Watch time

Watch time is the cumulative time viewers have spent watching a video, usually reported in minutes for the video as a whole and in seconds for an individual viewer. It is the most informative single number in video analytics because it is the hardest to inflate: a view can be an accident, and time cannot.

Its close relative is average view duration, the time a typical viewer spends before leaving, which normalises watch time against how many people arrived. A video with high watch time and low average duration was watched by many people who left early; one with the reverse was watched properly by few. The two describe different problems.

Benchmarks published as industry averages are worth very little here, because duration depends almost entirely on the length and purpose of the video. A ninety-second prospecting message and a forty-minute training recording cannot be compared on the same scale. The useful comparison is against your own previous videos and against the same video last quarter.

Watch time is also the metric most distorted by autoplay. If a page starts the video automatically, some of the recorded time belongs to people who were not watching at all, which is worth knowing before treating a rise as an improvement.

One more caution about comparing across platforms. A social feed, a hosted video page and an email link count time in different ways and start counting at different moments, so a figure that looks like a decline after a change of tool is frequently an artefact of the measurement. Compare within a platform, and treat a switch as a reset of the baseline rather than a result.