How do you choose a niche to run tests on to see if they are profitable or not?

How do you decide that you want to try a new niche in terms of actually 'discovering' the niche as it were to begin with?

Regards,

Lee