An extreme programming technique. If you do not know anything to make a prediction, use your own historic data to predict what will happen. You are allowed to learn and to adapt. Take your own experience and modify settings to tweak the outcome. This is a very simple PDCA cycle.
It is a little bit like differential calculus: To calculate the speed of something, you take the slope of the curve. If it is linear, it does not matter where you take it. If it is not linear, you can take the slope centered around the point where you want to know it. In mathematics that is fine, if it is a function of time and we want to predict future, well we can only take data from the past. They will be off, but if the curve we are looking at is smooth, we will be close. The same is true for acceleration (second derivative) of the curve.