New product portfolio management under uncertainty
In most organizations the decision making processes associated with the development of new products are broken down into three hierarchical levels: strategic, tactical and operational; that use data and decision support tools individually tailored. In each level, only the part of the process considered relevant to that particular level is modeled in detail, while the remaining parts are represented in an aggregated fashion or even disregarded. In principle, this decomposition strategy allows managers to break down the problem into smaller pieces and concentrate on the decisions that are relevant to each level, without significantly affecting the overall financial performance of the development pipeline. The aim of this study is to characterize the gap between the set of strategic decisions based on a decomposition strategy and those obtained by using a comprehensive approach, in which the decisions from all levels are considered simultaneously. For that purpose a multi-phase simulation-optimization decision support framework is developed and a specific instance of the framework is applied to an R&D pharmaceutical development pipeline. In the process of developing and implementing the framework a series of theoretical and practical issues were found, opening the door for the rest of the work undertaken in this thesis. On the theoretical side, it was found that the conventional net present value and decision tree analysis provide inaccurate measures of financial performance. The inability of these methodologies to properly discount the portfolio cash flows according the corresponding risk profile motivated the development of a valuation method that combines real options analysis and discrete event simulation. On the practical side, the need to realistically represent the flexibility in the allocation of resources without significantly increasing the computational burden of the simulation resulted in the development of alternative formulations of the multi-mode resource constrained multi-project scheduling problem. The formulations use continuously divisible resources and continuous time representations to minimize the increase in the number of binary variables caused by larger time horizons and portfolios. The simulation time was not the only practical consideration; the ability to deploy and maintain the framework in an industrial setting was also examined. This motivated the last part of this thesis, which consisted in the evaluation of commercial discrete event simulators from a simulation-optimization perspective.
Pekny, Purdue University.
Off-Campus Purdue Users:
To access this dissertation, please log in to our