I've also learned it's a classic under-specified problem. To solve it completely you need, for example: 1) distribution of arrival times and destinations of pa

I've also learned it's a classic under-specified problem. To solve it completely you need, for example: 1) distribution of arrival times and destinations of passengers; (For example, if late evening people usually go up, it's best to idle on ground floor) 2) a "satisfaction function", giving a measure of performance of your system: You can get low average waiting times and diverging maximum wait time for a busy elevator, simply trade time from a single passenger for others (by e.g. never going to a floor). As long as utilization is significant at all times, this phenomenon exists. Clearly it's not only average we care about, but also maximum and minima of transit times. --- It's a very illuminating modelling exercise.