How much research do you do before you implement a feature?

How much time you spend researching possible solutions and technologies to a problem before you decide on the best approach? How do you know when you know enough to make a decision?