Three-dimensional neutron depolarization (3DND) is a powerful but uncommon technique to characterize the magnetic microstructure inside bulk materials on a (sub)micron scale. The rotation of the neutron polarization measures the fraction of magnetic phase, while the shortening of the polarization provides a measure of the average size of magnetic particles. In this work we used 3DND to monitor not only the ferrite fraction but also the ferrite grain size during cyclic austenite-to-ferrite and ferrite-to-austenite transformations in Fe-0.25C-2.1Mn (wt.%) steel. These outputs from 3DND give an estimation of ferrite number density and confirm that there is negligible new nucleation events during cycling. Therefore cyclic experiments provide direct observation of austenite-ferrite interface moving. The movement of the interface in the present sample show that higher Mn concentration significantly retards interface moving and leads to a transformation behaviour distinct from that of a previous lower alloyed (< 0.5 wt.%) steel. To couple the microstructure evolution revolved by 3DND with the physics of the interface (e.g. mobility and partitioning of solutes), a computational cheap 3D model under mixed-mode interfacial condition has been developed to predict ferrite fraction, ferrite grain size distribution and interfacial concentrations. The results show that the combination of 3DND measurements with modelling 3D microstructure evolution model provides a versatile tool to analyse the intrinsic parameters of austenite-to-ferrite and ferrite-to-austenite transformations.