Comput. & Graphics, Vol, I, pp. 293-296
Pergamon Press, 1975, Printed in Great Britain
GRAPHICS IN BUSINESS DECISION MAKING* IRVtN M. MILLER International Business Machines Corporation, Poughkeepsie, N.Y. 12602, U.S.A. Abstract--Mountains of data can be converted into easily digested graphs so that decision makers can grasp the peaks and trends of economic situations and make rapid and confidential decisions. We will study how we can define simple models using a graphic-based Industrial Dynamics system (Forrester of MIT), then interact with them symbionically to obtain graphs of various manufacturing or sales considerations. We will proceed to apply graphic analysis techniques to these graphs in order to produce curves that identify the sensitivities of the analysis, to reinforce confidence in the decision to be made. Finally, we will discuss the progress and the potential of graphics in business decision making. Graphical representation of data is a rapid, concise and effective way of communicating complex information. Or, to rephrase, a picture is worth a thousand words. Computer graphics has been available for approximately twenty years; yet, in spite of a relatively significant impact on mechanical design, little impact has been felt in the business decision process. Why? Cost of equipment is often cited, as well as limited utility, but these are only superficial reasons. If business decisions can mean the difference between losses or gains of millions of dollars, the initial one hundred thousand dollar investment that would have been required five years ago would be justified. Expensive equipment and supposed limited utility have discouraged the business executive from risking research dollars to overcome these hurdles. Then how does he get his feet wet? The graphics developer could follow the example of the diving instructor, who starts his beginners at the edge of the pool with a jump, not with a double somersault off the twenty-foot board. Let us look at some graphical developments and evaluate their impact. Five years ago a business model was adapted to graphics to illustrate the power of graphics in solving a particular business application. A manufacturing retail firm which produced two products for four markets was modeled graphically so that we could show the ease with which we could obtain a solution of problems of optimum production and pricing. To illustrate the model with the distribution of one product, consider the following figures. Here (in Fig. 1) we have an index which allows us to select an option at random. Figure 2 results from selecting "Factory Costs" from the index, using a light pen. Here we can select the number of machines, and change them from 450 to 800. Then, returning to Fig. 1, we can select "Revenue Parabolas (A)" and obtain Fig. 3. Similarly, we can obtain "the marginal revenue and cost curves" in Fig. 4. However, with Fig. 4 we can point to the intersection of
T~BLE OF COHTEHTS
1 2 3 4 S 6 ? S
FACTORY COSTS I'tNRKET COSTS AND D£1%~ND ( R ) f't~RIC.ET COSTS AND DEffiAND ( B ) R I [ t E I g J [ PCL'R~SOI,J ~ CA) REUE~JE PAI~IIOI.A~ (B) MARGINAL REVENUE AND COSTS ( A ) MARGIHAI. REVENUE Af~ID COSTS ( B ) PRODUCTION AMD PROFIT
NEXT
Fig. 1. F~CTORY COSTS ,.~13f'lrNi ~TRAT I LJE COSTS TO PRODUCE Of~L~/ A TO PRODUCE B TO E N C ~ SECOND S H I F T MAI HTEHeI,K;E COST/r~HINE
4 ~ . , ~ ee ~ ~10 1S~ ~NP ~¢w4D00
PRODUCTIOH COST PRODUCT R ~ATERIAL ~CHINE ~ ~ HOUR~ PRODUCT i P~T~IAL Pto~CHINIE fIRS HOURS
!1~ 00 ~ $0 88000
4@ee. eg ¢~m, ~11 40e, 00
L A ~ 3 R COST F I R S T SHIFT(PER ~ HOUR~ SECOND S H I F T SECOND SHIFT OUERTIME
HUMBER OF A~JAILRBLE MACHINES
3.80 3 4S S 18
80e 00
BeCK I H~X NEXT
Fig. 2. TOTAL REVENUE ~ D
COST CURUES
I O~'f
7M
1M 0
*Paper presented at the Conference on Computer Graphics and Interactive Techniques, 15-17 July 1974, sponsored by the University of Colorado Computing Center and ACM/SIGGRAPH.
0
IK
~
3K
4K
~IK 6K
?K" ~ ( ~K INDEX
Fig. 3. 293
~
10K I I K I N
13~( 141( 15#,
294
IRVIN M. MILLER M~RGIN~
RE~JENUE ~,ND COST CURUES
~GIN~L
REVENUE ~ND COST CURVES
\
IK
0 1K
2;<
3K
4K
( PRODUCr a )
~(
FW,C
1K
TX
~
3K 4K (PRODUCT Fe}
SK
6K
~CX INDEX
INDEX
Fig. 4.
