The five video metrics worth looking at

Dashboards are generous with numbers because numbers are cheap to produce. These are the ones that change decisions.

1. Interaction rate

Of the people who saw the video, how many did something — opened it, tapped a product, unmuted it.

This is the honest engagement number, because autoplay does not inflate it. A video with many views and almost no interactions is being scrolled past, which usually means the thumbnail or the opening two seconds are not working.

If it is low: change the thumbnail first, then the opening. Not the placement — a video nobody engages with anywhere is a content problem.

2. Add-to-cart rate from video

Of the people who interacted, how many added something. This is the conversion step the whole mechanism exists to produce.

If it is low but interaction is high: the video is interesting and the products are not the ones it made them want. Check the tagging — in practice this is usually a tagging mismatch rather than a content failure.

3. Attributed revenue per video

What the video actually earned. The number to rank your library by, and the one that decides what to make more of.

What to do with it: look at the top five together and find what they share. It is usually a format rather than a product — a length, a framing, a way of opening.

4. Placement performance

The same video in different positions, compared. This is how you find out that your carousel is doing nothing and your product-page videos are doing everything, which is a common and expensive thing not to know.

What to do with it: move effort to where it works. Most stores have one placement doing the majority of the work and several that exist because they were set up once.

5. Mobile versus desktop, on all of the above

Not a metric so much as a split you should apply to every other metric. Behaviour differs enough that an aggregate can hide a serious mobile problem behind a good desktop average.

What to do with it: if mobile is much worse, the cause is nearly always layout or speed, not content.

What to ignore

• Total plays. Measures traffic.
• Average watch time in isolation. A ten-second video with eight seconds watched is doing better than a sixty-second video with twenty, and the raw average says the opposite.
• Completion rate on looping video. Means little when the video restarts automatically.

A rhythm for reading them

Monthly, not weekly. Video data is noisy at small volumes and a weekly read will have you reacting to randomness. One monthly look at the five above, one decision about what to make more of, and leave the rest alone.