How do I pick the right playbook for my system?

Start with your critical user flow (e.g., checkout) and trace it end-to-end with OpenTelemetry. If most time is inside a single DB call, use the Monolith + RDBMS playbook. If you see synchronous fan-out across services, use the microservices playbook. If it’s event-driven, optimize queue backpressure. For API endpoints on Lambda/Cloud Run, use the serverless playbook. For web vitals, use the SPA & edge playbook. For generative AI endpoints, use the inference playbook.

What KPIs should we monitor to prove business impact?

Tie engineering metrics to product KPIs: p95 TTFB and LCP to conversion rate, error budget burn to churn/retention, infra cost per transaction to gross margin. Use cohort analysis to verify improvements stick beyond novelty (e.g., mobile vs desktop, geographies).

Isn’t micro-optimizing code faster than changing architecture?

Micro-optimizations can help, but the highest leverage usually comes from fixing the pattern-level bottlenecks: fan-out depth, hot DB paths, cold starts, lack of caching, and backpressure. We routinely see 2–5x wins from these changes, while code micro-tuning nets 5–15%.

How do we roll these changes safely without new incidents?

Use GitOps with ArgoCD, progressive canaries with Argo Rollouts, and SLO-based alerts. Change one variable at a time, add a rollback plan, and watch p95/p99 and error rate during each step. Keep blast radius small and iterate.

Performance-optimization · Nov 22, 2025 · 10 minute read

The Performance Playbooks I Wish I’d Had: Pattern-by-Pattern, p95 Down, Revenue Up

Battle-tested optimization playbooks for the architectures you actually run—monoliths, microservices, queues, serverless, SPAs, and AI inference—focused on user-facing metrics and business impact.

Back to all posts

The Performance Playbooks I Wish I’d Had: Pattern-by-Pattern, p95 Down, Revenue Up

Key takeaways

Implementation checklist