A genetic algorithm system for product line exploration and optimization

We report development and industrial application of a genetic algorithm (GA) model to find near-optimal product portfolios for marketing management. This GA model uses consumer preference information that is typically available from marketing studies such as choice-based conjoint (CBC) analysis and other discrete choice model projects. Because a single result might capitalize on chance, the system does not simply find one optimal portfolio but instead allows individual-level and model-level bootstrapping of results. Examination of the resulting distribution of near-optimal portfolios is informative for strategic insight and generation of market hypotheses. We describe application of the GA model in a personal computer accessory product line for a major manufacturer using CBC data from N=716 respondents. The distribution of portfolio results suggested that the manufacturer's actual product line was potentially much larger than optimal and was missing two products that might be highly desired by consumers. Finally, we review the underlying computer code and its options. The model provides multiple methods of determining individual preference for the GA model along with various adjustable parameters.