Three features carry almost all the information. The initial cliff, in the first two or three seconds, which measures the hook and nothing else. The slope through the middle, which measures whether the piece keeps giving reasons to stay. And any sharp local drop, which is a specific moment that lost people and can usually be identified to the frame.
The common analytical error is treating average view duration as the metric. It is an average of a shape and hides everything useful: two pieces with identical averages can have completely different curves, one of which is fixable and one of which is not.
Rises matter too, and are under-used. A bump means people scrubbed back, which almost always means something was said too quickly or shown too briefly. That is an instruction to slow down at exactly one timestamp, which is the most actionable note a graph can give you.
What does a retention curve tell you?
Where viewers left. The first-seconds cliff measures the hook, the middle slope measures whether the piece keeps earning attention, and a sharp local drop identifies a specific moment that failed.
What does a rise in the retention graph mean?
Viewers scrubbed back, which usually means something was said too fast or shown too briefly. It is an instruction to slow down at that exact point.