Gogula Aryalingam has started a series on managed self-service BI. Part 1 provides an overview of the topic:
When putting together a business intelligence strategy using Power BI, Microsoft recommends three primary strategies that an organization can adopt. Out of these, the one that I tend to go with is managed self-service BI, which brings forth the concept of discipline at the core, flexibility at the edge. This concept is the dominant strategy used for BI at Microsoft itself; explained very nicely in this article. It’s my personal favorite, because I find it an effective means of onboarding customers once the core platform is built with the required standards (discipline), and then help them adopt the solution from the edge, thus providing them with the best of both worlds.
Part 2 takes us to the edge:
Now, what happens when an analyst, for instance, has a set of sales target spreadsheets and wants to compare the figures with sales metrics so that salespeople’s performances can be measured? It certainly needs a new dataset. However, flexibility at the edge has to prevail in the right way. This post will look at how we can go about this keeping to discipline at the core, flexibility at the edge.
Note: The analyst’s requirement is at current local to their group or department. It has not yet been made an organizational requirement. That’s how most requirements start out: A requirement at the departmental level, and then when enough people start reaping the benefits within and outside of the department, it can get absorbed into the core.
Part 3 returns to the core:
One problem that we may have overlooked when building a bunch of core datasets in that post, is that certain dimensions tend to duplicate across the datasets. Imagine a scenario where the single master data source of a managed self-service setup is a data warehouse, which sources all the required dimensions. When you have, for example, core reseller sales, internet sales, and finance datasets, each one will have a calendar dimension and a few others created in each of these datasets. This is not ideal if you think about the extent of the duplication and effort that is required.
This is where, once again, using DQ for PBI datasets and AS comes into play, where you could draw up a layered core dataset architecture. If we take the example of AdventureWorks’ fact tables in the data warehouse (single master data source) you can figure out what the business processes are.
Read on for Gogula’s thoughts. I think there’s a lot going for this particular strategy, especially in a large organization with hundreds (or thousands) of people actively using Power BI. At that point, doing everything through a central IT organization doesn’t scale very well.