Ship thresholds, not averages
Averages hide the user experience that matters. Thresholds — p95, p99, or the percentage exceeding a pain point — tell you what your product actually feels like.
When a dashboard shows 'average response time: 200ms,' both the team and the stakeholder feel good. But the user experiencing 3-second page loads on a slow connection or during peak hours is invisible in that number. If the p95 is 2.5 seconds, then one in twenty users is having a recognisably bad experience — enough to bounce, file a complaint, or churn. The average tells you nothing about that user.
The same principle applies beyond performance. Adoption rate of '65%' hides the distribution: some teams use the feature daily, others tried it once and never returned. Average revenue per user hides whether one segment is carrying the rest. Average NPS hides the passionate promoters and the furious detractors cancelling each other out.
Ship thresholds instead. Pick the one metric that, when it degrades, correlates with user churn or escalation. Track its p95, p99, or the percentage of users below an acceptable floor. Report that number prominently on the product dashboard. When it moves, you know something real happened.