Same here. I do mainly Python development - Django docs made things way easier then the Perl frameworks that I tried.
But for quick and dirty scripting tasks I still pull out Perl as the extra sysntax makes thing so much easier - backticks and built in regular expressions for example.
And I have still never seen anything as intuitive to use as a Perl hash tied to a berkley db file for simple persistence to disk.
I agree that Perl's autovivification is a timesaver, although I can usually hack together what I want with Python's defaultdict.
For "data analysis improvisation" though, I don't know. A good REPL is a key part of that, and I'm not sure that Perl has anything as good as the ipython cli or jupyter notebook.
The thing I find easier in Perl than in the other languages I might reach for when doing a lot of quick scripting tasks is declaring nested data structures. It's easy in Perl because often you don't have to declare them--you just use them. And that means that you can often make major changes to how your data structures work with just small source changes.
For tasks where you need to process data from logs or text files or similar, but don't actually quite know what it is you need to be doing with the data so need a good bit of exploring/flailing to get your bearings, and then maybe a few iterations trying different approaches to figure out what structures you need, and you don't have a lot of time to think about it up front--in other words, data analysis improvisation, Perl shines.