C. Lawrence Meador.

A case study of the use of a decision support system online

. (page 1 of 3)
Online LibraryC. Lawrence MeadorA case study of the use of a decision support system → online text (page 1 of 3)
Font size
QR-code for this ebook







it 14



Charles Lawrence _Meador
Sloan School, Massachusetts InTtitute of Technology

David N. Ness
Wharton School, University of Pennsylvania*

674-73 August 1973






Mfl.SS. INST. T'nn.

OCr 18 1S73



Charles Lawrence J^eador
Sloan School, Massachusetts InTtitute of Technology

David N. Ness
Wharton School, University of Pennsylvania*

674-73 August 1973

♦Formerly of the Sloan School, Massachusetts Institute of Technology


OCT 30 1973



Articles which describe the impact of interactive, conversational
computers on high-level management, especially in strategic planning
problem environments, are rare. It is also true that our understanding
of the complex issues related to human information processing, cognitive
behavior, and affective or motivational characteristics (particularly as
they relate to man-machine systems) is at a relatively primitive state
in terms of implementation/design support. This paper explores possible
reasons for, and an approach to solution of certain aspects of this
problem. This approach is operational ized by the construction of an
interactive computer model designed to support management problem solving
in a long range planning environment. Alternative experimental paradigms
are suggested and contrasted for studying human behavioral characteristics
and problems as they relate to management decision system (MDS) or manage-
ment information system (MIS) implementation. Finally, a case study is
presented where information relevant to the issues discussed here was
gathered during the development of a corporate divisional plan by the
corporation's president who used a specific computer planning model
described below.


Conversational computer systems and management science models appear
to have not had the profound impact on management that was predicted in
the early sixties [1]. One reason for this situation is that computer
systems are often designed without a clear knowledge of the problem
environment in which decision support is sought. Another reason is that

designers do not give computer systems much knowledge about the problem
environments within which they seek to support important decisions. Further,
designers often do not compensate for differences between their "style" and
that of the community that the system is designed to serve. This could
account for the lack of impact of computers on relatively "unstructured"
problems in environments such as strategic planning. Perhaps a partial
solution to this predicament may be to get more problem centered knowledge
into computers and thereby make them more "intelligent" within a given
problem context.*

The approach taken in this paper attempts to strike a convenient
compromise between providing support which is so general that it is of
little direct help to the person with the problem, and support which is
so specific that it can only be useful in solving (or operating on) one
particular problem. Thus, the system described below possesses some
(rather general) knowledge of the problem area which is being considered,
but it relies on the particular user to tailor its effect to his own

In developing this system, a substantial amount of attention was
paid to the fact that every new system environment which is proposed to
help or support a manager in a decision environment creates a new problem
for the manager: namely, to learn about the system. For very general
tools this cost is substantial. The manager is required not only to
learn how to understand the tool (a passive understanding) but also how
to use it to operate on his problems (an active understanding). This
"set-up" requires not only effort but also time. Thus the rewards of

See, for example, Minsky, M. and Papert, S., Artificial Intelligence
Progress Report , Artificial Intelligence Memo No. 252, MIT, January 1, 1972.

getting involved in the use of the new system are deferred in time, and
the manager must be willing to risk a substantial investment (in particular
of his time and energy) before any payoff is visible.

With the technology described here, we attempt to avoid this problem
by constraining the manager in some dimensions (which we hope, and will
present some evidence of below, are not terribly material to him) and in
doing so provide him with some prior structure. This allows him to obtain
return for his efforts with relatively little investment (either in terms
of effort or time) and thus substantially limits the risks involved in
deciding to try to use such a support system.


An important though somewhat hazy distinction can be made between

so-called programmed and nonprogrammed decisions:

Decisions are programmed to the extent that they are
repetitive and routine, to the extent that a definite
procedure has been worked out for handling them so that
they don't have to be treated "de novo" each time they
occur ... Decisions are nonprogrammed to the extent
that they are novel, unstructured, and consequential.
There is no cut-and-dried method of handling the problem
because it hasn't arisen before, or because its precise
nature and structure are elusive and complex, or because
it is so important that it deserves a custom-tailored
treatment. By nonprogrammed I mean a response where
the system has no specific procedure to deal with
situations like the one at hand, but must fall back on
whatever general capacity it has for intelligent,
adaptive, problem oriented action [2].

