There are two approaches to this case (and probably even more):
- model your exception handling in the process, so whenever something wrong happen you will trigger an error event and then model catch error event that will handle exception flow in the process - in my opinion this is recommended so business process covers all scenarios and not only positive flow so to say
- enhance your work item handlers to deal with exception and signal events that are awaiting error situations
Both most likely will provide you with same result but option 1 in my opinion is clearer as that is visible directly on the process level and not somewhere in the code.
HTH