To understand whats really going on
check out this link that talks about the properties of matrix multiplication and consider especially the non commutative property, A*B != B*A, and the associative property, (A*B)*C = A*(B*C).
Now consider you are looking to scale, then rotate, then translate a point P with matrices S, R, and T
running this as 3 separate operations is equivalent to ((P*S)*R)*T, thus it's the same to pre calculate the combined matrix of S*R*T and then multiply P by that.