.
Ideally, identification of resistance to change, and recommendations on likely courses
of action, should be an integral part of the SD modelling exercise. If SD is to flourish
in the next millennium we must address its historically poor implementation rate. This
paper has suggested a way to improve this.
Bibliography
Barlas Y. (1996). Formal aspects of model validity and validation in system dynamics
in System Dynamics Review, Vol. 12, No. 3, (Fall 1996), John Wiley & Sons Ltd., pp
183 - 210
Booz Allen & Hamilton (25 May 1993). Building a World Class Billing System in
Gigante. An internal report by consultants to management of Gigante
Bowman, C. and Asch, D. (1987). Strategic Management. Macmillan: London
Campbell B.R. (1998). Process Failure in a Rapidly Changing High-Tech
Organisation: A System Dynamics View in Tok Wang Ling, Sudha Ram and Mong Li
Lee (eds) Conceptual Modeling - ER’98: Proceedings 17th International Conference
on Conceptual Modeling, Singapore, November, 1998, Springer-Verlag Lecture Notes
in Computer Science series 1507. pp 291-301
Forrester J.W. (1961). Industrial Dynamics. M.LT. Press and John Wiley & Sons,
New York.
Forrester J.W. (1994). System dynamics, systems thinking, and soft OR. System
Dynamics Review, Vol. 10, Nos. 2-3, John Wiley & Sons
Forrester J.W. and Senge P.M. (1980). Tests for Building Confidence in System
Dynamics Models. In Legasto A.A., Forrester JW. and Lyneis J.M. (eds), System
Dynamics, TIMS Studies in Management Sciences, Vol. 14, North-Holland
Publishing, New York, pp 209-228
Galbraith, J.K. (1972). The New Industrial State. Pelican Books, Baltimore, Maryland
McGrath, G.M., Dampney, C.N.G. and More, E. (1995). MP/L1: An Automated
Model of Organisational Power and its Application as a Conflict Prediction Aid in
Information Systems Strategy Implementation. Proceedings of the 11th IEEE
Conference on Artificial Intelligence for Applications. Los Angeles, February 20-22,
1995, pp 56-64
McGrath G.M., Campbell B.R., Offen R.J. & More E. (forthcoming) Intra-
Organisational Collaboration in a Complex, Rapidly-Changing Information Services
Company: A Field Study. To be presented at The Fifth International Conference of
The International Society for Decision Support Systems. Monash University,
Melbourne, Australia, July 1999
Machiavelli, N. (1993). The Prince. Wordsworth Reference, Hertfordshire, UK
Markus, M.L. (1983). Power, Politics and MIS Implementation. Communications of
the ACM, Vol. 26, pp 430-444
Markus, M.L. and Bjorn-Andersen, N. (1987). Power over Users: Its Exercise by
System Professionals. Communications of the ACM, Vol. 30, 498-504
Pfeffer, J. (1981). Power in Organizations. Pitman Pub. Inc.: Marshfield,
Massachusetts
Pfeffer, J. (1992). Managing with Power: Politics and Influence in Organisations.
Harvard Business School Press, Boston, Massachusetts
Porter M.E. (1980). Competitive Strategy: Techniques for Analysing Industries and
Competitors, Free Press, New York
Senge P.M. (1992). The Fifth Discipline: The Art & Practice of The Learning
Organization. Random House, Australia
Vennix J.A.M. (1996). Group Model Building: Facilitating Team Learning Using
System Dynamics. John Wiley & Sons Ltd, Chichester, England
