Imagine you're assigned to look into the slow decline in response time of a production application, which is deployed across a Managed Instance Group consisting of five instances. What approach would you take to diagnose this issue with minimal overhead?