Back to the Future: Where Dimensional Modeling Enters the Modern Data Stack

dbt’s powerful capabilities allow data teams to deliver data products and analytics solutions to solve business problems faster than ever. Yet still, even with the best modern technologies, challenges arise. How can you be certain what your building will stand up to changing requirements? How can you connect disparate parts of your business to derive new insights? The answer may be a blast from the past-but the fundamentals never change. Learn how to apply fundamental techniques-like dimensional modeling-to modern tools, helping you to build scalable and reusable solutions to solve data problems today, and in the future.
Coalesce 2023 is coming! Register for free at coalesce.getdbt.com/.

[@ZachRenwickData]

Finally, a talk that actually provides us with solid arguments and examples of how and why we should use concepts of dimensional modeling. Really well done!

[@jerrynguyen7815]

Dave Fowler has to see this!!

[@larsoevlisen]

There is a lot of overlap with the points in this presentation and Data Vault. From focusing on data modeling as a business process focused project, to not creating a “wet” environment etc. etc. What does the presenters think about Data Vault as the base data modeling layer of business?

[@1988YUVAL]

Very interesting!

[@nosh3019]

great talk!

[@bfkgod]

I don't really see how someone could possibly argue against this. That level of process-driven modularity and low-level atomic data is not optional in my opinion. I genuinely think anyone who doesn't do this is wrong.

[@Abufaisal02]

Why not mentioning anything about the Inmon school?

[@tonydahlager]

There is value in normalization under certain circumstances, but we are promoting the value of Kimball's dimensional modeling. They are sometimes at odds with one another.

[@Milhouse77BS]

Not really a conflict. Inmon wants 3NF before the dimensional models, Kimball uses ETL to go straight to dimensional model.

[@tonydahlager]

I think this is only partly true. Inmon in his writings talks about aggregating when serving data to end users in a star schema. Kimball is very clear that overly aggregated data is one of the most fundamental mistakes a data warehouse designer can make. "Aggregated data in the absence of the lowest-level atomic data presupposes the business question and makes drilling down impossible." To me, reference to a shape of a star-schema does not fully encompass the differences in the principles of each approach.

[@hallkbrdz]

Good question - use whatever is needed. Store normalized, present wide-table and star.

[@mrRambleGamble]

Great talk but the first guy was way too quippy. Like a bad sitcom