Optimal Testing Strategy For Product Design Wei Wang Pengbo Zhang Apr /15/20091
Outline Introduction Pandora’s Box Role of Learning in sequential testing More complicated testing strategy 4/15/20092
Introduction Why are new product design testing needed? – Decreased market uncertainty – Optimal candidate solution What are sacrificed? – Direct testing cost (Wind tunnel etc.) – Indirect opportunity cost (testing time) How are tests carried out ? – Sequential – Parallel – Mixed Optimal testing strategy needed! 4/15/20093
Pandora’s Problem n closed boxes Box i contains reward x i with distribution F i (x i ) independent of all others x 0 initial wealth to start Cost c i to open box i with a time lag of t i Benefits discounted at rate r Pandora opens one box at each stage She can choose to stop searching and collect at that time the maximum reward thus far uncovered Wanted: A sequential decision rule that can maximize the expected discounted value 4/15/20094
Formulation 4/15/20095
Learning Erat, S., Kavadias S., Sequential Testing of Product Designs: Implications for Learning, 2008 Pandora’s Box assumes all testings (opening a box) are independent Interactions between design attributes and similarities between design configurations Outcome from testing one configuration relevant in understanding and predicting the performance of subsequent tests Similar DP formulation with correlation/covariance matrix added to model transferable learning 4/15/20096
Formulation 4/15/20097
Testing Strategies Loch et al. Parallel and sequential testing of design alternatives, 2001 Parallel testing Proceeding more rapidly Good when lead time is important Sequential testing Learning, lower cost Good when cost per test is important 4/15/20098
Formulation 4/15/20099
Formulation 4/15/200910