Data-driven project portfolio selection: Decision-dependent stochastic programming formulations with reliability and time to market requirements

Abstract

We develop stochastic programming models for project portfolio selection under the plan-driven waterfall approach and the more flexible agile approach. The models account for the requirement to earn return fast and to generate a certain return with high probability. The models take the form of static (waterfall) and dynamic (agile) disjunctive integer nonconvex chance-constrained problems. To make the models computationally tractable, we devise model strengthening approaches and decomposition methods. We also develop an algorithm to obtain an ideal investment plan that provides the targeted probabilistic return as quickly as possible whilst maximizing the excess return. Using a representative US-based software company data, we show the significance of the benefits given by the ideal plan. Our results show that the probabilistic return can be reached faster under the agile, as compared to the waterfall, approach. This can partially explain why agile approaches are popular in new product development. Counterintuitively, the results show that the agile approach, which includes more stochasticity sources than the waterfall approach, leads to less uncertainty regarding the time to reach a certain return than the waterfall approach. The reason for this outcome is the dynamic abandoning and re-starting of new projects protecting from downside risks, and hence, from outcomes that would result in longer time to reach the required return level. Furthermore, we introduce a visualization tool to guide a venture capitalist’s investment. The visualization tool highlights the company’s performance regions derived with the proposed models. The numerical tests show that the developed models are robust and computationally tractable and can be used for larger problems, with more projects, time periods, and uncertainties.