Indeed, Andersen-Lake method is much faster, but is limited to vanilla problems.

The finite-difference covers wider range of problems, including stochastic volatility models, like SABR or Heston.