You are right that there is no way to control the sound after it leaves the speaker. However, we can measure the effects of the reflections that arrive after the direct sound by looking at the time domain response. It has a certain pattern to it that will depend on the time of arrival of those reflections. Based on that pattern and the similarity of patterns across multiple measured locations we can identify the problems caused by reflections. Then, a filter is created to invert those problems as best as possible.

The key is to not think of the filtering in the time domain. It's not like a graphic equalizer that can only raise or lower the amplitude at certain frequencies. Our filters are in the form of impulse responses that operate on the audio signal through an operation called convolution. As such, they are affecting the signal in each channel in the time domain--hence the name. An additional benefit of such filters (also called FIR) is that they operate in the time-frequency domain as well. That means they can be used effectively to lower the ring down time of room modes in the low frequency range.