Problem environments can be characterized as structured or unstructured to
the extent to which programmed or nonprogrammed decision procedures apply
to them. Most interesting management decision problems appear to fall into
the relatively unstructured category. A typical attitude which has prevailed
in the past is that substantially unstructured problems are either 1) too
trivial to require decision support or 2) so complex that it is impossible

to make support system technology relevant. A more useful design hypothesis
might be that such problems are in general neither trivial nor impossible:
they are just difficult to solve.

In addition, it may be much more constructive to initially attack
the problems in terms of sub-problems and sub-decisions that can be
supported rather than solved by a computer-aided decision process. This

suggests that we look for aspects of the total problem where structure
is recognizable and use the computer to improve and mechanize those aspects.
Such sub-problems may include comparison operations, production of graphs
or other data presentation, arithmetic or primitive logical operations.
Here we are using our understanding of relatively common sub-problems to
help us "parse" our real problems into more manageable parts. Thus,
breaking a problem apart into sub-problems accomplishes two distinct things:

1) Problems rarely look alike, but often share common parts which
can effectively be operated on, and

2) We are helped to structure our thoughts in an effective and useful
way by suggesting what kinds of sub-problems might be particularly
useful to attack.

Many human problems influence the success or failure of attempted MIS
or MDS implementations. Decision makers may attend to the wrong criteria in
solving even their most important problems and may have substantial resistance
to change even in light of unsatisfactory current performance [3]. Individuals
or groups who control certain information may not want to relinquish their
control through adoption of a system that exposes their data to other
functional areas of the organization [4]. Decision makers may distrust
technically innovative ways of dealing with their problems, first, because


they can't afford the investment required to understand the innovation (a
new mathematical model for instance) and, second, because they may fear a
machine related takeover of some aspect of their job or responsibilities [1].
Perhaps more important, the decision maker cannot afford the risk of
accepting model results as an input to an important decision process without
understanding the model. Furthermore, people are often more sensitive to
computer errors in data processing than to the same errors made by human
processing [5].

Also, as was suggested briefly above, most information processing
technology has, in the past, been "packaged" in such a way as to require
that a manager assume substantial risks in attempting to acquire and use it.
These risks often involve:

(1) Acquiring expensive computer hardware

(2) Building and supporting expensive software

(3) Incurring substantial time delays involved in accomplishing (1)
and (2) and thus delaying potential returns.

(4) Incurring the costs and delays inherent in attempting to link such
systems together.

It is small wonder, then, that successful applications of decision
support technology are reported so rarely.

A common assumption in many MIS implementations is that expanding the
availability of the data base to managers or increasing the quality and quantity of
information available should lead to better decisions. In fact, most
managers do not need more information and the models that they employ in
dealing with this information are often primitive, simple, historical
models [6, 7]. In many problem environments it may be much more fruitful
to improve the information processing ability of managers so that they may


deal effectively with the information that they already have, rather than by
adding to the reams of data confronting them, or by attempting to directly
improve the quality of those data [6]. The system discussed here is directed
towards this objective.

It is also important to note that in any unstructured environment, one
of a managers most difficult tasks is to "find" a useful problem to work on.
Pounds [ 8] discusses this problem at some length. One of the most valuable
aspects of a decision support system is that it often suggests to managers
(cf.the case study which follows) what problems might usefully be considered.
This represents an important contribution to the managers ability to handle
problems, particularly those which are "novel" in some sense. In this
fashion, a system may generaly aid and support the managers problem finding
(and problem solving) process by giving him the confidence to tackle new



In order to test many of these ideas, an interactive planning support
system, called PROJECTOR, was implemented and tested on a number of
students (including several practicing managers) at the Sloan School of
Management, MIT. In order to understand the case study which will be
presented, it is necessary to describe this model in a functional fashion.

The PROJECTOR system is a long range financial planning model for
new enterprise, acquisition and project analysis. The system was designed
and implemented by the authors . The primary thrust of the effort was
twofold: first, to bring together many of the advanced tools for long
range financial planning into a single interactive computer model and,
second, to provide an effective learning tool for graduate students and
professionals studying financial management and long range financial
planning. The underlying assumption was that graduate students of
financial planning and development who were exposed to a realistic,
professional planning tool would be better trained as a result. An

overview of this planning system is given in the next section. More detail
can be found in The PROJECTOR On-Line Planning System [9]. The design and
implementation of PROJECTOR took place over a period of many months
beginning with an early predecessor at a major U.S. chemicals and plastics
producer. Throughout the entire design and implementation process, managerial
users of the system were continuously consulted. The system went through
several evolutionary stages at both the chemicals and plastics company
and at M.I.T., largely as a result of strong continuous model user input
from professional managers, professors and graduate students of management.

