Anytime where you are trying to "collect" features of an array into one final value, and the collection function can be sensibly applied to the elements in any order as long as some interim value is passed along (e.g. summing, maximizing), the reduce paradigm is appropriate and can be more readable.

The bonus upside is that these paradigms also make parallelism quite simple, should it later be necessary.