The P50 and P99 rule

When implementing observability, the metrics you focus on depend heavily on your business requirements. While P50 is usually a good starting point for measuring baseline latency, certain scenarios demand a focus on P99 metrics. Many companies ignore P99 latency, dismissing it as too niche or expensive to justify fixing. However, the 1% of users affected by those delays could easily be your highest-revenue customers.

This is because the clients falling into that 99th percentile are typically power users who heavily utilize your system, making them the most likely to trigger these edge-case performance bottlenecks.