The 4472 uses A/D converters that are not particularly stable at DC. More than likely the offset has drifted with time and temperature and you're seeing the result (the spec indicates a max offset of 3 mV, coincidentally). Try running a self-calibration on the board; that ought to take care of the problem.
Usually a small amount of DC is not a problem when acquiring data with AC coupling. The purpose of the AC coupling is either to allow DC offsets which would otherwise exceed the input range or to suppress low frequency noise. I use AC coupling myself, for example, when I'm looking at power supply noise. The vast majority of the time, I'm using some sort of frequency-domain processing, such as a power spectrum, and DC is ignored altogether.
Offsets might cause trouble when looking at time-domain artifacts, particularly if you're counting on a residual DC level for triggering. In general, I would not expect DC offset performance better than a few mV on a dynamic signal acquisition card such as the 4472.
Hope this helps,
Ed L.