Having a display which 'jumps all over the place' is bad for user confidence and also bad for meeting the typical use-case of "Tell me: ABOUT what voltage is present?" in many cases.
"Once upon a time", and quite likely still, DMMs use(d) "dual slope" analog to digital conversion. This has the advantage of being able to achieve a large number of bits of accuracy (and/or resolution) at relatively low cost.
It has the disadvantage of tending to be slower to much slower per conversion cycle.
DMM's once (and probably still) used to often make the conversion period a multiple of the mains cycle time (60 or 50 Hz mains) to minimise the impact of mains AC noise.
AND if you have a somewhat suspect conversion system, not only averaging the result but doing some sort of sliding window averaging when the input signal is 'a little noisy' and/or the conversion process is pushing its limits somewhat will add apparent stability.
All these (or some subset) add up to a high resolution, OK accuracy stable display.
If you look in the spec sheet (that small sheet of close typed thin paper that often comes with meters) you'll find that on a 1999 display (which thus has 0.05% resolution), the accuracy is usually around 1% (ie merits a 199 display.)
On a digital display of actual voltage, the last digit "should" hunt between two adjacent digits, with the time on each proportional to the undisplayed remainder. eg a voltage of 1.9876 V on a perfect accuracy "three and a half digit" 1999 display should alternate between 1.987 V and 1.988 V. On avergae it should display the former 40% of the time and the latter 60% of the time as the undisplayed 6 is 6/10 of the way between 1.987 and 1.988. Many meters do NOT have the least significant digit "hunting" in this manner.