I tell people to start implementing when they are pretty sure there aren't more important stories out there. An iteration's worth of data is worth months of speculation.