We do know what kinds of situations genetic algorithms do well at (though generally other metaheuristics like hillclimbing do slightly better.) The fitness landscape has to be as smooth as possible and not have bad local optima. You want as few variables as possible. And you want them to be independently correlated with fitness. If an improvement requires multiple things to mutate all at once it's unlikely to happen.

There are various other heuristics. But you definitely need to have an understanding of the solution space (as well as the alternative) to know for sure.