Is it worth modeling these customers?

[kangaroo]

Let's say we have an order form. Customers fill out the online order form with their first name and last name; email address and phone number are optional. At a later point they can execute this order. At the time of execution, the customer first name, last name, phone number, and physical address are required; email is optional.

We have two business processes: orders and order fulfillment. We have a customer dimension on order fulfillment.

The business analyst says they want to be able to view a customer's order fulfillment percentage. This person also said he may want to view additional information with the brief customer information collected on the order forms. My question is how do we model these order customers -- we'd either have a dimension with 800 records of "john doe" or a dimension with 1 record of "john doe" that means nothing since they are not the same person? So, should we even model these order customers since there's really no way to always make this connection to the order fulfillment customer; for example, there is no way to connect a customer who fulfilled an order one time without an email address as the same customer who placed an order a few times without fulfilling the order and providing just a name (and sometimes an email address)?

We do have a loyalty program. Where their loyalty program ID is provided on the order form (if available). I've tried to explain that this is the point of the loyalty program. What should I tell this business analyst? That we can determine a customer's order fulfillment percentage, but it won't be 100% accurate since email address is not required on both orders and order fulfillment?

[Jeff Smith]

2 choices - create rows in your Customer dimension to reflect the information as it is entered. Add only the complete information into the Customer dimension and have a record in the dimension you assign to orders that have bad Customer information.

I prefer to add everything to the Customer Dimension whether it's complete or not. At least this way, you have the data as it existed in the source system. You never know how someone is going to use this data. You may develop the ability down the road to map all or some of the partial customers together. At the very least, it gives someone the ability to analyze how complete the data is and what might be down to improve the quality and completeness.

One man's garbage is another man's gold.