Surrogate vs natural keys and smart columns

Some of my co-workers have proposed that we use natural keys in our slightly normalized data warehouse table structures. These table will be used to populate our fact and dimension tables. In example, one of our primary key structures would consist of a source system code (VARCHAR2(10)) and a policy key which is a smart column consisting of the policy number concatenated with the policy term effective date (VARCHAR2(50)). I am not sold on this approach. What are the cons of this approach? What are the cons of using natural keys? What are the cons of smart columns?

One of many Ralph articles on the subject