It eliminates bias by centering the distribution at $0.5$ regardless of the observed data.

It provides a more conservative variance estimate near the boundaries of $0$ and $1$ by adding pseudo-observations.

It forces the distribution to be exactly normal for small $n$.

It replaces the standard error with the exact Fisher information.