Several years ago a large European energy company was able, for the first time, to analyze the profitability of every commercial transaction on a consistent basis. This had previously been difficult owing to the different allocation of costs in the subsidiaries each country. They discovered that in one country there were a surprising number of low volume, low profit transactions, which was odd because generally you would expect that a higher volume deal would be less profitable than a low volume deal to the seller. (A buyer can demand a better price if buying 1000 units of something than buying 10). As a consequence the company raised its prices overall in this market. This reduced volume slightly by losing some marginally profitable deals but improved their overall profits in this market by 25 per cent. They were able to do this through the introduction of a federated enterprise data warehouse.

It goes without saying that complex international enterprises need consistent high-level management information and a clear view of how the business is performing. Unfortunately few companies have achieved this today. The usual approach to improve visibility into business operations is to standardize reporting structures and data models from the top down.

However, the standardization of business structures is entirely impractical for most enterprises. It costs too much, takes too long and is prone to failure because very few areas within a given business will be simple or uniform enough not to require local customization. While it may be possible to standardize (say) currency codes across an enterprise, completely standardized product hierarchies are rarely feasible, and may in some cases not even be desirable. The sales of a local confectionery product sold only in Japan would not be of interest at the global level, but the performance of the confectionery category as a whole probably would be.

Even if full standardization could ever be made to work, the processes for delivery consistency also impedes the flexibility of local operating units to adapt to changes in their local business environment. In the end a “central office versus local autonomy” war breaks out, and operating units end up developing “shadow” business intelligence solutions alongside the official central office one that they are obliged to feed data into. As the central system does not give them what they want, the incentive to check data quality carefully is low, and there is a serious danger of erroneous data being used back at head office to make important decisions. Somehow this “central vs local” logjam has to be broken, and a few visionary companies, such as Shell and Unilever, are doing so. These visionary companies have taken the approach of using a ‘federated’ model for enterprise data warehousing. This approach is one in which a hierarchy of linked data warehouses can exchange data, business models and reporting structures, to allow local autonomy and customization, but also deliver global control and a degree of standardization. A 'federated data warehouse' consists of a set of data warehouse instances that operate semi-autonomously, and are generally geographically or organizationally disparate, but which can be thought of, and managed, as one large data warehouse. Since a federated data warehouse can be built one step at a time, it offers a ‘start small, think big’ approach to enterprise data warehousing. The federated approach significantly reduces risk in a global roll-out, because each local warehouse is smaller in scope, delivers quickly on local requirements, and can be operated by local business units.