Why do we replicate data? Would it not be a good idea to keep one database of customers, products and other infrastructure data? The answer is not a short one. Based on the transactions performed, the requirement for  data quality may vary. For example, transactional data tends to contain late-coming data values or absence of data. If the customer provides an incorrect phone number or does not know the ZIP plus four code, we still need to create a transaction or pending-transaction. Thus data is captured with some compromise.

On the decision support side, we may live with less-than-perfect data when generating a marketing list. However if a specific question about a customer is asked, the ad-hoc query may need better quality of data. OLAP analysis is interactive-- you find out a fact and then dig in more for details or conclusions. If the data quality is bad, it may mislead the what-if analysis when performing drill-up or -down exercise. When it comes to data mining, we are asking the data to tell us rules we are not aware of. If the data quality is bad, we will find incorrect patterns.

As we improve the data quality, we need to coordinate snapshots, standardize value sets and transformation rules and keep more trending rules. This results in different degrees of data replication. Data warehousing is the technique and technology to replicate data for various transactional and analytical needs. When to do it? How to do it? To what degree do we replicate data? Data architects need to evaluate the situation and apply precise disciplines to ensure long-term gains and fulfill short-term needs.