When the transactions are themselves little pieces of revenue, the boss can ask, "How much revenue did we take in?" by adding up the transactions. But in most businesses, there are many transactions that are not little pieces of revenue. Deposits and withdrawals in a bank are transactions, but they are not revenue. Payments in advance for subscriptions and for insurance coverage are even worse.

In the past year or two, data warehouses have become increasingly operational, increasingly fine-grained, and increasingly critical to day-to-day operations. In last monthýs column (see "Relocating the ODS," DBMS, December 1997), I argued that the old, separate operational data store (ODS) must now be brought directly into the enterprise data warehouse. Another way to say this is that the enterprise data warehouse will sub-sume the needs of operational inquiries and reporting anyway. Letýs call this the operational data warehouse.

When you take an operational view of our data warehouse, you find that there are some characteristic structures that are probably independent of your specific kind of business. Virtually every operational data warehouse needs two separately modeled versions of the data: a transaction version and a periodic snapshot version. Furthermore, the snapshot version has an additional special feature called the current rolling snapshot. This current rolling snapshot is the echo of the formerly separate ODS. The current rolling snapshot is properly integrated into the overall enterprise data warehouse. 

LINK