Hi Ashish,
I'm not an AMS expert, but I guess the reason may be similar to block level budgeting done at top level:
For example, in backend implementation most of the time a top level is partitioned into blocks in order to cope with complexity. During partitioning, timings at inputs & outputs of the block are budgeted as a rough estimation. However after implementation of blocks finalized, and you try to plug them at top level one can encounter timing problems, which were impossible to see at block level (better budgeting means less deviation from this). Making an analogy with this, one of the reasons maybe to verify co-habitation of analog & digital blocks in addition to running simulations with "behavioral" models, and see whether the circuit behaves exactly it is designed for. Otherwise you'll never know if there is any issue at the analog-digital interface, as behavioral models are perfect models not including any timing/transition information which are very important especially after circuit layout is done.
As I said previously, I'm not a mixed signal expert, so my answer is based on an analogy. Hope it helps.
BR,
Gokhan
---