APP Speed Evaluation Methods and the LazyPerf Tool

In order to understand why it rains and thunders, users often search on their phones. The simple search request triggers a long interaction process, similar to a bank transaction that requires multiple steps. Any excessive waiting in an app can cause user churn.

Speed is a key factor for user retention. Studies show that a one‑second increase in page load time can lose 10% of users (BBC), and reducing perceived wait time can boost conversion rates (e.g., Lazada, GYAO).

Therefore, we regularly evaluate the speed of critical app scenarios with two main goals: (1) prevent performance degradation, and (2) identify gaps compared to competitors.

1. Log‑based speed evaluation – By inserting log points at key nodes (e.g., app launch, first paint, first contentful paint, full render, time‑to‑interactive) and aggregating the data, we obtain speed metrics without heavy instrumentation. This method is low‑cost, easy to implement, and reflects real‑world usage, but it cannot measure competitor apps and may miss fine‑grained user perception.

2. Manual evaluation – The app is operated while recording the screen. Afterward, the video is frame‑by‑frame analyzed to locate the start and end frames of interest. This yields the most user‑centric metrics and allows competitor comparison, yet it is labor‑intensive (days per scenario) and subject to network/device variability.

3. Automated evaluation – Automation tools (Appium, ADB, WDA, etc.) drive the app while recording the screen. Algorithms then detect the start/end frames, dramatically reducing human effort and improving repeatability. However, challenges remain: writing stable test cases, ensuring execution reliability, and developing accurate frame‑recognition algorithms.

To address these challenges, we built LazyPerf , an integrated APP speed testing platform that lowers both the cost of writing automation cases and the manual effort of frame calibration.

Key features of LazyPerf:

Case creation by recording real‑device interactions, eliminating the need for script coding.

Layout‑based widget addressing that is resilient to UI changes and supports pattern‑based matching.

Hybrid addressing using UI hierarchy, OCR, image matching, and similarity comparison for complex screens.

Built‑in first‑ and last‑frame detection algorithms covering 20+ scenarios with >90% accuracy.

Support for multi‑scenario annotation within a single recording, reducing overall test cycles.

“Ten‑frame” calibration mode that keeps manual verification under 10 seconds.

The platform also highlights current limitations of open‑source automation tools (performance overhead, device compatibility, maintenance) and proposes a future direction where physical automation (robotic arms, vision‑guided touch) complements software‑based testing.

Overall, the article presents a comprehensive view of APP speed measurement techniques, their trade‑offs, and a practical solution (LazyPerf) to make performance testing more efficient and scalable.