*In the process of developing this system. Professors John D. C. Little, Gerald
A. Pogue, Devendra P. Garg, and Henry M. Paynter provided much useful advice. They
are, however, not responsible for any errors or omissions.


PROJECTOR is a user-oriented, interactive computer planning model designed
to facilitate financial planning. It has report generation capabilities
for decision makers who are not necessarily familiar with computer languages
or operations. The primary thrust of the system is to support the cognitive
skills of the sophisticated financial planner with the computational and
storage capabilities of a high speed electronic computer.

Of the multitude of functions that high-level decision makers must
perform in attempting to realize organizational objectives and goals,
long range planning has perhaps the most profound impact on the success
of the organization. Clearly then, it is essential that this key planning
function be executed in a systematic, consistent and technically sophisti-
cated problem solving environment. PROJECTOR is designed to provide that

The model is designed to allow the long range planner to express his
view of alternative capital investment policies or other financial ventures
via a remote computer time sharing terminal and evaluate alternative strate-
gies in light of factors he feels are important. Rather than force the
decision maker to evaluate strategies and alternatives on the basis of a
single arbitrary decision criterion such as net present value, discounted
cash flow rate of return, profitability index, minimum cost, maximum
profit, benefit/cost ratio, return on investment, return on sales, profit
margin or payback period, the planning model allows complete freedom of
decision criteria selection according to whatever the planner feels is
appropriate to the particular situation. In its most fundamental sense,
PROJECTOR is a decision-support system, relying heavily on the expertise
and experiential background of the financial planner. Although the model
is designed to address normal long range planning and report generating


functions, it is also valuable in "crisis" planning situations where time
is short or when sudden changes in the long range planning or investment
horizon become apparent.

Since PROJECTOR is a generalized model, with application to a wide
range of long range planning problems, it is possible for the user to use
a variety of options to obtain a custom tailored version of the model.
All of this can be done without learning to program the computer and
without dependence on a programming or operations research staff. The
time spent in developing a model (at least initially) is small, and yet
useful results can be obtained. All interactions between the computer
model and the decision maker consist of English language words, sentences,
or their abbreviations, and of course the input data needed to describe the
financial planning problem to be explored. The computer, through the remote
time sharing terminal, will ask in a conversational form for the relevant
project data, for the computational options that the manager may or may not
use in a particular situation and for the other financial factors
that the manager considers important to his particular planning problem.
The computer will request data only on factors that the manager indicates are
relevant to the current application. In addition to the capabilities avail-
able in the planning model, it is possible for the manager or his staff to
build their own models, submodels and other computer executed procedures to
be used in conjunction with the standard PROJECTOR system. These special
procedures might include, for example, a unique report generating function,
a user supplied computational algorithm or a standardized data management
routine. The special procedures are added in the form of FORTRAN language
subprograms. Also, the experienced model user may enter input data and


model configuration parameters in the form of a brief data file to minimize
the time used in setting up and evaluating alternative financial ventures.
This is particularly useful when working on one particular problem over a
period of time.

In addition to new enterprise, new product and project planning
capabilities, PROJECTOR implements some relatively sophisticated management
science models for forecasting. These include univariate and multivariate
regression analysis as well as exponential smoothing with trend and seasonality
analysis. Optimal exponential smoothing model parameters can be

Merger and acquisition analysis is facilitated by the implementation of
a model for determining the optimal mix of various merger target goals such
as earnings per share dilution, post merger debt-to-equity ratio, percentage
ownership, working capital requirements, and package fractions of common
stock, cash, bonds and convertible preferred stock. The goal programming
technique is used to aid the management decision maker and planner in
realizing optimal strategies in the process of pre-merger planning and
negotiation. Refer to Figures 1 through 6 and Tables 1 through 4 for
further information on the organization of the planning system.


The Scenario

The divisional planning project described here took place at a small
New England manufacturing company which will be called Acrofabrication
Industries, Inc. (Acrofab). The president of Acrofab was in the process of
considering the incorporation of a new subsidiary, which will be called
Vehicle Security Systems (VSS) for purposes of this discussion. Approximately


eighteen months before the formal process of divisional planning has been
initiated, certain members of the Acrofab management group discovered what
they believed to be a potentially lucrative new product need in a market
segment adjacent to and complementing a rapidly growing consumer durable
goods industry. Shortly afterwards, a research and development program
was initiated to determine whether a product could be produced to satisfy
reasonable cost constraints and the technological constraints associated
with the perceived product need.

