The last piece of the puzzle is the objective. To make good recommendations, the machine needs to know what value you associate with a potential set of decisions. This is usually straight forward: In the can design problem above, e.g., you want to minimize the total surface of the cylinder, which can be computed from the height and diameter of the can (do you still remember the math from 8th grade?).
Next, see how we put everything together and design the perfect can.