Postselected weak measurement is a useful protocol to amplify weak physical effects. However, there has recently been controversy over whether it gives any advantage in precision. While it is now clear that retaining failed postselections can yield more Fisher information than discarding them, the advantage of postselection measurement itself still remains to be clarified. In this Letter, we address this problem by studying two widely used estimation strategies: averaging measurement results, and maximum likelihood estimation, respectively. For the first strategy, we find a surprising result that squeezed coherent states of the pointer can give postselected weak measurements a higher signal-tonoise ratio than standard ones while all standard coherent states cannot, which suggests that raising the precision of weak measurements by postselection calls for the presence of "nonclassicality" in the pointer states. For the second strategy, we show that the quantum Fisher information of postselected weak measurements is generally larger than that of standard weak measurements, even without using the failed postselection events, but the gap can be closed with a proper choice of system state. . It involves weak coupling between the system and the pointer, but the postselection on the system leads to a surprisingly counterintuitive effect: the average shift of the final pointer state can go far beyond the eigenvalue spectrum of the system observable (multiplied by the coupling constant), in sharp contrast to the projective quantum measurement. The mechanism behind this effect is the coherence between the pointer states translated by different eigenvalues of the system observable, which has an enlightening interpretation based on superoscillation [2].