The research and development efforts were directed toward the development
of an easy to operate and inexpensive security device for protecting certain
classes of consumer durable products. The new security device was given the
name "Interceptor" and a comprehensive market feasibility and research study
were initiated as soon as the technical team at Acrofab were convinced that
they could build Interceptor for a reasonable price. After several months •
of research and development activity, a prototype of Interceptor was completed
and tested. Results of the prototype testing, which was done by Acrofab
personnel, indicated that in fact Interceptor offered a much higher-level
of security under several possible theft/intrusion variations than did
existing competitive systems designed for the same purpose. These existing
systems currently dominated ninety five per cent of the high security
market for such products. Substantial increases in theft rates of
products for which Interceptor protection was planned indicated that
existing security systems certainly did not provide adequate protection
for the target products. Further testing indicated that Interceptor was
technologically superior to all existing competitive products. Interceptor
prototypes functioned well and the security system was not frustrated by a


wide variety of commonly used intrusion or theft strategies. On the basis
of a favorable product test and this market research program, a decision to
go ahead with the division plan was made.

The Divisional Organization

It was decided that the Vehicle Security Systems (VSS) division would be
incorporated separately as a wholly owned subsidiary of Acrofab. This strategy
was based on the high degree of risk and uncertainty of launching even a
technically superior product into a market dominated by well entrenched
nationwide organizations. As a result of the VSS divisional organization,
VSS would have to seek support from the financial community on the basis of
its own merits, without the benefit, for instance, of parent company joint
liability arrangements. Of course Acrofab would be willing to invest some of
its own capital in VSS but the president felt that the risks were too great
and too poorly understood to make unlimited funding available from the parent
company to VSS. At this point the president decided that he needed to obtain
some support to aid in the development of a comprehensive divisional plan
that would be essential for internal planning purposes and would be a
valuable selling point when he was seeking formal external financial support
for the VSS division.

The president was interested in examining several different issues
related to launching the new division and the Interceptor product line. The
historical objectives of Acrofab had been to offer its clients good delivery
service (minimum lag between order entry and shipping date) and high quality
manufactured merchandise while trying to maintain minimal working capital
investment and low variable costs. Various sub-objectives included maintaining
capital structure and organization product interdependency within prespecified
bounds and controlling the level of risk involved in corporate ventures through


various classical methods of planning and analysis. With such conflicting
and dimensional ly incompatible objectives and subobjectives it was obvious that
no optimal plan could be determined even if there were much less uncertainty
in the new product market.

Initial Steps

The president had some formal analytical training since he had
obtained an M.B.A. degree from a technically oriented graduate school
of management. He decided to obtain technical assistance in the form of a
professional consultant to help in setting up some kind of formal planning
model to aid in the analysis of the new division. At the time of his original
outside contact with the consultant, he was not sure of what type of formal
model he needed or in what way such a model might be implemented at Acrofab.
He did, however, have very well defined questions about issues that he wanted
to explore in relation to the VSS division and he had collected (or estimated) a
substantial amount of data relative to the major divisional issues. In fact,
at that time he had substantially more data than he had ideas of what to do
with it.

The president wanted to use an analytical model to gain insight into
questions such as what levels of inventory and other working capital components
would be appropriate for a given level of production and sales. More
importantly, what would happen to his cash flow situation if these variables
began to change rapidly or failed to meet the expected levels? In mapping
out the original version of the divisional plan, the marketing department
had specified a marketing life cycle analysis to be used in the financial
calculations (see Figures). The president felt that it was very important
to determine the sensitivity of total project profitability and value to
changes in the deterministic product life cycle assumptions. Specifically,


what would be the impact of changing the slope of the life style curve or
expanding or contracting the estimated horizon to product maturity and
decline? What about the impact of increases or decreases in total product
life sales on the profitability of the new division?

The president knew that these issues could have a profound impact on the
value of the proposed new division to Acrofab. Furthermore he was concerned
about basic relationships between investment in working capital, total
production, variable costs, actual cash flow and project acceptability.
The dynamics of the market place itself were of major interest since changes

1 3

Online LibraryC. Lawrence MeadorA case study of the use of a decision support system → online text (page 1 of 3)