...looks like a kind of substitution.
It's perhaps more similar to encoding, say to ASCII, Baudot code or "dits & dahs" (morse code). And as the name implies, fractionation breaks up the message characters into smaller pieces (fractions) that then exactly represent those characters.
It was used in the more classical ciphers like transpositional ones. Probably before the common use of computing machines. The transpositions occurred on a character by character basis. The coordinates on a Polybius square can serve a similar function to that of an index into an S-box/lookup table. Or they can be otherwise manipulated in ways that a single character can't be. Since the advent of modern computers, this fractionation occurs automatically when the characters are represented as bits within a block of bytes. The term does not seem to appear in contemporary cryptography.
I suggest that the benefit is not for elimination of message redundancy, but for assisting with diffusion. Substitution and permutation are entirely mathematically possible with a human alphabet. You'd just have multi-character substitution boxes, so a three character substitution box would accept input like
ggn and output
yqy. You'd permute single letters between them as we do today. It would be cumbersome though as such an s-box would be at least 19,683 entries long ($(26 + 1)^ 3$). Numbers extra. And their efficient generation via polynomials would be problematic. Algebraic qualities, avalanche and linearity were not well appreciated then.
So now we fractionate/encode to numbers and bits, and manipulate, substitute and permute them instead.