The word optimisation has been used so much in software marketing that its meaning has worn off. What a glass optimisation program does is actually plain: it tries a large number of ways to place the parts on a sheet and picks the one that uses the fewest sheets. There is no magic, there is search. Knowing that sets the expectation correctly too.
The limit of the search comes from how glass is cut. The cut is guillotine, meaning every cut crosses the sheet from edge to edge. That makes the solution space far narrower than free placement: some layouts look perfect on paper but cannot even be candidates, because they cannot be cut. The job of a sheet optimisation program is to find a good arrangement under that constraint.
The number of attempts and the time spent are adjustable. The engine produces candidate plans for a given period and returns the strongest candidate it has found. As the time grows the result improves up to a point and then stops, because once every part fits on a single sheet there is nothing better left to find. Beyond that point different arrangements give the same result.
A misunderstanding is common here: the utilisation figure is taken for a quality measure. In fact, if all the parts fit on one sheet the utilisation is fixed and does not change whatever the arrangement. What should be compared is not utilisation but the number of sheets used and whether the remaining waste is reusable.
It is worth saying plainly that the engine is search based, because exaggerated descriptions circulate in this field. There is no learning model here; there are layouts tried, results measured and a best candidate chosen. That is how the inside of a glass nesting program works, and being understandable makes the result easier to trust.
Blade allowance and edge trim are part of the arithmetic too. Every cut takes a few millimetres and the border of the sheet cannot be used; when neither enters the calculation a plan that fits on screen does not fit at the bench. These parameters arrive with defaults, but it is possible to see how the result changes when they are adjusted, and running the same job with two different allowances takes seconds.
Finally, the input to optimisation is the order itself. The measurements are not typed again somewhere else; the engine takes whatever the order record says. That wholeness is what is meant by a cutting optimisation program for glass workshops: when the place the measurement is entered and the place the plan comes out are the same record, nothing is left to get lost in between.