First let's justify the relation, which using the model of functions to represent the states, says that
$$
\frac{\int \left(\phi^\dagger(x) + \epsilon \alpha^\dagger(x)\right)A(\psi(x)) \,dx}{\int \left(\phi^\dagger(x) + \epsilon \alpha^\dagger(x)\right)(\psi(x)) \,dx} - \frac{\int \phi^\dagger(x) A(\psi(x)) \,dx}{\int \phi^\dagger(x) (\psi(x)) \,dx}
= \epsilon \frac{\int \alpha^\dagger(x)A(\psi(x)) \,dx}{\int \phi^\dagger(x) (\psi(x)) \,dx} - \epsilon \frac{\int \phi^\dagger(x) A(\psi(x)) \,dx} {\left(\int \phi^\dagger(x) (\psi(x)) \,dx\right)^2} \int \alpha^\dagger(x) A(\psi(x))\, dx + O(\epsilon^2)
$$
Using perturbation methods, expand (for small $\epsilon$) the first term on the left using the usual trick of multiplying the denominator by
$$
\frac {\int \left(\phi^\dagger(x) - \epsilon \alpha^\dagger(x)\right)(\psi(x)) \,dx}{\int \left(\phi^\dagger(x) - \epsilon \alpha^\dagger(x)\right)(\psi(x)) \,dx}
$$
This rearranges into the $\epsilon=0$ piece being subtracted off, plus the expression on the right of your relation. However, to get this rearrangement, two properties of the functions $\phi, \alpha, \psi$ are required:
All three must fall off sufficiently rapidly that the needed manipulations of the integrals are not rendered meaningless.
Unless I have done something wrong, I need to assume $\alpha$ commutes with $\psi$ (or, it turns out, with $\phi$) in order to manipulate to get that nice expression.
So when you have an operator which commutes with the test states, you relation gives an expression for the derivative of the matrix element for that operator. But I think you have not seen this relation much because it is wrong for operators in general, and it would be "trappy" to blithely point it out with the caveate that you must be careful to use it only in the commutative case.