Preaching to the converted here, but this is one way in which I have found
conlanging useful for me.  My defense of the art.

I just sat down to do the first couple hours work on my next conlang, and
started by listing my goals for it:

1. make it sound natural
2. make a balanced use of space
3. figure out what's up with the natlang I'm documenting

These are essentially the goals I had for my last language, which was a
reimagined and relexified version of the language I was documenting at the
time, Michif.  I kept my phoneme inventory roughly the same size, and then
worked out how to fit roughly 120 suffixes (3 maximum at a time) into two
syllables' or less worth of space for every transitive combination of
animate and inanimate pronouns.

By the time I had a working draft I understood for the first time just how
fine a line a natural language walks, trying to be brief yet transparent,
while also making balanced use of acoustic space.  I realized that not
only does Cree or Michif create its entire inflectional system using a
reduced phonology, but that that space is perfectly filled, maintaining
maximum distance and clarity in a way that I hadn't previously understood
(despite having read everything I could get my hands on about the language
for years, and being half way through a masters in linguistics).

Today I'm working documenting a new language in a very different family.
It has a large number of lexical suffixes, almost all of which contain no
vowels and add on to already complex finals, and which can be themselves
followed by more consonant only clitics, suffixes and connectives. There
seem to be a million ways for this to be awkward and unwieldy, but it
sounds effortless coming from a native speaker.

Knowing that I need to figure out how all this works I could start by
looking at the answers (the language as documented so far) and try to
figure out the questions answered, OR I could go straight to the questions
by relexifying the language to the lowest morphological level possible.

While it might be seen as extremely labour-intensive, the approach has
definite advantages in that once you start trying to build a system that
works the complexities involved throw themselves in your face, one after
the other after the other.  As each question is answered, new questions
are created, whose answers depend on the answers invented for the previous

Based on my experience I believe that conlanging as a method of
understanding language in general, and as a method of understanding the
full system of balances at work in any specific language, is not only
useful but in many ways more direct than most other approaches to the
study of linguistics, and I am hoping that my next conlang will reinforce
my suspicions.  Sure it won't be a complete relex, and I'll probably try
to force in too much complexity out of sheer curiosity, and it likely
won't be near as smooth as I'd like simply because I won't have the
opportunity to revise and polish it for some time, but with every
shortcoming of my conlang I will find the questions to answers I didn't
even know I possessed, and for that I know the task will be worthwhile.

tldr, I use long sentences when I have a lot to say.