The pipeline we built is fairly fast and robust. Depending on the nature of your service, you may take an optimistic approach with regards to the quality of the code the team produces and always deploy code to production. With good monitoring around your logs and application metrics, you will be able to catch issues minutes after the code is deployed. You will then rely on CodeDeploy and its ability to deploy older releases quickly to recover from that situation.
If you take that approach and a problem is detected, simply roll back. You may know exactly what's wrong and know that it's an easy fix, but knowing that there is an ongoing issue impacting users usually produces enough anxiety to make more mistakes and make the situation worse.