It allows the likelihood to be factored into a part depending on $\theta$ and $T(X)$, and a part independent of $\theta$.

It is the only statistic that can be used for estimation.

It always has the minimum variance.

It is a statistic that is always unbiased.