Thursday, April 12, 2007

Beware Distributed Analytics (You Read It Here First)

What might be called the “standard model” of business intelligence systems boils down to this: many operational sources feed data into a central repository which in turn supports analytical, reporting and operational systems. A refinement of the model divides the central repository into a data warehouse structured for analysis and an operational data store used for immediate access. (Come to think of it, I don’t see the term “operational data store” used much anymore—have I just stopped looking or did I miss the memo that renamed it?)

No one has been attacking this model directly (unless I’ve missed yet another memo), but it does seem to be under some strain. Specifically, much more analytical power is being built into the operational systems themselves, reducing the apparent need for centralized, external business intelligence functions. Examples abound:

- channel-specific analytical systems for Web sites and call centers (see last Thursday's post).
- more extensive business intelligence capabilities built into enterprise software systems like SAP
- rule- and model-driven interaction management components built into customer touchpoint systems
- more advanced testing functions built into operational processes (okay, we haven’t seen that one yet, but I did write about it yesterday).

Generally speaking, more analytical capability in operational systems is a Good Thing. But if these capabilities replace functions that would be otherwise provided by a centralized business intelligence system, they reduce the incremental value provided by that central system and therefore make it harder to justify. The resulting fragmentation of analytics is preferred by many departments anyway because it increases their autonomy.

This fragmentation is a Bad Thing, especially from the perspective of customer experience management. Fragmentation leads to inconsistent customer treatments and to decisions that are optimal for the specific department but not the enterprise as a whole.

The problem is that analysts, consultants, journalists and similar entertainers are always looking for new trends to champion. (Yes that includes me.) So coining a phrase like “distributed analytics” and calling it the Next Big Thing is immensely attractive. (FYI, a Google search on “distributed analytics” finds 64 hits, compared with 410,000 for “predictive analytics”. So the term is definitely available.)

We must all resist that temptation. A consolidated, centralized data view may seem old-fashioned, but it is still necessary. (Whether the data must be physically consolidated is quite another story—there’s nothing wrong with federated approaches.) By all means, analytical capabilities should be extended throughout the organization. But everyone should be working with the same, comprehensive information so their independent decisions still reflect the corporate perspective.

No comments: