We Run Our Own Channels. Here Is The Retention Data Nobody Publishes.
Most studios report views. Views go up on their own. Here is what we track instead, and the retention patterns we see across our own portfolio.
Almost every production company reports views, impressions and engagement. All three go up on their own if you spend enough on distribution, which is what makes them comfortable to report and useless to act on.
We own and operate our own channels, which means we live with the numbers rather than presenting them. Here is what we actually track, what the patterns look like, and how to read your own.
The Four Numbers Worth Tracking
Average view duration, in seconds and as a percentage
Both, because they answer different questions. Seconds tell you how much attention you earned. Percentage tells you whether the video was the right length. A sixty percent retention on a two minute film and on a ten minute film are very different achievements.
The thirty second hold
What share of people who pressed play are still there at thirty seconds. This is the single most diagnostic number we track, because it isolates the hook from everything else.
The shape of the curve, not its height
A cliff, a steady slide and a specific drop mean three completely different things and have three completely different fixes. The average hides all of it.
Action rate among viewers versus non-viewers
On a page with a video, compare the conversion rate of people who played it against those who did not. This is the only number that says whether the video did a job.
The Patterns We See Consistently
The first thirty seconds decide the rest of the curve
Across every format we run, a video that holds its audience through thirty seconds retains far better through the middle than one that does not. It is not that the middle of those videos is better written. It is that the audience remaining after thirty seconds has already self-selected.
The practical consequence is that improving your opening improves your whole curve, and improving your middle mostly does not.
Length is a promise, and breaking it costs you
Videos that run meaningfully longer than the viewer expected from the thumbnail and title show a distinctive shape: normal retention until roughly the point the viewer assumed it would end, then a sharp drop. Matching stated length to actual length recovers most of that.
The back half is where watch time is actually earned
Most teams put ninety percent of their effort into the opening. That is correct for whether someone stays, and wrong for how much time you bank. Total watch time is dominated by minutes four through nine on a long-form video, and those minutes are usually the least edited part of the film.
Captions change the numbers more than production value does
Of the cheap interventions available, adding accurate captions has moved our numbers more reliably than any equivalent spend on production quality. A large share of first views happen with sound off.
How To Read Your Own Curve
- A cliff in the first ten seconds. The thumbnail and title promised something the opening did not deliver. This is a packaging problem, not a video problem.
- A steady decline with no features. Pacing. There is no forward tension and nothing the viewer is waiting to find out.
- A drop at one timestamp. Something concrete is wrong there. Go and watch that thirty seconds. It is usually a tangent, a slow transition, or the moment the film starts selling.
- A rise partway through. People are re-watching something. That is worth understanding, because it is usually the most valuable part of your film.
We report these numbers on client work as a matter of course, including the videos that underperformed. Roughly half of everything anyone tries does, and a report that never says so is marketing rather than measurement.
Work With UsThe Two Mistakes Teams Make Reading Retention
Comparing across formats. A ninety second product video and a ten minute documentary have no business being on the same chart. Percentage retention falls with length as a matter of arithmetic, so a long film always looks worse and is often working harder.
Reacting to one video. A single result is mostly noise: publication time, thumbnail, what else was competing that week. The signal lives in the trend across eight uploads. We do not change a format on one data point and neither should you, which is the same discipline as not pivoting a channel at video four.
Why We Publish This At All
Because the alternative is a portfolio page with no numbers on it, which is what most of this industry offers, and it asks a client to buy on taste alone.
Running our own channels means we are the client for a large amount of our own work. It changes what we optimise for, and it is the reason we own channels at all rather than only taking briefs.
Common questions
- What is a good average view duration?
- It depends on length and format, which is why a single benchmark number is misleading. Track your own trend across the last eight videos rather than comparing to an industry figure.
- Should I optimise for views or watch time?
- Watch time, in almost every case. Views measure packaging. Watch time measures whether the thing was worth watching.
- Do you share retention data on client projects?
- Yes, as standard, including the ones that underperformed and what we would change.