Raw metric queries are hard to read during an incident, so you need a picture that shows health at a glance. Grafana turns queries from Prometheus, Loki or Elasticsearch into panels on a dashboard. Build one per service with request rate, error percent and p95 latency on the top row, so a spike can be matched with a deploy. Avoid a wall of 60 graphs; a dashboard should answer one question quickly, and alerts, not dashboards, should tell you when to look.