The Basic Approach

I t is still too early t o speculate about t h e final form of t h e decision sup- p o r t system (DSS): t h i s is a long-term goal. In expanding t h e basic approach, we will rely upon t h e findings of psychologists, m a n a g e m e n t scientists, a n d specialists in DSS, a n d upon experience gained from t h e i m p l e m e n t a t i o n of models of R&D project selection. We expect t h e development of o u r s y s t e m t o take a few years. Working closely with t h e decision m a k e r s of t h e firm (who should welcome efforts t o make decision making m o r e objective), we will have t o decide a t each stage whether i t is worth while t o c o n t i n u e t h e basic approach. Caution in development is recommended because few successful applications of models of R&D project selection a r e known, and n o t many of t h e recently suggested approaches have been t e s t e d in a wide r a n g e of practi- cal situations (multiobjective decision making u n d e r u n c e r t a i n t y , fuzzy analysis, e t c . ) .

Our procedure was selected because of i t s flexibility a n d because of t h e g r e a t n u m b e r of successful applications t o r e a l problems, including t h e m a n a g e m e n t of R&D, t h a t have been reported. However, f u r t h e r progress in our r e s e a r c h will very m u c h depend on t h e success of our efforts t o s t i m u l a t e support from decision m a k e r s and from specialists in a n u m b e r of disciplines, s u c h as multiobjective decision making, decision analysis, a n d c o m p u t e r techniques.

A basic principle i n t h e development of a 1)SS is t h e modular principle:

all of t h e techniques applied have t o be compatible s o t h a t t h e analyst (or decision m a k e r ) c a n combine t h e m a t will. This principle g u a r a n t e e s t h a t t h e model can be adapted to new r e q u i r e m e n t s or t o new findings i n t h e rapidly changing fleld of DSS. We a r e still searching for t h e best approach t o o u r problem and t h i s is why we a r e testing a n u m b e r of approaches (Section 2.3).

In practical applications, hybrid models have often proved successful (Hogarth 1974, Bunn 1978, Chapman 1979).

With o u r p r e s e n t under-standing of t h e decision situation we t h i n k t h a t modules with t h e following functions a r e necessary lor t h e decision s u p p o r t s y s t e m :

to process d a t a on innovations t h a t have already been phased o u t o r t h a t have r e a c h e d an advanced s t a g e ;

t o r e p r e s e n t how t h e m a n a g e r envisions t h e development of c e r t a i n ongoing or new innovations;

t o forecast i m p o r t a n t quantities needed for planning innovations (technological a n d scientific t r e n d s , supplies of m a j o r r e s o u r c e s , development of t h e firm's capabilities, e t c . ) ;

t o c r e a t e scenarios;

t o t e s t long-term effects of s t r a t e g i c decisions on R&D programs (e.g. risk);

to process t h e judgments of e x p e r t s ;

t o r e p r e s e n t t h e m a n a g e m e n t view of t h e long-term objectives of t h e firm; a n d

to s e c u r e a n efficient man-machine dialogue.

While all of t h e modules a r e closely i n t e r r e l a t e d , t h e y m u s t o p e r a t e indepen- dently s o t h a t c h a n g e s in t h e organization of t h e Arm will n o t have c a t a - strophic consequences for t h e s y s t e m a s a whole.

Our p r e s e n t r e s e a r c h is focused mainly on module 2, a r o u n d which t h e o t h e r modules will be developed s t e p by s t e p . Experience suggests t h a t t h e process of developing a DSS m u s t be iterative, adaptive, a n d flexible (Keen 1980). In presenting o u r first ideas a b o u t t h e DSS, we s t i m u l a t e t h e decision m a k e r t o specify m o r e precisely his expectations of t h e s u p p o r t h e is t o be given .

4 - 2 Interactive Mode of Operation

Before we discuss t h e basic model a n d several versions of i t , we m u s t say why a n interactive mode of operation is necessary for o u r c a s e s t u d y . When faced with t h e evaluation a n d selection of innovation projects u n d e r t h e cir- c u m s t a n c e s of multiple objectives, u n c e r t a i n t y , a n d t h e prospect of long- lasting effects on t h e company a s a whole, t h e decision m a k e r is often u n a b l e t o a r t i c u l a t e h i s p r e f e r e n c e s well enough for u s t o c o n s t r u c t a utility func- tion. In most c a s e s , t h e first presentation of t h e problem will be very vague a n d will have t o be c o r r e c t e d via feedback loops. This is t h e m a i n reason for involving t h e decision m a k e r in t h e problem formulation and solution a n d in t h e evaluation of t h e r e s u l t s . The decision m a k e r m a y wish t o c h a n g e s o m e of t h e d a t a on which t h e decision t r e e is based, s u c h a s t h e r e s o u r c e require- m e n t s for a c e r t a i n project p a t h in a particular period, t h e expected benefits of realizing a p a r t i c u l a r project, or t h e probability of c e r t a i n c h a n c e nodes.

This m i g h t r e q u i r e a r e a s s e s s m e n t of t h e i m p a c t of t h e c h a n g e s on t h e final o u t c o m e . He m a y even wish t o change whole b r a n c h e s of t h e decision t r e e .

The g e n e r a t i o n of a feasible s e t of a l t e r n a t i v e s is i n s o m e cases m o r e i m p o r t a n t t h a n t h e solution itself because it p r e d e t e r m i n e s t h e final choice.

Our approach is intended to be process-oriented and should allow for any changes t h e decision maker wishes to undertake. Many models for project selection have been rejected because t h e decision maker felt t h a t his prefer- ences were reflected inadequately. An interactive procedure greatly increases t h e decision maker's confidence in the method. Zeleny (1980) stated t h a t t h e "human decision-making paradigm m u s t be amplified r a t h e r than ignored, respected rather t h a n degraded." Interactive decision making is t h e b e s t way t o m e e t this demand.

