Normally everything in mapreduce/hive would be a aggregate query but you can have non-aggregate queries. That would just be a query that had no "reduce" operation.
Hive can do a fairly complex query using multiple queries and windowing functions, etc. So, not sure the statement "mapreduce is suitable with simple aggregate queries" is completely true.
The types of business queries not appropriate for mapreduce/hive are real time queries. For example, trending query such as the top hash tags for twitter, etc. The overhead would make them inefficient.
Or if the data has to be normalized for some reason, the mapreduce/hive requires them to be basically be in one table. For example, if you had a highly normalized "point of sale" database and want to do any sort of query that would be painful unless you denormalized the data first.