The first dashboard usually tries to impress. It should try to calm.
During a bad week, nobody wants a wall of clever charts. They want to know whether traffic is normal, errors are rising, latency is changing, jobs are stuck, and money is still moving. That is enough for version one.
Dashboards fail when they become museums of metrics. Every chart had a reason once. Nobody remembers it. The page loads slowly and answers nothing.
Start with five panels
Use five boring panels:
- Traffic. Requests, jobs, or active users.
- Errors. User-visible failures, not just exceptions.
- Latency. The workflows people wait on.
- Queues. Anything that can silently accumulate.
- Money or trust. Payments, exports, messages, deploys, whatever the product cannot afford to fake.
Put links to logs and runbooks beside the charts. A dashboard that cannot lead to action is wallpaper.
The first dashboard is not there to show that you measure everything. It is there to make the next incident shorter.