Effect of skewness in data

What model are you using?

Ideally, one should always scale / normalize data before feeding them into some ML / statistical model. By using Z-scores you should be able to control for skewness of your variables.

Data does not necessarily have to be standardized and mainly from the model, which we want to use.

Normality it's in many cases an asumption. In this situation normality means that the error between the predictions and the actual answers is distributed normally.