To use set analysis or not to use set analysis?

[#1]

I am a great fan of using set analysis, but is it always the right approach? On the plus side you can control the exact query that is being used by the end user to ensure there is no subjectivity around the output but on the down side are you under-selling the overall experience a user can gain by making their own selections.

I use a combination of both approaches now, usually a front sheet dashboard full of set analysis for static KPI's backed up with other more detailed tabs to allow the end user more autonomy in their experience.

What is best practice?

[petter]

Well you seem to have a very good practice. My experience is that it's always a balance between the more static quick glance dashboards and deeper flexible sheets where the user and analyst can do their own data/business discovery. Often different customers and different use-case demands a different mix. It is also a matter of training and enabling the users by providing applications that are intuitive and well-documented. Not only that but giving the basic and focused introduction and training to empower the users/analysts is extremely important.

Set Analysis or rather Set Expressions (which is the right term) is a very flexible and powerful tool. The challenge is to not over-use it and over-complicate by doing too much set expressions. Many times things that you initially develop with set expressions would later be better to move into your data model and simplify your expressions by using flags and general data model improvements.

I would say that logic that cuts across the entire application and data model would often be best to pre-calculate with various flags, attributes and dimensions at load time. More specific fine-tuned analytics that are specific to one or very few charts could be better to solve with set expressions.

Testing, testing and testing is very important when you employ set expressions make sure that you have some good test-data which will reveal logic problems in your expressions. Make sure to use Barry Harmsen's brilliant "clear-variable" function to clear out non-relevant selections. This can be very demanding to do manually when the complexity of your data model grows. This is especially relevant when you are using extended calendars.

Having one or more good quality master-calendars will aid any set expressions. Then you can rely on simpler and more maintainable set expressions.

A good data model which is almost always a Star Schema will always make it easier to take full advantage of your set expressions without over-complicating them.Be sure to test the set expressions step-wise.

It's a huge and interesting topic.

[#3]

‌Thanks for your reply Petter, as you point out, too much use could make an application intensive to maintain and great economies can be made by building an efficient data model with logic that is global across the app.

I'll definitely look into the "clear-variable", thanks for the tips!