Depends if you want to go the vendor route, where "cube" is very specific to the implementation of the tool you're implementing. Dealing with semi-additive measures and drill-down/drill-through will differ between SSAS, Pentaho, or Cognos.
I think the real value in the process comes from all the data modeling decisions, which involves a lot of interactions with the business users that are asking for data/reports. Something as simple as an e-commerce order can look very different between CEO KPI reports, marketing, purchasing, and accounting. For example, accounting cares only when the product ships while marketing cares when it's sold. Multiply this by a hundred, and you end up with nuanced data pipelines that encapsulate all this domain-specific logic.
Multidimensional modeling is very useful even if you're not adopting cubes specifically, and just using dbt on a read-only replica of your database to create aggregate table for a few dozen reports.
Thanks this helps. It seems like it shines when you have a lot of non technical staff requesting the data. At my job, the data analysis folks have to write complex ad-hoc queries all the time. We store the queries for consistency reasons (e.g. to make sure we can be consistent and run the same query each year for the same report) It seems like OLAP wouldn't be a good fit there.
Consider the cube as the end product of a process of automating data the business needs to make decisions. It starts with "How much did we sell yesterday?" and expands to include week-over-week/year-over-year comparisons, breakdowns by category, leaders/losers and other frequently looked at answers.
I think the real value in the process comes from all the data modeling decisions, which involves a lot of interactions with the business users that are asking for data/reports. Something as simple as an e-commerce order can look very different between CEO KPI reports, marketing, purchasing, and accounting. For example, accounting cares only when the product ships while marketing cares when it's sold. Multiply this by a hundred, and you end up with nuanced data pipelines that encapsulate all this domain-specific logic.
Multidimensional modeling is very useful even if you're not adopting cubes specifically, and just using dbt on a read-only replica of your database to create aggregate table for a few dozen reports.