The main idea behind interactive decision making is jointly to solicit t h e decision maker's preferences and investigate t h e feasible alternatives for t h e eventual determination of an optimal solution. The most important facet of a n interactive procedure is the ability of the decision maker t o answer t h e questions asked by the algorithm. One cannot expect him to answer ques- tions t h a t a r e difficult even with a computer.

Larichev and Polyakov's (1980) classification of interactive procedures is based on t h e distribution of the work between t h e decision maker and t h e machine. They distinguish between unstructured, pseudostructured, and s t r u c t u r e d procedures, which differ in t h e degree of involvement of t h e deci- sion maker in finding a solution. In this report we consider only structured procedures because of their relative simplicity. Structured procedures reflect t h e results of psychological investigations, t h a t human capabilities for com- paring multiattributed alternatives a r e very limited. Hence interactive pro- cedures should ask simple questions.

4.3 The Basic Model a n d Difierent Versions

The basic model c a n be formulated in several ways, depending on t h e size of t h e problem. This is determined by:

t h e number of projects under consideration;

t h e complexity of the decision trees (numbers of decision and chance nodes and corresponding branches);

t h e number of periods;

t h e number of types of resources formally included.

If t h e problem is not too large, t h e evaluation and selection problem takes t h e form of a stochastic linear programming problem in which uncertainties about t h e f u t u r e a r e incorporated into t h e objective function.

The decision variable is zij, j t , which is the j t h path of a project i in period t , and i t is assumed t h a t t h e future s t a t e , j , of t h e world occurs, which is determined by t h e outcomes of chance nodes up to period t (further details are given by Gear and Lockett 1873). The constraints of this model version e n s u r e t h a t not more t h a n one project path will be selected for each project, t h a t resource availabilities a r e not exceeded i n any of t h e periods, and t h a t whole paths a r e either adopted or rejected. Since t h e values of t h e end points can be expressed in monetary t e r m s , the objective can be formu- lated a s t h e maximization of t h e overall expected s u m of t h e final values of t h e projects. Similar expressions can be found for t h e other two objectives

mentioned a t the beginning of Section 4. Other constraints arising from pecu- liarities of the firm can be included easily.

The first version of our model takes into account t h e order in which deci- sions are due and uncertainties arising in each project over time. Projec- tions of all possible future states of t h e world a r e obtained with one computer run. By defining the nodes of the decision t r e e in a n appropriate m a n n e r , one can take into account uncertainties about resource requirements, project durations, and project outcomes. The results of t h e calculations indicate how to allocate t h e available resources to certain selected projects in period 1 in order t o be on t h e optimum path. If t h e number of decision variables is large, difficulties may arise in t h e analysis of the solutions.

From our viewpoint, this first formulation of t h e problem is well suited for interactive multiobjective decision making and can be combined easily with t h e s t e p method or the reference point approach (Wierzbicki 1979a, b, 1980, Kallio e t al. 1980), because all uncertainties involved in t h e decision t r e e s a r e represented in t h e objective function. In solving t h e problem interactively, the decision maker can manipulate factors only in t h e objective space.

We hope t h a t because of t h e relatively small number of projects in o u r case study we need not exceed t h e limits of solution with t h e existing stan- dard packages. Moreover, i t is possible t o reduce the size of a problem t h a t has become too large by reducing t h e number of chance nodes in the decision trees. The necessary theory for t h e single-objective case is provided by Lock- e t t e t d. (1980). Finally, one should be satisfied with a good feasible solution having upper and lower bounds on t h e expected value of t h e optimal solution instead of strong optimization, which h a s no real sense (Lockett e t al. 1980).

As t h e size of t h e linear programming problem increases with t h e size of t h e decision t r e e , i t becomes more and more difficult t o solve t h e problem with existing standard solution packages, even when a branch-and-bound method is applied. For this reason, a n alternative approach h a s been developed. It combines linear programming, simulation, and heuristic interpretation of t h e results. Each path of a given innovation project is deterministic linear programming. This simulation results in some addi- tional constraints, their n u m b e r corresponding to t h e n u m b e r of project paths in which chance nodes are incorporated. The constraints differ only in t h e right-hand sides of t h e linear inequalities. This approach was first reported by Lockett and Freeman (1970). The application of Monte Carlo tech- niques has also been proposed by Allen and Johnson (1971). We developed a

n u m b e r of innovation projects with complicated s t r u c t u r e s and highly disag- gregated resources and periods. A weakness of t h e method is t h e problem of final choice. Also, i t is not suited for a n interactive mode of operation.

Nevertheless, i t c a n be used a s a convenient starting point for a n analysis using the approach discussed above combined with man-machine dialogue.

We think t h a t a combination of both approaches is t h e best way t o arrive a t t h e m o s t realistic picture of t h e whole decision process in innovation proj- e c t s .

The basic model discussed h e r e is l i n e a r . Many detailed studies have indicated t h a t linear models provide good simulations of real R&D situations (Bell a n d Read 1970, Allen and Johnson 1971), a r e easy t o handle, and a r e easy for decision m a k e r s t o understand. They c a n be easily expanded for multiob- jective decision-making problems. (The t h e o r y a n d a n u m b e r of c o m p u t e r programs for multiobjective linear problems a r e widely discussed in t h e l i t e r a t u r e , e.g. by Evans and S t e u e r (1973) a n d Zeleny (1974).) In c o n t r a s t , nonlinear problem formulations do n o t add t o o u r understanding of reality and often c a n n o t be solved by s t a n d a r d computerized solution techniques.