Flavors of Self-service BI

My Fall newsletter entitled “Why a Semantic Layer?” is out and it will be e-mailed to subscribers on Monday. The reason why I wrote is to bring the focus back on organizational BI. Based on my experience, decision are makers are confused about what their company’s BI strategy and roadmap should be. I’ve asked a BI manager recently if he knows that a semantic layer could be a true self-service BI enabler and he said he didn’t know. After talking to some well-known BI vendors and listening to their “pure” self-service mantra, the focus was shifted from a comprehensive BI architecture that truly addresses the company needs to selecting a tool that has the most visualization fluff.

In general, there are two self-service BI paths your organization can take:

  • Semantic layer + ad hoc reporting – If your organization decides to invest in implementing a comprehensive organizational BI framework (see the diagram in my newsletter) that features a semantic layer, the end user’s focus will be where it should be: on data analysis and not on building models. For example, end users can use Excel, Power View, or whatever third-party data visualization tool your company felt in love with, to connect to the semantic layer and create insightful ad hoc reports with a few clicks. No need to import data and build models before the business users ever get to data analysis.
  • Self-service models – In this scenario, you put the responsibility on end users to create models and their version of the truth. Following this path, an end user would use a tool, such as Power Pivot, to import data, create a model on the desktop, and then derive insights from the data. Don’t get me wrong. There are good reasons when this path makes sense, including the ones I mentioned in my blog “Does Self-service Make Sense”.

Ideally, your BI roadmap should consider and plan for both paths although the focus should be on organizational BI first and then finding gaps that self-service BI could address, such as testing ideas requiring external data. If your organization relies only on “pure” self-service BI, it won’t be long before you’ll find out that the pendulum has swung the wrong way. Many have taken this road but the outcome is always the same: “spreadmarts” that might get the work done in the interim but would fail in a long run. As a rule of thumb, I recommend the 80/20 principle, where 80% of the effort is spent on organizational BI (DW, ETL, semantic layer, dashboards, operational reports, data mining, big data, etc.) and 20% is left for self-service BI. But each organization is different so the ratio might vary.

Let me know your thoughts.