A common pattern I see is deployment (or build) automation that routinely fails due to a set of common issues. When the deployment fails, people log into the box they were deploying to, perform the well known fix and get on with their lives. This is a trap that minimizes the value of the automation.
--Eric Minick, UrbanCode
Eric goes on to say that your strategy here should be to ban developers from logging into the affected server and trying to fix it manually. The first response should instead be to fix the automation to account for the new scenario that's causing the broken build.
You can read the full post here.