Conjoint Design Generation On-The-Fly

Many companies and marketers need to decide which features and characteristics to include in a product. This research develops a novel algorithm for Choice Based Conjoint Analysis On-The-Fly (CBC-OTF). The goal of CBC-OTF is to find a global optimal alternative among thousands or even millions of possible products with respect to an individual and population level criteria. We iteratively drop uninformative levels, features, from the design in order to quickly converge to the best configuration. A novel approach using Fedorov algorithm for faster and more efficient designs is used. By using an aggregated logit model we manage to get utility part-worths for the non-removed levels with very low computational costs and high precision. Repeated simulations show that the candidate global optimum reached deviates on average less than 15% from the individual optimal product. If a standard approach is used i.e. CBC with no level removal, the deviation increases to 35%. If optimality at population level is considered, precision is not very high, 35% of predictions are correct.