You could spend months or years to know that your approach is correct before trying it. If the cost of failure is high, say you're firing a rocket into space, then you want to put enough time into being sure you're right. But if it doesn't take long, and it's easy to revert, why not try it and see what happens?
I do this in programming all the time. Whether something works or not is data. I don't like to leave a solution in place that I don't understand, so I will usually do the work to know it's correct. But it's easier, less stressful, and takes less time if you already know it works.
And of course if it doesn't work, that's data too.