One rule of thumb I've found when constructing interviews is take how long it would take you (or any other member on your team) to do the problem cold and then

One rule of thumb I've found when constructing interviews is take how long it would take you (or any other member on your team) to do the problem cold and then multiply that by 2 to 3x. This gives two main advantages. The first is you have plenty of time for a conversation. Even assuming ideal conditions it takes between 2 and 3 times longer to walk someone through a solution as you implement it than just to incept an idea. Second is the interview platform is all ready stressful and deeply unnatural so building in some buffer for that necessary context switch helps you extract more signal. That all being said, if you beat a 2% rate on your total funnel you're doing pretty well. Most people just aren't going to be a good fit; that's just the reality of this industry with huge choice for the candidate (you almost certainly will not be anyone's first pick; the numbers are against you), near limitless specializations, and no real qualification body.