$L(\theta; x) = g(T(x), \theta) h(x)$ for some functions $g$ and $h$.

$L(\theta; x) = \frac{g(T(x))}{h(\theta)}$.

The score function depends only on $T(x)$.

$T(x)$ must be an unbiased estimator.