the marginal revenue and cost curves and obtain the optimum operating point, resulting in Fig. 5. Then, indexing to Fig. 6, we can look at our production and profit figures. On the surface this graphic model represented an easy interactive method for changing numbers and observing results quickly. Though ease of problem solving was one objective, our main objective was to provide the user with insight. For example, in Fig. 3 we observe that the cost curve terminates abruptly just below the peak of the parabola. This observation gives, along with interactive conjectures, the effect on costs and on production capability of the number of manufacturing machines. In Fig. 4, we will occasionally see discontinuous jumps in our marginal cost curve, indicating errors in our numerical differentiation algorithm. One does not have to be a mathematician or a programmer to observe the reliability of the model. This model has proved valuable as a demonstration and educational tool. However, its impact on the observer is like watching a cliff diver--my, that is impressive, but I would never do it myself. The executive could see the value of the model, but it represented far too great a change in his development process for him to reach the stage of using the graphic model. Furthermore, the distrust of models, which the graphic model had the potential of reducing, still existed. Recognizing that a more elementary approach was required, an experimental system was developed to give the user a simple problem-solving capability. In Fig. 7 we show the index of that program. We shall consider only those menu options which represent the simplest analysis thread. We obtain Fig. 8 by choosing the axes option in Fig. 7. We can now select axes limits, scales, and the number of points. The y limits represent the maximum window size, so that curve values less than the window size will be scaled to use limits less than the y limits, to give maximum display viewing. Next we return to the index, select the define option, and obtain Fig. 9. In this display we indicate whether we are defining the graph number for the curve to be plotted, or we indicate which curve we wish to define. We then obtain Fig. 10, which
Fig. 5. PlARkET I PRODUCT A UOLUME Pi~ICE
~ET
8
6~4 ~309
S~ R386
PRODUCT B VOLUME 0 PRICE 17~86
17231
et~RKET 3
@
MEN RRW MATERIRLS TOTRL COST
L'S4371S
SEFOR~ T~X PROFITS
19(11311
r'W~KET 4
S)93 8407
t(WDi 8398
e L~lOe
S ~.78$?
I~3 s6"l~wr'/9
IN~'X
Fig. 6.
c u p t , 'r,¢~ L~BL
PNtNS[TERS
Gie#ffM 1234
~,'aXS
•
kin
1 ~
C3 e4 CS C6 C7 C8 CS CIO Cll C12 el3 C14 CIS C16 C17 C18 C19 CL:~ DEFIh~ STOP
IO kalJl[L
~M AXIS
DEI.~ WINDOW
Fig. 7.
X~AXI~ LOW HIGH L03 LIH NPTS
Y-RXIS
0 ~.¢~eeee
e
1 5e
1
EHD
Fig. 8. allows us to define a standard curve or to create a special curve. We select operations, which gives us Fig. 11. This operation option provides us with a set of analysis operations which prove to be very powerful. Among the standard arithmetic operations we also offer differentiation and integration. Returning to the index (Fig. 12), we now see the results of defining several curves. Finally, in Fig. 13, we see the plot of these curves.
Graphics in business decision making
295
DEFINE C142 LINE 1 O143 LIrIE C144 LINE C14S LINE C146 L I N E C 1 4 7 OPER C 1 4 8 PARR 1 C149 OPER O1~00PER C1~10F~R CIS~ OPER C153 OPER C 1 5 4 OPER C15S I~ER O I E ~ OPER C157 O ~ E R C1S80PER CIS90PER C160
-0 e?2 - 0 e~37~ 0 0454 -0 036 -0 286 1 5 • ~ ' M ~ C148/C147 C149/C147 ClSe/Cl4? C151/C147 92 C142,_C14B C143-C149 C144-C150 C145-C151 C146-C15~ 0
• 003
a E 2 2
64 I 1 1 I
~ 2 E
300
2 I E
200
2
100
GRAPH 1 2 3 4
CURVE
I I I 1 t
34E9 3265 311 ~16.2
E~D
0
Fig. 9.
Fig. 13. OPt[RATIOt~
C160CURVES OPERRTIOt4 NURIIIrI~ Cl + e C142 t C143 / 2 C144 :It 3 C14S $Z 4 C146 MAX S C147 RIM 6 C148 LOG ? C149 EXP g C1~ $UIq 9 CIS1 DATA CtS? Ibrf E CIS3 IrLItlC C 0154 coPY REDO 01$S DIF FIMl Cir~ ll¢l~ C157 DIglI
,- 1 6 0 FUNk'T IOMS I
PARABOLA
2 3 4 S 6 ? 8 9 10
HORF1AL GROWTH UHIFORfl WEIBULL ~T LINE OPERATI O~ POINTS C U ~ FIT DRAW
Fig. 10.
