Models for making good decisions
Current computational approaches to decision-making suggest solutions for the best outcome on average. For example, standard tools could produce a call center workforce schedule that results in a small number of unhappy customers on an average day. However, the schedule may not help the center be equipped to deal with days that have an abnormally high number of calls, and therefore an abnormally high number of unhappy customers. This may be an unacceptable risk for the manager.
Industrial and Systems Engineering Assistant Professor James Luedtke is working to develop new algorithms in a field known as stochastic programming to specifically address uncertainty in decision-making settings, while allowing for individual preferences for risk. These new algorithms, which will address constraints that limit the probability of bad outcomes, will offer alternative solutions to the best-on-average solutions produced by current models. “When you aren’t making a decision thousands of times, being best on average doesn’t matter to you,” explains Luedtke. “If a decision-maker has just one shot, she may be willing to give up making a choice that is best on average to reduce the risk of being one of the bad outcomes.”
Luedtke’s methods could have a broad range of appliations in fields such as medicine, business and finance.