System Dynamics '95 — Volume II
Quantifying the Soft Issues:
A Case Study In the Banking Industry
Henk Akkermans
Eindhoven University of Technology
Graduate School of Industrial Engineering and Management Science
PO. Box 513, 5600 MB Eindhoven,
The Netherlands
Abstract
Quantification of causal models that contain many so-called "soft" variables is often
problematic because so few "hard" data are available to calibrate the model. This paper
describes a case study in which different techniques were used to quantify a causal
model that contained a number of such soft variables, such as "level of expected
customer irritation", or "effort required to reach branch office". The case study itself
concerned the development of a decision-support system to assess branch office
viability for a medium-sized bank. The specific techniques used for quantification are
part of the standard "tool set" of the Participative Business Modelling (PBM) Method,
the synergistic blend of system dynamics and group knowledge elicitation techniques
developed by the author in a series of six case studies, of which this was the fifth.
Introduction
Modern commercial banks possess a wealth of data regarding the behaviour of their
clientele. Marketing managers of industrial firms would give anything to know in such
detail when, where and how their customers spend their money, and when and where
from these customers receive their money. Thanks to advances in information
technology, banks can dig up these data with relative ease from their own accounting
systems, which may well be described as "digital gold mines".
These computerised data can be of great value in strategic decision-making.
And, nowadays, European banks have quite some strategic decisions to make. After
decades of relative tranquillity and prosperity, the banking industry finds itself now in a
highly competitive market with rapid decreasing customer loyalty on the one hand, and
a broad supply of promising innovative technology on the other hand.
Unfortunately, as it turns out, these computerised data are not enough for
strategic decisions. Useful as they may be, they provide information on only a part of
most strategic issues, for almost every strategic decision consists not only of so-called
"hard" issues, where these computerised data can tell a great deal about, but also of
many "soft" issues, on which they provide little direct information. Examples of such
soft issues are intangibles such as “perceived customer effort for reaching branch
office", "customer irritation", "attractiveness of the office. building" and so on.
Moreover, these soft issues are often of more crucial importance than the hard data:
"while hard data may inform the intellect, it is largely soft data that generate wisdom"
313
Parallel Program
(Mintzberg 1994). This is what managers in the banking industry complain about when
they talk about inadequate decision-support tools: they receive loads of data on some
of the hard issues, but they have nothing to confirm or challenge their managerial
intuition with regarding the soft issues.
Clearly approaches are needed that can address and integrate both the hard and
the soft aspects of strategic decisions. This paper describes a case study in which such
an approach, based upon system dynamics modelling and group knowledge elicitation
techniques, was applied successfully in the development of a decision-support system
for a European commercial bank. Special attention is given to practical guidelines and
examples of how to translate a complex, "soft" issue into a quantitative decision-
support system.
The Client Company and Its Problem
The client company was a medium-sized European bank. Its management structure
was strongly decentralised, with much autonomy for management teams at local
branch offices. At the time of this project, the company was going through an
extensive streamlining operation, in which the local branches were being examined by a
project team of internal consultants for possible cost-cutting opportunities. It was often
found that some of the smaller neighbourhood offices of the local branch were loss-
making: too few clients utilised them for too few services. Would it not be better to
close such offices and refer the customers to another office in the vicinity? Analysis of
the available hard data could easily show the direct savings of doing so, but often the
local bank managers would object to such a financially oriented conclusion. Their main
worry was: "The direct savings are fine, but how will our customers react to this
closure?" Here the available information was less helpful. Although there were
extensive data on such facts as wealth distribution and service usage patterns for
different consumer categories in different areas, these still said little about how people
would actually react. Nevertheless, the local managers had to make a decision taking
into account ail aspects, both soft and hard. In the past, in the absence of better
information discussions on these issues had dragged on for years, but under the present
cost-cutting program swifter decision-making was required.
The Participative Business Modelling Method
The modelling approach used in this project is called "Participative Business
Modelling" or "PBM" (Akkermans 1995). In PBM, a group of people facing a
strategic problem develops a model of that problem in a series of group model-building
sessions, facilitated by one or more experienced modellers / process facilitators.
