How Important is Feature Engineering in Machine Learning?

Some say feature engineering can make or break your model. How critical is it in your experience? Do you have any favorite techniques to share?