The algorithm complexity went down in the function they patched (6x improvement in their benchmark), but in the context of how they benefited with how they were using the algorithm the impact was much larger (improved to taking 1% of the time), which is plausibly exponential (and figuring out the actual complexity is neither relevant nor an economic use of time).
> figuring out the actual complexity is neither relevant nor an economic use of time
The fix was replacing a nested loop with a map. Figuring out that this goes from O(n^2) to O(n) (modulo details like bucket count) is immediate if you know what the words mean and understand enough to identify the problem and make the fix in the first place.