Elitism is basically required to preserve some members of current population ( Usually the best ones) to move onto the next generation without any modifications. 

When crossover and mutation operators operators will be applied to these candidate solutions, then these potential solutions might result in better efficiency and optimality of the algorithm.

If we skip the elitism process, then the accuracy and efficiency of GA will decrease and it will take more time to reach an optimal solution, because we are not preserving the current best solutions. The fitness of these solutions might decrease after crossover and mutation operations.

It might eventually reach an optimum global, but its time complexity will be very high.