Having recently categorized seven different kinds of database, let me now make a similar effort for business intelligence. To a first approximation, I’d like to split BI use cases into 2×2 = 4 groups, along two dimensions:
- Is there an operational business process involved?
- Is there a focus on root cause analysis?
That could lead to the categories:
- Both operational and root-cause: Real-time monitoring (or more precisely human real-time).
- Operational but not root-cause: Operational BI without that monitoring aspect — e.g., checking whether an expense submission makes sense.
- Root-cause but not operational: Investigative/exploratory BI.
- Neither operational nor root-cause: Monitoring BI without an immediate operational aspect — e.g., checking the dashboard periodically.
Those, in turn, could be more descriptively named:
- Real-time (preferably with a modifier such as “quasi” or “human”).
- Tactical (or task-oriented/-centric).
- Investigative (or exploratory).
- Traditional BI.
To complete the list, I’ll add a fifth category, as explained below.
Notes on those categories include:
- Real-time BI generally is based on dashboards, although push alerting also plays a role (and could play a lot bigger one if the technology got better).
- The whole point of task-oriented BI is to tie it to an operational application, in lieu of a dashboard. This can be wholly automated, e.g. via a rules engine, or have a pop-up interface for a human who’s in the midst of using the operational app.
- Investigative BI is generally launched from a dashboard. Flexibility of drilldown is a key differentiator.
- Traditional BI has been all about dashboards for years, although that’s changing some due to the new mobile form factors.
- At the higher end, third-party BI can be delivered via dashboards. Often, however, it just looks like a single parameterized report. (E.g. Google Analytics.)
The problematic category in all this is traditional BI — i.e., BI that you sort of just look at and don’t do a whole lot with. In an actual categories graph, the “traditional BI” quadrant would be the lower left-hand lame one. So let’s soften that a bit and split traditional BI according to three kinds of user set:
- Enterprise-wide. This is the area where traditional BI really is lame. There’s no way to take a few metrics, sprinkle them across the whole company, and magically transform your business. Yet that’s pretty much the marketing pitch the BI industry used for years.
- Third-party. If you’re dishing out information to third parties, they may be satisfied just to get it, and shallow BI technology may suffice. So I think it makes sense to split this out as a fifth category.
- Departmental. Assuming your organization has some kind of enterprise-wide BI — and by now most do — then why would you want separate departmental BI? Four reasons are:
- You want to do better analysis — i.e., you’re headed into the investigative zone.
- You want to pull in some kind of data that your corporate BI standard doesn’t handle well. In that case, you’re probably in either the investigative or the third-party area.
- You want to meet some operational need. Once again, that takes you outside the realm of traditional BI.
- Not only do you want lame BI — for some corporate political reason, you want your own particular brand of lameness. In that case, you really are in the realm of traditional, departmental BI.
Summing that all up — and subject as always to Monash’s Third Law of Commercial Semantics — I’ll go for now with a five-fold business intelligence split:
- Real-time BI (preferably with a modifier such as “quasi” or “human”)
- Tactical BI (or task-oriented/-centric)
- Investigative BI (or exploratory)
- Traditional BI (which is lame)
- Third-party BI
- Business intelligence industry trends (February, 2012)
- BI served to third parties (February, 2012)
- Some issues in BI (mobile, departmental, etc.) (January, 2012)
- “Operational” vs. “investigative” analytics (November, 2011)
- Reinventing business intelligence (May, 2009)
- BI segmentation, and an anti-dashboard rant (October, 2006)
- More anti-dashboard ranting (November, 2007)