You spent $3,000 on traffic last week. Your video host reports 4,812 plays and an average watch time of 41%. Your cart reports 62 sales.
Now answer a simple question: which of those 4,812 plays produced the 62 sales?
You cannot. Nobody can, from those numbers. And that is the whole problem, because every decision you are about to make depends on the answer.
A play count is a measure of your ad, not your video
Think about what a play actually is. A play means someone landed on the page and playback started. That is a fact about your traffic. It tells you the ad worked, the click worked, and the page loaded.
It tells you nothing about whether the video sold anything.
Average watch time is not much better. Average is where information goes to die. A 41% average watch time is the same number whether half your viewers watched 82% and the rest bounced instantly, or every single viewer drifted off at the four minute mark. Those are two completely different videos with two completely different problems, and the average hides both.
Play counts tell you the ad worked. They are silent on whether the video did.
The number that matters is the second they left
A retention curve is a play count that has been made useful. Instead of one number for the whole video, you get a number for every second of it. And the shape tells you what to do.
A gentle slope is normal. People drift. A cliff is not normal. A cliff means something specific happened at a specific second and half your audience decided they were done.
Go and watch that second. It is usually one of four things:
- The hook ended and the proof did not start.
- You asked for the sale before you earned it.
- You said something that broke belief, and they stopped listening.
- Nothing happened for thirty seconds.
You cannot find that second in an average. You can find it in a curve, and once you have found it you have a concrete edit to make instead of a vague feeling that the video is underperforming.
Watch time is the intent signal your pixel is missing
Here is where most funnels quietly leak money.
By default, your Meta pixel fires on page load. So every single person who lands on the page, including the ones who bounce in two seconds without ever pressing play, is fed back to Meta as a signal. You are teaching the algorithm to find people who load your page. That is not the same as teaching it to find people who buy.
Now fire the pixel at 75% watched instead. Suddenly you are only feeding back people who sat through three quarters of your pitch. Those people are qualitatively different. They are interested, and Meta's optimisation gets a much cleaner signal about who to go and find next.
Nothing about your ad spend changed. You just stopped lying to the algorithm about who your good viewers are.
Then close the loop with the cart
Watch time is a proxy for intent. A sale is not a proxy for anything. It is the thing.
So the last step is to stitch the sale back to the play that earned it. When your cart fires a postback on a completed order, that order gets attributed to the specific session, the specific video, and the specific variant that produced it.
That is the number you have been missing. Not "this video got 4,812 plays" but "variant B produced 38 of the 62 sales on half the traffic". One of those sentences tells you what to do on Monday morning.
What to actually measure
- The retention curve, for the second people leave. That is your edit list.
- Watch-time milestones, wired to your pixel, so the algorithm optimises toward attention rather than page loads.
- Sales attributed back to the play, so a split test is decided on revenue instead of click-through.
Play counts are fine. Keep them. Just stop making decisions with them.
This is the reason TrackPlay exists. The conversion layer came first, and the player was built around it. If you want to see it on your own video, the free plan gives you 1,000 plays a month with no card.