Synonyms: response time, system delay, lag, load time, perceived latency
Do not index
Definition
System latency is the gap between a user action and the system's visible response. It covers everything from a button tap to a full page load, and it decides whether a product feels fast or feels broken.
Use cases
Ignore latency and users assume the product is dead, then bounce. Jakob Nielsen's thresholds set the stakes: 0.1s feels instant, 1s keeps their train of thought, 10s is the edge before they give up.
"Did my click even register?" A checkout button with no instant state change gets tapped twice, and now you have duplicate orders. Flip the button to a loading state on tap, before the server replies.
The frozen feed. A blank white screen during a 3-second fetch reads as a crash, so people refresh and make it worse. A skeleton screen shows the layout right away and they wait instead of rage-quitting.
AI chat lag. An LLM takes 8 seconds and the user thinks nothing happened. Streaming tokens word by word holds attention, even though total time is identical.
How it's used in practice
Set a latency budget per interaction: under 100ms for taps, under 1s for transitions, visible feedback past 1s.
Acknowledge instantly: fire a state change (spinner, disabled button, optimistic update) within 100ms of the action, before data returns.
Match the loader to the wait: spinner under 1s, skeleton or progress bar for longer, step count ("Uploading 3 of 12") for multi-second jobs.
Measure real numbers: pull field data from Core Web Vitals (LCP, INP) and test on a throttled mid-range phone, not your fast laptop.
🪄
Pro tip: Optimistic UI is the cheat code for perceived speed, but it needs a rollback path.
If you show a "liked" state before the server confirms and the request fails, revert quietly and tell the user. A fake success with no error handling does more damage than a slow, honest loader.
Challenges & limitations
Some latency is out of your hands. Slow networks, third-party APIs, and model inference times cap how fast you can ever go. Design hides the wait, it doesn't delete it.
Fake progress backfires. Timer-based bars that aren't tied to real work lose trust the moment users watch them stall at 90%.
Lab speed lies. A page can pass Core Web Vitals in testing and still drag on a user's three-year-old phone with 15 tabs open.
Free resources
NN/g: Response Time Limits — Jakob Nielsen's foundational 0.1s / 1s / 10s thresholds and what users feel at each.
