Both normal mode models and Fast-Field Programs are based on a contour integral representation of the acoustic pressure. Normal mode models evaluate the integral by residues which involves finding the poles of the Green's function. FFP models evaluate the integral directly by stepping along the contour.
In terms of efficiency, the question is whether one can find poles more rapidly than directly integrating. It takes about 15 Green's funtion evaluations to find a pole. The number of Green's function evaluations for an FFP model increases linearly with range.
On the other hand, the normal mode series neglects certain contributions which tend to be important in the near-field (say within 10 water depths). Indeed some problems have no modes at all such as the problem of a point source in free space. Also for very complicated problems (with elasticity) it can be difficult to reliably find the modes. Then an FFP model is a good alternative.
A complete explanation of when to use which model would require many pages. A rule-of-thumb is to use SCOOTER when you are concerned about the field within 10 water depths and KRAKEN otherwise. SCOOTER can be used for larger ranges but will generally require more CPU time. KRAKEN can be run for closer ranges but requires some insight in setting up the environment to make sure that the modes are adequate for describing the field. This is done either by extending the model of the ocean bottom in depth, introducing a false bottom, or computing leaky modes.