Migration Velocity Analysis is a seismic imaging method developed in the image domain, rarely applied on real data sets. The artefacts created by the classic migration and inappropriate choices of user parameters are the two main limitations. We propose here to replace the adjoint operator by the inverse and to apply a three-step workflow on a marine real data set acquired in West Africa. In the first two steps, we use the low frequencies to limit the numerical cost. The results of the first step are then used to determine the optimal user parameters such as the ranges of surface and subsurface parameters. We run new iterations to obtain a more consistent velocity model. In the third step, higher frequencies are introduced. We retrieve a geologically plausible velocity model, with an improved data fitting.