etr~
CISI;
Fig. 11. PROGRN'I PARAI'~TER'S
CLOt.= T'~PE LABL
¢I
A
>,A•5
CI'13 Lithe C144 LIt<£ C145 LIME C146 LINE C147 OPI[R C14" P ' B C,49 OPER CISO O P E R C1$10P~R C1~ OPIER C153 OFIER CIS40PIER ClS50PI[R CIS60PI[R CIS? O~R CIS80PER C159 OPER CIE~
o 0
• ~7= 0 04S4 -0 036 -0 ~ I $ 0 ~--'P~ C148/C14? C149/C147 elSe/C14? C151/C147 ~ C14B C148 C143 C149 C144 ClSO CI~"Clll C146-C15a 0 DEFINE STOP
IO LABEL
B LIM l.IN
ORRPH 1~34 1
C
1 ~
34~ 9 3~E $ 311 ~ 2 ' "3
t
' 2 I E I B
64
I 2 a 1 = I 2 I E E 3 3 3 3 3 3
~ AXES
D£LETE WINIX~J
tool of industrial dynamics, DYNAMO, we built a graphic implementation which added several key features to it. First, graphs could be defined more easily and more subjectively; second, the model could be built more easily with fewer errors; and third, the results could be presented with high resolution graphs. A series of panels have been established that make it easy for the user to build, run, and test his model. The first panel, Fig. 14, shows the master index for selecting the various options randomly. First we select "Number of Points," which allows us to specify the number of intervals for running the model. Next we select "Labels." The user then has the choice of reading the labels from cards or of typing them from the screen. A blank label ends the label entering session. One can then assign values (Fig. 15) to these labels for execution of the program. Next the user can type in tables (Fig. 16) or draw in graphs (Fig. 17) to be converted to tabular form. It is this table generation and modification technique which gives the graphic approach a significant user-oriented advantage, particularly when one is interested in shape rather than numerical values. Once a graph or table is generated, the other mode may be used for modification. Once the constants, variables, and tables have been defined, the user can create his program. If he has neglected to define a label or table, he can always do so. In the program creation mode, he is lead through the creation of a statement by being given a choice of labels when appropriate, and subscripts and operations as needed (Figs. 18-20). If a statement is to be deleted or inserted, he can easily perform that task. Once the program is created,
Fig. 12. In our first graphic application we showed a rigidly developed model which allowed us to solve a problem by changing parameters, by specifying solutions, and then observing results. However, if we wanted to change the basics of the model, we could not do so without reprogramming. This second program, the analysis approach, not only allows us to solve simple models, but also allows us to build models such as the first one and to solve them. In Fig. 12, we saw such a model created with solution provided. A compromise of the two graphic approaches leads us to Industrial Dynamics (Jay Forrester of MIT). Using a
PO [ N T 5 LABELS •rNZTI~#. TABLES PLOT ~ EXECUTE D I S K I IRT D I SKREI~D IEXT DB END
Fig. 14.
T DT ..-E~', TtqO lid TH~ C ~ Pit
0 1 0 2 100 200 1000000 0 0
D I lid TI R TR PR aD END
0 0 0 0 0 0 0 o CONTEHUE
Fig. 15.
296
IRVIN M. MILLER pt"t
500K
,r I~'S T L,.~ST STEP
1
12 1
400K
200 31~5 405
1
,2 4 5 6 ? • 9 10 11 12
300K
273 161 262 36S 405 344 263
200/(
END
10~
Fig. 16. .
2
500
.
.
.
:3
.
.
4
.
.
,
S
.
,
$
?
,
,
8
.
.
•
.
.
.
10
.
.
.
11
12
Fig. 21. 400 f
300
200
100
I i
,
i 3
.
.
.
4
.
.
.
II
.
.
•
.
.
?
.
II
.
.
.
•
.
.
.
10
.
1t
II
Fig. 17. r LI~ R
Dr PN TR
ZERO P PR
TI4O D ~D
HD $
THD ND
E TI
Fig. 18. P.K*HD.KeC.K
P.KmHD
Fig. 19.
.J .K
/
NIN LOG J*
Fig. 20. he can punch the labels, tables, and programs on cards to be read back at a later date. He can also write it out to disk, if he does not want to use cards. Before executing the program, he specifies the independent variable and the variables to be printed and plotted. After the program is executed, it can then be displayed (Fig. 21). The results on the graph, scales, windows, and grids can be changed easily. If one is not satisfied, a value or table can be
changed readily. Once the model is built, one uses Figs. 14-17 to change input parameters and to rerun the model. Thus anyone may use the model. An interface has been provided between the graphic DYNAMO and the analysis program, so that further analysis can be performed on the results. So far this graphic DYNAMO program has been used only for demonstrations. We find that it is not graphics, but the awareness and use of economic techniques, that is in short supply. The business executive should be educated in such techniques, preferably with case studies in his own field. One should know how to swim before learning to dive. To further extend the usability of graphics, the analysis program has been rewritten in APL and in other terminal languages. We have also written data base access programs to provide a total business analysis effort. In reviewing, we find that C. P. Snow's two cultures are bearing full fruit. The optimism of the scientist forges ahead of the traditionalism of the non-scientist. The engineer is part of the rapidly changing technological environment. His training has disciplined him to take risks and to adapt to changes. Thus graphics has been accepted in the engineering ranks, though with difficulty and expense. The business executive, on the other hand, views technological progress conservatively, so that he can extend to himself the experience of the wild duck wading and surviving the technological turmoil. Thus graphics has yet to make significant inroads in his ranks. Once accepted by the executive, however, business graphics should inspire serendipitous observations which will convert decisions based upon insufficient knowledge into sophisticated, successful strategies.