An alert that does not change a decision only interrupts. Observability begins when a signal helps explain the state of the system and points toward an action.
Start with the questions
What is failing? Who is affected? Since when? What changed? Your metrics, traces and logs must reduce the time needed to answer those questions.
Fewer panels and better relationships between signals usually produce more reliability than collecting telemetry without an operational hypothesis.