Rolling back first is the part people resist and the part that ends the outage. Debugging a live site is how a ten-minute incident becomes an evening. Works the same on Vercel, Netlify or Railway.
Workflow
Copied 5 times
1. Is it you or everyone? Open the site in a private window on mobile data.
2. Status pages: your host, your database, your auth provider. Two minutes, saves an hour.
3. What changed? Last deploy, last migration, last environment variable edit, in that order.
4. If the last deploy is under an hour old, roll back first and debug afterwards.
5. Read the real error in the function logs, not the friendly message in the UI.
6. Only then hand an agent the log line, and ask for causes rather than a fix.