People have automation backwards - automation doesn't solve problems. In order to automate, you have to solve all the problems. The machine has to know how to detect and handle any variation it encounters - or summon human help. If it has to summon human help too often, or fails to detect a problem too often, it is simply not worth it to automate.
I run into this all the time in my day job with software automation. Everyone thinks you wave some kind of magic wand when you say 'automate' that makes it just work. I have to put out the speed brakes once in a while when chatting with my VP to remind him that automation doesn't necessarily mean accurate, as much as it means fast. Sure, I can provision a bunch of network gear quickly, but if the automation is wrong, that just means I'll wipe out an entire portion of the infrastructure in seconds.
