poor data modeling, or data modeling that is well suited to the needs of the application, but not to the needs of analysts
Great addition, and a very important point. Analysts are expected to use the exhaust and incidental output of applications to try and generate insight instead of their needs being better considered at the design stage.
This is all within the context of analysts also needing to reconcile across multiple source (SaaS and other) systems leads to very brittle pipelines, and consumers downstream completely dissatisfied.
This highlights the primary issue for most tools in the data analytics space: poor data modelling, leading to the exposed data being "too difficult to use in Business Tools".
I've gathered some thoughts [1][2], but it boils down to the following: What users experience as the problem, is probably not the cause of the problem. Most builders respond to the user's poor experience at the interface as sufficient signal to start building solutions that improve that interface, while the underlying cause is deeper within the system (and the primary method for improving that underlying, currently, is data modelling).
Here [3] is a snippet towards an example of Facebook building a solution that addresses the complexity issue. Also common for ex-FAANG engineers to their surprise, discover that none of this type of infra exists outside of FAANG.
[1] https://news.ycombinator.com/item?id=33804938
[2] https://twitter.com/rdrn_/status/1604528546784870402
[3] https://seattledataguy.substack.com/p/a-zero-etl-future#:~:t....