Retention: Reading the Curve and Fixing the Opening
Compute average viewed, the 30-second intro share and the biggest drop from two retention curves, and use the shape to fix an opening.
In this lesson you will
- Describe the retention report and its key moments
- Compute average viewed and the intro share
- Read shapes as questions
- Fix an opening step by step
About 30 minutes
Where do they leave?
Kofi’s views are growing, but he feels uneasy. The average view duration of his basil video is almost five minutes of eight, which seems good. Yet in the comments someone writes, “Lost interest after the intro.” Abena says, “An average hides where people leave. You need the shape.” The shape is the audience retention curve: a line that shows, for every moment of the video, how many of the viewers are still watching. It is one of the most useful pictures in Analytics, and you can understand it with simple arithmetic.
What YouTube shows
In YouTube Studio, the audience retention report is available for each video on its own, and its data typically takes a day or two to appear, according to YouTube Help. It draws the curve and highlights key moments: the intro, which is the share of the audience still watching after the first 30 seconds, plus top moments where retention is strong, spikes where the curve rises and dips where it falls. YouTube also lets you compare with typical retention for your recent videos of similar length. Help says that a high intro share could mean the start matched what the thumbnail and title promised, and it advises against treating one number as good or bad on its own.
Curves you can compute
The program below takes two made-up curves for an eight-minute video, with a reading every tenth of the way through, 48 seconds apart. For each it calculates the average share viewed, using the area under the curve; the share still watching after 30 seconds, by drawing a straight line between the two nearest readings; the biggest single drop between readings; and the average view duration in seconds.
video = 480
a = [100, 79, 67, 60, 56, 53, 50, 47, 45, 43, 40]
b = [100, 64, 58, 55, 54, 53, 52, 51, 50, 49, 48]
step = video / (len(a) - 1)
def average(curve):
return sum((x + y) / 2 for x, y in zip(curve, curve[1:])) / (len(curve) - 1)
def at(curve, seconds):
i = int(seconds // step)
frac = (seconds - i * step) / step
return curve[i] + (curve[i + 1] - curve[i]) * frac
for name, curve in (("A", a), ("B", b)):
drops = [(curve[i] - curve[i + 1], i) for i in range(len(curve) - 1)]
worst, where = max(drops)
print(name + ": average viewed " + str(round(average(curve), 1)) + " percent")
print(name + ": still watching at 30 s: " + str(round(at(curve, 30), 1)) + " percent")
print(name + ": biggest drop " + str(worst) + " points, at " + str(round(where * step)) + " to " + str(round((where + 1) * step)) + " s")
print(name + ": end of video " + str(curve[-1]) + " percent")
print("average view duration A:", round(average(a) / 100 * video), "s")
print("average view duration B:", round(average(b) / 100 * video), "s")
A: average viewed 57.0 percent
A: still watching at 30 s: 86.9 percent
A: biggest drop 21 points, at 0 to 48 s
A: end of video 40 percent
B: average viewed 56.0 percent
B: still watching at 30 s: 77.5 percent
B: biggest drop 36 points, at 0 to 48 s
B: end of video 48 percent
average view duration A: 274 s
average view duration B: 269 s
Video A averages 57.0 percent viewed, and video B averages 56.0. They are nearly the same. But their shapes differ greatly. At 30 seconds, A still has 86.9 percent of viewers and B only 77.5. B loses 36 points in the first 48 seconds, against 21 for A. After that, B holds its viewers well: it ends with 48 percent, while A ends with 40. In plain words, B has a weak opening and a strong middle and end. A has a stronger opening and then a steady slide. The average view duration is 274 seconds against 269, almost identical. If you looked only at that number, you would think the videos were the same.
Reading the shape
| Shape | What it may mean |
|---|---|
| Steep drop at the start | Opening did not match the promise |
| Steady slide | