Transport in models of the global atmosphere can lead to severe time step restrictions, particularly over the poles of a latitude-longitude grid or where convective updrafts are resolved. This has been solved before by the use of semi-Lagrangian advection and by the use of quasi-uniform grids such as the cubed-sphere instead of latitude-longitude. However semi-Lagrangian transport is not conservative: the transported quantity is not conserved. And the use of quasi-uniform grids does not solve the problem of time step restrictions due to strong convective updrafts. Better ways are needed to ensure model stability either where winds (or updrafts) are strong or where grid spacing is small. Implicit time stepping for transport is not popular in atmospheric science as it has been assumed to be expensive, complicated, inaccurate and does not scale well on large parallel computers. I will challenge these assumptions. Implicit time stepping can actually be simple to implement and, if it is used only where needed, then accuracy is maintained. Implicit advection is conservative, can be used with any grid structure of the sphere, can exploit improvements in solver efficiency and parallelisation and can provide stable solutions for large time steps. I will show how this can be achieve and will show solutions on pure transport cases and for solutions of the Navier-Stokes equations.