Modelling in PBM moves gradually from very informal, qualitative and conceptual
models to more and more formal, quantitative simulation models. PBM contains
techniques and guidelines for this whole modelling process. Probably the easiest way of
discussing these is by looking at the PBM method as being articulated on four different
levels:
Level 1: PBM consultant attitude
Fundamental to any PBM project is the attitude the consultant must bring to work with
PBM successfully. Crucial as the 'right' attitude is to successful conduct of any PBM
314
System Dynamics '95 — Volume IT
project, this is, unfortunately, something that can only be learned experientially, not
from a textbook. The three essential aspects of the 'right' attitude are:
¢ professionalism;
* process consulting; and
¢ systems thinking.
These three aspects are interrelated in the following manner. In any serious
management consulting — or business modelling — project, one needs a professional
attitude. Within management consulting, different kinds or styles of consulting are
distinguished. One such style is "process consulting", which is often contrasted with
so-called "expert consulting". All consultants who conduct process consulting projects
should display what Edgar Schein, the developer of the concept, calls "a helping
perspective" (Schein 1969). Finally, PBM can be seen as one of a small number of
management consulting approaches in which, in addition to a process consulting
attitude, systems thinking, and in particular the system dynamics methodology, is
considered very important.
Level 2: The PBM Tool set
Armed with the proper attitude, the PBM consultant is ready to apply the various
techniques that are in the PBM tool set, i.e. the set of individual techniques that are
employed in PBM projects, which are summarised in Figure 1.
Project Start PROJECT DEFINITION
Cognitive Mapping
Hexagon Brainstorming
PROBLEM CONCEPTUALISATION
Workbooks
Causal Diagramming
Stocks-and-Flows Diagramming
Reference behaviours and Archetypes
Preliminary Models
(Likert-scale) Propositions
MODEL FORMALISATION
Pareto-Analysis
Graphical Functions
System Dynamics Simulation
Discrete-Event Simulation
Sensitivity Analysis
Time Series Validation
KNOWLEDGE DISSEMINATION
Control Panels
Microworlds
‘ Learning-Wheel Workshops
Project Finish
Figure 1: PBM techniques by project phase
These techniques themselves are mostly well described in the literature (e.g.
Richardson and Pugh 1981, High Performance Systems 1994, Morecroft and Sterman
315
Parallel Program
1994), Here we will focus on techniques that are used specifically to quantify — soft —
conceptual models, the main theme of this paper.
Level 3: Generic PBM project design
How these are combined into a generic project design is explained at the third level.
This generic design described "the average PBM project", knowing well that it does
not exist. A generic PBM project consists of four stages, which were already identified
in Figure 1.
1, Inthe problem definition phase the consultant finds out, together with the group of
participants, what this problem is really about and how the rest of the project is
best executed.
2. In the problem conceptualisation phase, a conceptual, i.e. non-quantified model is
developed of the problem (in this case, assessing branch office viability in banking),
using a number of different graphical modelling and knowledge elicitation
techniques from the PBM tool set.
3. In the model formalisation phase this conceptual model is translated into a
quantified simulation model with which various policy experiments are conducted.
4. In the knowledge dissemination phase findings from the three previous phases are
disseminated over a wider part of the client organisation.
Level 4: PBM Project Design Guidelines
At the highest level we find various kinds of PBM design guidelines, which aim to give
advice on how one deals with various contingencies, and what sorts of trade-offs one
has to make when tailoring the method to a specific problem and a specific
organisation. These are out of the scope of the current paper, but are discussed at
length in Akkermans (1995).
Project Synopsis
This project was the fifth in a series of six case studies in which the PBM method was
gradually refined (Akkermans 1993, Akkermans, Vennix and Rouwette 1993,
Akkermans 1994, Akkermans and Bosker 1994). This fifth case was far larger and
more ambitious than the four preceding projects. Therefore, it made sense to cut it up
into three separate phases, the end of each phase being marked by a steering group
meeting which took a 'go no-go' decision for the next phase
In the first phase, a qualitative conceptual model was developed in a number of
structured workshops with internal experts in the content manner at hand and two
local bank managers. The resulting model contained all the main factors and
relationships the participants had defined as relevant to the issue.
In the second phase, the conceptual model was quantified in a further series of
workshops, and was applied, refined and validated in two actual decision-making
processes by local branch management teams.
In the third phase, this refined model was then embedded in a more user-
friendly decision-support system (DSS). Also, a structured policy workshop format
was developed to lead local management teams through the various steps in the DSS.
And finally, the internal consulting group was trained in the use of the DSS and the
policy workshop.
