# Dashboards Nobody Opens

> Ninety dashboards, four in use. Build for a question, not a system.

Source: https://christwentyfour.com/blog/dashboards-nobody-opens

Most teams have far more dashboards than they use. Someone builds one during an incident, it helps for a week, and then it sits. In the next incident, the responder has to pick one. A bad pick wastes the most expensive minutes of the event.

## Build for a question

"Is checkout healthy right now?" is a question. "Payments service" is a topic, and topic dashboards turn into walls of metrics. Test it: could someone who does not know the service tell, in ten seconds, if things are okay?

## Three tiers, top to bottom

The top shows the few user-facing numbers: success rate, latency at two percentiles, throughput. The middle shows likely suspects: dependencies, errors by type, queue depth. The bottom holds detail for confirming a hypothesis.

## Show percentiles, not averages

A healthy average can hide one request in ten timing out, and those are the users who complain. A widening gap between the middle and the tail usually means queueing has started.

## Delete aggressively

Once or twice a year, archive what nobody opens. An unused dashboard costs a responder time in the next incident.