The objective of the GOALI award is to integrate models of customer behavior and product design constraints in a multidisciplinary design framework to simultaneously develop marketing strategies and design families of feature-based products. The relative effectiveness of inferring market segmentation from discrete and continuous representations of variation in customer tastes will be assessed in terms of impact on both solution quality and computational tractability. Information on product design constraints will be used to develop strategies for simplifying the optimization problem formulation. Heuristic multiobjective optimization techniques will be applied to determine both the optimal number of product variants and the configuration of each variant. The uncertainties inherent in new product technology deployment will be accounted for by estimating the likelihood of diverting customers to other variants in the product family - or to competitors' products - when the feature content of the optimal solution is changed.
If successful, this research stands to redirect current thinking on optimal market-based product family design. This integrated and context-specific approach offers inherent advantages in balancing breadth of market coverage against operational expenses incurred in engineering and manufacturing a diverse lineup of product variants. This proposed work is particularly relevant to American manufacturing enterprises as they can no longer afford to implement exhaustive strategies to product proliferation. Instead, they must be strategically selective in targeting products to well-defined customer groups with strong motivations for purchase. Additionally, the dissemination of this work through creation, circulation, and ongoing development of a comprehensive library of case study problems will benefit the entire engineering design research community. Collaboration of academic and industrial partners is essential to the project?s success as the complexities encountered through real-world market conditions and design constraints could not otherwise be identified and addressed.
Each consumer has a unique set of preferences that leads to different wants and desires. A product’s customer-perceived value is driven by product specifications, which along with price drives market share. To accommodate market heterogeneity, a significant product design challenge becomes balancing the need to offer product variety while controlling the costs of product development and production. So that opportunities for marketplace success are not forfeited, designing competitive products requires correct representations of market heterogeneity and the ability to explore the large number of possible solutions. To offer some perspective, an automotive feature packaging problem can involve hundreds of available features and hundreds (or thousands) of product variants. For American manufacturers competing with those in developing nations who enjoy inherently favorable cost structures, product lines must be designed in such a way that the number of offerings is minimized, and that the offerings are configured to maximize both profit and market share. If incorrect model forms are used to represent the preferences of the market, a market-based product design strategy may be unsuccessful. Results from the work completed under this award suggest that hierarchical Bayes mixed logit models capture a wider range of preferences than latent class multinomial logit models. High levels of positive correlation observed within class part-worths estimated by the latent class model appear to mask preference heterogeneity, leading to product lines with significantly less variety. Further, the hierarchical Bayes mixed logit model had greater prediction accuracy when used with validation data, indicating that this model form is more representative of actual respondent preferences. Outcomes from this work also suggest that the hierarchical Bayes mixed logit model is capable of estimating accurate respondent preferences even when non-compensatory behavior is exhibited. If validated after additional investigation, this outcome could signify that American manufacturers can use this model form as a robust tool for almost all product design problems. While the rich representation of heterogeneity obtained from a hierarchical Bayes mixed logit model was found to be effective at describing the marketplace, significant computational challenges when optimizing product lines also had to be addressed. The combinatorial nature of this problem led to billions of possible solutions that had to be navigated by optimization algorithms. Non-tailored optimization approaches were shown to be unsuccessful at interrogating entire regions of the problem space. The work completed during this award addressed this computational bottleneck to ensure that the most effective design solutions could be discovered in reasonable amounts of time. A targeted initial population strategy was developed for heuristic optimization techniques that provided a novel and effective way of bridging marketing and engineering information. Providing a better starting point for the optimization algorithm led to significantly reduced computational costs while improving solution quality. By extending this approach to accommodate pricing variables and multiple objectives, this work improves the realism and plausibility of using such optimization techniques to develop product lines. Adoption of this approach by the industry partner led to reduced run-times (by several days) and produced solutions that were undiscovered using previous techniques. Computational challenges associated with product line optimization were also addressed by exploring the role of design constraints and how they should be handled and represented. Initial efforts explored if such constraints could be handled solely in the market research domain. Results of this investigation showed that models of customer preference estimated with only feasible product combinations were able to arrive at a solution with reduced computational effort. However, these models were not able to consistently ensure a feasible product line in all cases. Rather than focusing on a single domain – market research or engineering – such constraints need to be considered jointly. That is, constraints must also be included in the optimization phase. To reduce computational costs, the optimization should start with a feasible starting point and repair the solutions on-the-fly to maintain feasibility. The outcome of this investigation highlights the need for increased integration between a firm’s market research and engineering activities if customer preferences are to be accurately estimated and the solution space is to be effectively explored. By combining all outcomes of this work, American manufacturers can make better informed decisions about their product proliferation strategies. In doing so, American companies will be more competitive on the global scale and consumers will get products that they are more satisfied with. As part of this award, three graduate students were supported. This support led to the completion of one MS degree, one PhD degree, with an additional PhD degree still underway. Further, support from the industry partner supported one MS student for two semesters, and internship opportunities with the company for two other students. This work has led to the publication of 8 conference papers, 1 journal paper (with 5 under review), and the presentation of this work to over 400 industry practitioners.
