It may help more to define more precisely a few scenarios where you experience problems, so people can be more specific with suggestions.
Lacking more specific information, you might consider putting automatic self destruct / deprovisioning for created cloud resources, e.g. 7 days, so you don't worry as much about what is created. What also helps is idempotency in these scripts, to just tell developers to retry things if there are transient errors.