How to read video analytics and act on them
Four numbers cover almost every decision: play rate (thumbnail and placement), the retention curve (the edit), completion rate (length and payoff order) and CTA click-through (offer and timing). Read them in that order — a video nobody starts cannot be fixed by re-cutting the middle.
The chain
impressions → plays → watch time → completion → CTA click → leadFind the weakest link before changing anything. Each link has one honest owner:
| Symptom | Metric | Fix |
|---|---|---|
| Nobody starts it | Play rate | Poster frame, position on the page, the headline above it |
| Cliff in the first 10% | Retention | Cut the intro; open on the payoff |
| Step down mid-video | Retention | Watch that exact second and cut it |
| Long slow decline, low completion | Completion | The video is too long, or the best part is last |
| Watched but no clicks | CTA CTR | Wrong offer, wrong moment, or no CTA at all |
| Clicks but no leads | Form | Too many fields, or the promise is unclear |
Retention curves, read properly
The curve is the share of viewers still watching at each 1% of runtime. Compare a video against itself over time and against your own other videos — not against a published industry average, which is measured on different content for a different audience.
Two readings people get wrong:
- A cliff at the very start is normal to a point: some of it is accidental plays. What matters is where it flattens.
- An upward bump is information: viewers are rewinding. Something went past too fast — usually a screen with text on it.
Play rate is a page problem
Play rate is plays ÷ impressions. It is decided by the poster, the surrounding copy and where on the page the player sits — not by the video's content, which nobody has seen yet. Test the thumbnail first; it is the cheapest experiment you have.
Beware of averages
"Average watch time" hides the shape. Two videos with identical averages can be a steady decline and a cliff at 0:20, and they need opposite fixes. Use the curve; use the average only for reporting.
Turning it into a decision
A workable loop: pick the weakest link, make one change, wait for a comparable number of plays, compare like with like. With per-video plays, completions, curves, referrers and CTA click-through in Videokr's analytics, that loop takes a week rather than a quarter.
Frequently asked questions
What is a good video completion rate?
It depends far more on length and audience than on quality — compare a video against your own others rather than an industry average, and watch the retention curve for where it loses people.
Which metric should I fix first?
Play rate. A video nobody starts cannot be improved by editing it.