This was a very similar conversation that the team at my last job had. Our data were customer-facing so it was a slightly different problem, but the question of

This was a very similar conversation that the team at my last job had. Our data were customer-facing so it was a slightly different problem, but the question of "our application cares about what the present state is, but our analytics cares about trends and point-in-time" is a more universal problem. We used Debezium and Change Data Capture + Outbox patterns in some cases, and solely outbox patterns in others. (See https://debezium.io/blog/2020/02/10/event-sourcing-vs-cdc/ https://debezium.io/blog/2020/02/10/event-sourcing-vs-cdc/ ) - You then can use the change events to do the analysis. That solves your "log all actions" problem. It's still a lot of work to put together the analysis (a tool like DBT helps) but it is a solved problem with enough labor at that point and doesn't disrupt the actual application. If I was starting from scratch on an application today that I thought might have analytics needs in the future, I would get this early on even if I didn't need it yet.