Introduced on day 1
Prescriptive analytics
Analytics that recommends an action: it combines a prediction or estimate with the costs of each error and the constraints, and outputs what to do - for example how many seats to oversell on each flight.
In detail
The four analytics families answer different questions: descriptive (what happened), diagnostic (why), predictive (what will happen) and prescriptive (what should we do). Only the last one ends in an action.
An overbooking limit is a prescriptive output. It needs a no-show forecast with uncertainty, the cost of a denied boarding and of an empty seat, and constraints such as cabin capacity and the airline's tolerance for denied boardings. Change the costs and the recommended limit changes even if the forecast does not.
Prescriptive recommendations about interventions inherit every causal assumption behind them. A pricing recommendation built on a demand model that confuses price with expected demand will recommend the wrong price, however good the optimizer is.