Estimates are always the elephant in the room. Especially large scope estimates. I don't think estimating more accurately is the solution, making software more

Estimates are always the elephant in the room. Especially large scope estimates. I don't think estimating more accurately is the solution, making software more predictable to write in the first place might be worthwhile. But the realistic solution is to manage around the practical reality that software estimates suck. Maybe don't plan release dates or milestones based on estimates, perhaps you could do it based on progress reports and a revised estimate after someone has actually started working on a task.