As will become clear from the following, this project was a veritable showcase
of PBM techniques. Not only was every technique from the toolset used (this is in
316
System Dynamics '95 — Volume IT
itself not surprising, bearing in mind the size of the project, which was three times that
of most of the preceding projects), but also. the techniques worked well up to very
well. Evidently the PBM method had matured by the time this project was started; the
lessons from the previous cases were learnt well.
Quantifying the Soft Issues
In this paper, special attention is given to techniques by which many of the soft issues
that were relevant to this problem were captured in a quantified decision-support
system. In fact, there were at least five different techniques that were used for this
purpose:
1. Mapping soft issues in causal diagrams
In the conceptual modelling phase, all the factors and relations that appeared to be
relevant from the interviews and group sessions were mapped down in causal
diagrams. Figure 2 gives an example of such a diagram for the part of the model that
dealt with the effects of branch office closure on the effort required from customers to
reach the next nearest branch office, that is, on office accessibility for customer groups.
customer_eff
customer_effort_new_branch
jproduct_loss_including_competition
number_of_products_per_type
Profit_per_product_type
Loss_per_product_type_
Profitability loss_due_to_increase_in_efort
Figure 2: Causal diagram showing effects on profitability of increased customer effort after closure
This diagram can be read as follows: the bigger the difference (3) between the effort required to reach
the old (1) and the new branch office (2) (increase_in_customer_effort), the higher the losses of|
products — such as savings accounts, or insurances — due to increased customer effort (4) . The better
the accessibility of the nearest branch office of a competitor (5: customer_effort_competitor_branch),
the higher these losses will be (6), Not every product type is equally sensitive to such a decrease in
accessibility (7), For instance, very few people will cancel their mortgage loan because their local
branch office is closed down, but many parents will switch the savings accounts of their children if
these haye to walk a long way to the bank. In this way, losses per product type can be determined (8),
which can be multiplied by the number of each product type sold in this area (9) and the profitability
per product type (10). In this way, an estimate of overall profitability loss due to increased customer
effort can be obtained (11).
317
Parallel Program
2. Converting soft relations into scales
In fact, the diagram shown in Figure 3 appeared in the final version for the. model.
Earlier diagrams included many more factors, such as the location within a shopping
centre, or the quality of the shops nearby, the availability of parking space, the
proximity of a large road, and the age and wealth distribution of the local population.
But, gradually, the team worked to incorporate all such considerations into a small
number of five-point scales, such as the one shown in Exhibit 1 for levels of potential
customer irritation, another crucial element in the estimation of expected profitability
losses after office closure (A similar five-point scale was developed for branch office
accessibility, variables 1,2 and 5 from Figure 2.)
1. Minimal irritation. No reaction on closure. Atmosphere of silent agreement.
2. Modest irritation: Customer irritation is voiced by clients complaining at the counter. Verbal
reactions, which do not result in actions.
3. Normal irritation: Considerable number of verbal complaints from customers. Some local bank
managers are approached individually by customers.
4. High irritation: Great number of individual complaints, voiced also in local newspapers. Local
shareholders start asking questions at local board meetings. This level results in actions, in the
preceding three irritation remained verbal.
5. Very high irritation: The main difference with high irritation is that here organised group actions
take place. A great deal of publicity, organised opposition in board meetings and other political
activities.
Exhibit 1: A five-point scale of levels of client irritation in response to branch office closure
The definitions of these five-point scales were once again very much a group activity,
taking place in workshops, with the bank experts jointly agreeing on adequate
formulations of each subsequent level.
3. Developing graphical functions
When the variables in the conceptual model have been given quantified ranges in this
manner, it becomes possible to create graphical functions for the relations between
variables. For instance, Figure 3 shows the graphical function for the relation
"customer_irritation > lost sales_of_product_type_1", with customer_irritation having
possible values between 1 and 5, as was shown in Exhibit 1.
This particular graph — and several others — were derived once again in the group
modelling workshops that have such a central place in PBM projects. Constructing
graphical functions for such soft relations in a group session with six experts was not
always easy, but, in the end, full consensus could be reached.
Procedures for arriving at a graphical function — even for a relation as soft as
this one — have been described elsewhere (High Performance Systems 1994,
Akkermans 1995), but the main idea is that:
« One starts by assigning ranges to the X and Y axis for the two variables under
study. If these are hard to determine, one can take a range between 0 and 1, with 0
being "minimal" and 1 being "maximal" (or, as in this case, between | and 5).
« Then one determines the "average" situation: If X is currently at this point, then
what is Y? If it is hard to denote a single value, one draws a hi-low bar to indicate
the plausible range of Y values.
