P99 latency went from "we don't know" to monitored, alerted, and acted on. First week we found a database query that was running 400ms on every API call.