When we evaluate the Feynman diagram containing a loop, we commonly use the identity: \begin{align} \frac{1}{A_{1}^{m_{1}} A_{2}^{m_{2}} \cdots A_{n}^{m_{n}}}= \int_{0}^{1} d x_{1} \cdots d x_{n} \delta\left(\sum x_{i}-1\right) \frac{\prod x_{i}^{m_{i}-1}}{\left[\sum x_{i} A_{i}\right]^{\Sigma m_{i}}} \frac{\Gamma\left(m_{1}+\cdots+m_{n}\right)}{\Gamma\left(m_{1}\right) \cdots \Gamma\left(m_{n}\right)} \end{align} to combine denominators of propagators, and make shifts in the loop momentum so that the denominator has a form: \begin{align} \left[\ell^{2}-\Delta\right]^{m} \end{align} where $\ell$ is the loop momentum, and $\Delta$ doesn't depend on $\ell$.
However, I'm not sure if this procedure is always valid. Would someone know the conditions for shifts to be allowed or give me an example we cannot make the shifts?