Best Practices

We've had the K1000 for years (maybe since v 5.0ish...), and are currently on 6.2. A long standing issue we have after each update is that a small number of machines stop checking in (maybe 2 out of every 100) Upon investigation of the machines in question, we find that the old agent was uninstalled, but the new agent failed to install... So we manually install it via the file share and it checks in, and all is well again. So my question is why does this happen, and are there any best practices or strategies to prevent it?

Background on our environment...  We are Win 7 machines, K1 version 6.2. Agent is installed on our machines as a post install task during the K2 imaging process. The "Update Agents" is turned on via Settings/Provisioning/Update Agents on the K1.  When there is a new version, we let the agents update themselves and manually create a new post install task for use on our K2 images. I thought about using the Provisioning Schedules feature to schedule daily pushes of the agent to an IP range, that way it continually catches any machines without the agent. However in testing this, I can get it to push, and say successful, but then the next push fails because the client is already there, so it looks like the agent is not installed when it actually is...  So I didn't like that, plus I didn't want to put extra, unnecessary, strain on the box.  Hence...  The reason for this post, wondering if others deal with this and how.

Thanks!