318
System Dynamics '95 — Volume II
30
% lost sales of service type 1
a
1 2 3 4 5
customer irritation
Figure 3: A graphical function for the "soft" relation of the effect of customer irritation on profitability
« Next one looks at extreme values of X: X=0 and X=maximal and determines
corresponding values of Y in a similar manner.
« One repeats this procedure for various intermediate points,
« At this point it is useful to consider whether there is any reason to suppose the
relationship would have a particular general shape. Is it linear or non-linear? In this
case, an S-curve was judged as most plausible: at low levels of irritation, minimal
losses will occur; then there is a level of irritation at which most of the "defections"
to competitors will take place. But in the end, most customers will not go so far as
to actually take their deposits or debts away from the bank, regardless of how
irritated they get (because doing so costs them money, of course).
¢ Armed with this knowledge, one now tries to draw a line through more or less the
mid points of the scatter bars of Y values that have been determined for the various
values of X. This becomes the graphical function.
4. Constructing control panels
In this manner, the team was able to develop a runnable simulation model in a system
dynamics language, which was used and calibrated in two internal consulting projects
where actual decisions regarding branch offices had to be made. Nevertheless, this
simulation model still needed a person well-trained in the use of the specific modelling
methodology and software, whereas the model would have to be used by internal
consultants, not by modelling experts.
Therefore, the simulation model was embedded in a user-friendly software
shell. In this shell, interaction with the model took place via so-called "control panels"
(High Performance Systems 1994, Akkermans 1995). Figure 4 show what became of
the causal diagram of Figure 2 in this shell.
319
Parallel Program
ie See nang Mee en ee
i: sa
Aangestoten Bank: :
ie
Pee lates
2 o
i ae = se
‘s
henting Bi late
= | 23 a4
qs
ae = :
os
- eee o
ES
e ot oe
a i
S Large increase customer effort | 2} Edit = Print ‘OK
= itis fy af So nee
omens = neers! eee ee
Figure 4:_A control panel to quantify effects of increased customer effort on profitability (cf, Figure 2)
What remains from the causal diagrams are the variable names and the connections
between them. Added are visualisations of the values used in the current scenario. All
values, also those of the graphical functions, can be changed by the users by mouse-
clicking on the relevant part of the screen.
5. Conducting learning-wheel workshops
The internal consultants were trained in the use of this software and the associated
manual. Also, they were trained in how to conduct a so-called "learning-wheel
workshop" (Byrne and Davis 1991, Akkermans 1994) with the aid of the software. In
a learning-wheel workshop, values for key parameters are systematically changed.
After each change, participants are asked for their expectations of resulting system
behaviour. Then the model is run with the new values. If the model outcome is
identical to the predicted outcome, then so much for the better: management intuition
is confirmed and confidence in the model is increased. If the two differ, the group
investigates the structure of the model to find out why behaviour was different: an
opportunity for management learning occurs. Next the wheel recommences, a new
variable is changed, estimates are made and the process is repeated.
This also happened in the learning-wheel workshops in this case. Figure 5
shows the different scenarios that were run every time the model was used with
management.
320
System Dynamics '95 — Volume II
MOST LIKELY SCENARIOS ALTERNATIVE:
SCENARIOS,
SCENARIO 0: direct savings
SCENARIO 1; Customer effort .. SCENARIO 2: Higher customer
old and new te Git new
SCENARIO 3: Higher sensitivity.’
(SCENARIO 4: customer effort “to effort for certain services.
competitors) Ce ea
SCENARIO 5; Expected customer |e SCENARIO 6: Higher irritation
irritation Se GEC Ron,
SCENARIO 8: Higher sensitivity
to irritation of certain'services
SCENARIO 9: expected oe Aes ae Bs :
dampening from new outlook |" (SCENARIO 10-Less dampening
(SCENARIO 12: Broadening || (SCENARIO ° Demographic
and deepening of customer base) “ developments)
Figure 5: The structure of the learning-wheel workshop in this case
Here all the scenarios shown in the right hand column are basically sensitivity analyses. For each of |
the values determined for the scenarios in the left-hand column, the "base values" one or two more
pessimistic values were tested. For instance, in Scenario 5 the expected level of customer irritation is
determined, but in Scenarios 6 and 7 the effects of higher levels of irritation are discussed as well.
