IBM researchers Walter J. Doherty and Ahrvind J. Thadhani proposed this figure in a 1982 paper studying computer response time and productivity. 400ms sits near the minimum delay people perceive as slow — faster than that, and people feel like they're driving the system, not waiting on it.
When real server response can't stay under 400ms, skeleton screens and optimistic UI become the practical fix — they reduce *perceived* latency, which is a different axis from actual latency.
Left responds within 400ms and the user flows straight into the next action. Right crosses the threshold, and the user's attention visibly drifts elsewhere.
When to use
When a screen depends on a slow API, aim to at least *react* within 400ms — even if the real work takes longer.