Spent a few years doing population projections for regional planning and the thing that always got lost was how sensitive the long run numbers are to small fertility and migration assumptions.

Exponentials, let alone super-exponentials, will tend to do that. As the article notes, the biggest issue is assuming that the system you're modelling doesn't have any negative feedback loops that would stabilise it.

The same is true of investment and retirement calculations. Little differences in the estimation of an exponent will compound to huge differences in the long run.