This way of interacting with the model turned out to be quite successful; not only did it
structure discussions with management considerable, it also proved to be very
reassuring for bank management: whenever there were critical assumptions in the
model they knew where to find them, whilst they also knew which assumptions
appeared not to be critical.
Project Results
This project was a very successful one in at least two respects. Firstly and most
importantly, it was deemed successful from the perspective of the participants: post-
project interviews have pictured the process as effective and a highly instructive
experience for all involved. Despite the presence of the external consultants, there was
a strong feeling of ownership of the final result. Despite the inherent 'softness' of many
of the issues, there was confidence in the quality of the model, as became apparent
from the case evaluation process that was carried out for this project and which was
primarily based on in-depth analysis of the evaluation interviews with participants (See
Akkermans (1995) for a more detailed description of the case evaluation process,
which leans heavily upon the qualitative research framework developed by Miles and
Huberman 1984).
Secondly, the project has also been a success from a business and
implementation perspective: the DSS and the policy workshop format have
subsequently been applied to several dozens of decisions by different local management
teams at the client company.
321
Parallel Program
Conclusion
This paper has tried to show that it is always possible to quantify a causal model, even
for the softest of issues, and that such models can be put to good use in practice (cf.
Sterman 1991, Forrester 1994). Does that mean that one should always quantify? No,
certainly not. There are at least two good reasons for confining oneself to the
development of a good, qualitative, conceptual model:
1. Quantification represents a considerable investment of time, so it has to be worth
the effort. In other words, if conceptual modelling delivers adequate answers to
the problem the client was originally facing, further modelling will not yield
additional value.
2. Quantification may appear artificial if one is modelling a very soft issue. In many
cases — unlike the company described in this paper — clients will not expect a
quantified model for very soft issues, in contrast to the expectations for a very
‘hard’ problem. However, this does not mean that model builders / consultants
should not stand up for their own convictions; if one feels that the group may still
be missing an essential dynamic insight, one should of course try to get the group
into the model formalisation stage, however difficult that may be.
Literature references
Akkermans, H.A. (1993) “Participative Business Modelling to Support Strategic Decision Making in
Operations - A Case Study", international Journal of Operations & Production Management Vol.
13, No. 10, pp. 34-48.
Akkermans, H.A., Vennix, J.A.M., Rouwette, E. (1993) "Participative Modelling to Facilitate
Organisational Change: A Case Study”. In: E. Zepeda and J.A.D. Machuca (eds.) : Proceedings
System Dynamics '93 Cancun Mexico, pp.1-10.
Akkermans, H.A. (1994) "Developing A Logistics Strategy Through Participative Business
Modelling". In: Platts, K.W., Gregory, M.J. and Neely, A.D. (eds.) : Operations Strategy and
Performance; Proceedings 1st EOMA Conference. University of Cambridge, Cambridge, pp. 137-
144,
Akkermans, H.A. and Bosker, J. (1994) "Design Guidelines for Participative Business Modelling
Projects; Lessons from an Unsuccessful Case Study". In: Proceedings 1995 International System
Dynamics Conference. Problem Solving Methodologies, pp. 15-24.
Akkermans, H.A. (1995) Modelling With Managers; Participative Business Modelling For Effective
Strategic Decision-Making. Doctoral Dissertation, Eindhoven University of Technology.
Forrester, J.W. (1994) "Policies, Decisions, and Information Sources for Modeling". In: Morecroft and
Sterman (1994), pp. 51-84.
High Performance Systems (1994) Introduction to Systems Thinking and Ithink. High Performance
Systems, Hanover NH.
Miles, M., Huberman, A.M. (1984) Qualitative Data Analysis. A Sourcebook of New Methods. Sage,
London.
Mintzberg, H. (1994) The Rise and Fall of Strategic Planning. Prentice Hall, New York.
Morecroft, JD.W. and Sterman, J.D. (eds.) (1994) Modelling for Learning Organisations,
Productivity Press, Portland.
Richardson, G.P., Pugh, A.L. (1981) Introduction to System Dynamics Modeling with DYNAMO. MIT
Press, Cambridge MA.
Schein, E.H. (1969) Process Consultation: Its Role in Organization Development. Addison Wesley,
Reading Mass,
Sterman, J.D. (1991) "A Sceptic's Guide to Computer Models", in: G.O, Barney, W.B. Kreutzer, M.J.
Garrett (eds.) : Afanaging a Nation. The Microcomputer Software Catalog, Westview Press,
Boulder CO.
322
CC BY-NC-SA 4.0