After time to time experiments with background inputs, it looked to me that it's much easier to test different messages first rather than expecting a specific solution to just work.
The largest hinder was often incorrect window handle, but sometimes the program requires a specific sequence of inputs and won't react on the very first attempt.
This is not the best way to answer, but I did a python tool which specifically can send multiple commands to multiple handles at once (and shows all possible handles too) for easy initial tests if the program will react on anything at all.