Split your retention curve into new viewers and returning viewers, and the average stops hiding two opposite stories. New viewers judge the promise, returning viewers judge the payoff, and a blended curve can look healthy while both groups are unhappy for different reasons.
Most creators never open the segment view. The platform buries it two clicks deep and the unfiltered line sits right on the main screen, so the average wins by default.
An average is a compromise between disagreeing groups. If new viewers leave hard in the first thirty seconds while returning viewers watch to the end, the middle curve shows a gentle slope and no alarm bell rings.
That matters because the fixes point in opposite directions. New viewers leave when the video assumes context they do not have. Returning viewers leave when the video repeats what they already know. One curve, two completely different rewrites.
The platform usually hides the split behind an audience filter, so it feels optional. It is not. One click separates two groups with opposite behavior, and the blended view actively conceals the disagreement that matters most for the next edit.
A retention average is two audiences agreeing to disagree, averaged into silence.
You do not need all four on every upload. Pick the split that matches the question you already have and read only that one. Segment views are fascinating, and a twenty-minute review becomes an hour of wandering fast.
If you can only ever check one split, choose new versus returning. It is the most stable, the most widely supported, and the least likely to be a platform quirk. The others earn their keep when a specific question points at them, not as a weekly ritual.
The pattern I trust most: new viewers drop early, returning viewers hold. That is a packaging-to-content mismatch. The thumbnail promised a wider audience than the script served, and the fix is either a narrower promise or a friendlier opening.
When returning viewers drop first, something else is going on. Usually the topic repeated a recent upload, or the video spent long on setup for people who had already read the community post. Loyalty drops are rarer and louder than stranger drops.
Watch the sample size before any of this. A video with four hundred views has too few returning viewers to compare honestly, and the segment line will swing on single viewers. Small samples produce confident nonsense.
Segmenting takes about a minute and changes the conclusion almost every time. That ratio is why it is the first thing I check and the last thing I remove from a weekly review.
Averages are fine for tracking a channel over a year. For diagnosing a single video they are too polite. Split the line and let the two audiences speak separately. The habit pays for itself the first time a rewrite clearly fixes one group's segment without touching the other. One click, two truths, and a clearly better next edit.
Platform analytics tools are fine for the basics, but third-party tools only earn their price when you need cross-channel comparison or retention views.
Most analytics tools show you fifty numbers. Three of them answer most questions: click-through rate, retention at thirty seconds, subscribers per view.
Most creators build dashboards they never open. Analytics tools only earn their space in your routine when the screen answers one question you ask every week.
Analyze a channel, extract its structural DNA, and generate scripts engineered from patterns that actually work.
Analyze a Channel