Our users answer our surveys faster than our analytics can measure.

We ran in-product surveys with a single question and a 5-point scale. Median time from survey appearing to answer submitted: 1.8 seconds. Median time from page load to survey render: 2.4 seconds.

Users were answering before the survey had fully rendered on our analytics timeline, which meant our event logs showed responses arriving before the impression event. Our data pipeline was dropping 12% of responses as "impossible" and flagging another 8% as bot traffic.

The responses were real. Our instrumentation was wrong. We had been discarding human answers because our clocks did not agree with human speed. When we fixed the event ordering to use client-side timestamps instead of server-side, response volume rose 21% overnight without any change to the survey.

What data are you throwing away because it arrived faster than your system believed possible?