This is hard! For me, it's a mix of having a good understanding and confidence that you have
(1) a clear hypothesis that you can test with a minimally viable product that is shaped by data and customer/market research,
(2) confidence that you have a potential solution that can prove the hypothesis correct, and
(3) an understanding of the risk and opportunity for building that solution, including the time it'll take to build, the availability of users willing to try your solution.
When in doubt, it's always a helpful rule of thumb (in my opinion) to simplify the scope of your solution as much as possible, to both reduce engineering time/complexity AND to not squander the opportunity cost of shipping too late (see the Reid Hoffman quote question posted as well!). The hardest one in my opinion is 1 - making sure that you get down to the real essence of an unmet customer need and being incredibly clear and specific on how you think your product can solve it, and how you will know.