Addresses in Customer Dimension

As part of an Orders dimensional model, I am attempting to design the Customer Dimension table. The customers have multiple sold-to addresses and multiple ship-to addresses. While the OLTP allows any Order to have any combination of sold-to and ship-to addresses related to that customer (many to many) in practice that is not the case very often. Usually (but not always) when an item is sold to address A, it ships to address C.

The question is: Is it better to have a separate dimSoldToShipTo table (including a role column) and exclude that information from the dimCustomer, or to include the SoldTo and Ship To addresses (multiple combinations for each customer) in the dimCustomer table? The understanding is that either or both would be linked to the fact table via the appropriate surogate key.

The number of customers is usually less than 20,000. While the capacity exists for 9999 ship to and 9999 sold to address per customer, typically there are less than 10 sold-to and less than 10 ship-to addresses per customer.

If, other than address, the customer information is common across addresses, I tend to have separate customer and address dimensions. The address dimension would contain one row per unique address.

Fact tables would contain FKs to both with appropriate roles as necessary. If, for dimensional reporting, you need to track addresses for a customer, I would have a factless fact table with customer key, address key, address type (degenerate dimension indicating role... ship to, bill to, etc.) and effective and expiration dates.

Agree with ngalemmo. I'll also add that most businesses don't really report at a street level. They're usually at a higher level, zip + 4, zip, or city. I would investigate this need and add additional dimensions as needed.

Thank you very much. I appreciate both of your responses. Now I can go down the right path.