"lost bits" is a bad heuristic - all microphones, recordings, playback electronics have a Analog noise floor - human hearing has a noise floor, your room has noise, you can't even hear as deep into the human noise limit while wearing headphones
and for a good understanding of noise specs, audio consequences it helps to add in frequency to the description since human hearing has considerable frequency response variation for noise threshold
as far as Psychoacoustics has tested there is a limit below which added noise, even correlated quantization becomes inaudible
24 bits is already way more than needed for keeping lsb, quantization effects below the analog noise, your hearing limits in any reasonable gain structure system
the basic audio DAC function doesn't necessarily have any volume control - mostly done in digital processing in front of the DAC, with today's high integration, upsampling processing on the chip some digital volume control is often built in but commonly doesn't have sufficiently fine step size to avoid "zipper noise"
digital volume control in your digital source software should be done with adequate resolution and dither - but 32 bits may be used just because that's a common modern processor/standard computer math word size