Can you elaborate a bit on what you mean by rolling back? You can roll back any changes to the scenario itself from Version control.
If you are looking at rolling back the datasets that were potentially built or dropped by previous steps wouldn't typically be possible.
There are a few things you can do to reduce the possibility of a failure :
1) Don't run other steps in the scenario if a particular step failed.
2) Add checks and metrics before running build steps. E.g check if the input dataset has 0 rows
3) Add retries to particular steps where appropriate in case it was transient issue.
The following tutorial will provide some examples on how to use check and metrics in scenarios: https://knowledge.dataiku.com/latest/courses/automation/automation-hands-on.html