I know a true ETL solution like this will be more powerful and flexible, but how do you decide when it's worth the investment? Are there key use-cases or reports that only this type of solution can provide? Is the main issue with external analytics (like Mixpanel) that you need to know what data to collect before you can report on it?
The 2nd thing is that MixPanel is write-once, read-forever. aka if you misspell something or change definitions you end up in a bad place. These ETL processes are much more capable of adapting and normalizing a schema over time.