A performance requirement is a specific, measurable commitment for how fast or resource-efficient a system must be in a user-relevant situation. It replaces vague requests such as “make search faster” with a testable target such as “95% of search queries return within 200ms.” The requirement should state the flow, threshold, population or context, and measurement condition. section 3.0 section 3.4
Treat the user journey as the unit of performance. Users experience searching, checkout, startup, exports, and editing—not isolated database queries, bundles, or endpoints. A requirement for a complete flow prevents teams from improving one component while an unaddressed bottleneck still delays the outcome. section 3.1.3 section 3.4.1
Write an effective performance requirement by making it:
- Specific: name the operation, threshold, and measurement, such as filtering within 500ms for 95% of users. section 3.1
- User-centered: tie the target to a visible frustration or job rather than an internal metric with no observable effect. section 3.1.3
- Context-aware: distinguish relevant devices, networks, file sizes, project sizes, or user segments. section 3.4 section 3.4.1
- Realistic and jointly owned: account for engineering constraints and set the target before optimization begins. section 3.0 section 3.6
- Maintained: review and tighten targets as product maturity, infrastructure, and user expectations change. section 3.4.3
From component tuning to a flow requirement
A team reports smaller bundles, faster queries, and lower memory use but cannot tell whether filtering is good enough. A better requirement is: “Filtering completes within 500ms for 95% of users, with no individual step exceeding 200ms.” Profiling can then reveal sequential requests as the limiting step, directing work toward parallel requests and caching rather than unrelated optimizations. section 3.1
Requirements also make trade-offs explicit. Pair a feature commitment with its performance constraint, such as delivering real-time commenting without page load exceeding three seconds. If a prototype temporarily regresses performance, record the accepted limit and the planned target rather than letting performance debt become permanent. section 3.4.4
Table 3.7 Performance targets by product maturity
|
Platform
|
Early Stage
|
Growth Stage
|
Maturity Stage
|
|
Web
|
LCP < 4s, no failures
|
Meet Core Web Vitals
|
Exceed Core Web Vitals
|
|
Backend
|
P95 < 1s, 99% uptime
|
P95 < 500ms, 99.5% uptime
|
P95 < 200ms, 99.9% uptime
|
|
Mobile
|
Cold start < 5s, no crashes
|
Cold start < 3s, 60fps
|
Cold start < 2s, smooth animations
|
|
Desktop
|
Launch < 10s, responsive
|
Launch < 5s, efficient resources
|
Sub-second launch, minimal resources
|
Compare example target levels for web, backend, mobile, and desktop products at early, growth, and maturity stages. section 3.4.3
Put requirements into everyday delivery: choose one business-critical flow, establish a measurable baseline and target, make it part of the definition of done, and use measurement before and after changes to validate user impact. Performance requirements can affect adoption as well as satisfaction: services that fail buyer load tests or respond much slower than alternatives may be rejected before features are evaluated. section 3.5 section 2.2.3
