Performance Test Plan Template for Agile Teams

For HTTP API load testing: k6 (scriptable in JavaScript, good CI integration), Locust (Python-based, distributed), or Gatling (Scala, good for complex scenarios). For frontend performance: Lighthouse CI, WebPageTest, or SpeedCurve. For monitoring during tests: your existing APM tool (Datadog, New Relic, Grafana). Pick tools your engineering team already knows. The testing framework matters less than running the tests consistently.

Start with three inputs: customer SLAs (contractual commitments), user expectations (industry benchmarks: pages should load in under 2 seconds, APIs should respond in under 500ms), and business requirements (checkout must complete in under 3 seconds or conversion drops). Set internal targets 20-30% tighter than SLA targets to create a safety margin. The [Technical PM Handbook](/technical-pm-guide) covers how PMs should negotiate performance requirements with engineering.

Run a subset (critical path latency checks) on every deploy via CI. Run the full load test suite weekly or before every major release. Run soak tests (8-24 hours) monthly to catch slow memory leaks and connection pool exhaustion. Run stress tests before expected traffic events (product launches, marketing campaigns, seasonal peaks).

First, reproduce the failure and confirm it is not a test environment issue (insufficient resources, network throttling, shared environment contention). If the failure is real, use the bottleneck analysis table to identify the layer causing the problem. Common patterns: database query without an index (add the index), N+1 query pattern (add eager loading), memory leak (profile and fix), third-party API timeout (add circuit breaker). File a bug with the test results, bottleneck analysis, and the [defect density](/glossary/acceptance-criteria) trend.

Yes. P99 means 1 in 100 requests experiences this latency or worse. At 10,000 daily active users, that is 100 slow experiences per day. At 100,000 users, it is 1,000. P99 often reveals infrastructure issues (garbage collection pauses, cold starts, database lock contention) that P50 and P95 mask. PMs should set P99 targets and review them alongside feature metrics. Use the [RICE Calculator](/tools/rice-calculator) to prioritize performance fixes by estimating the user impact of latency improvements. ---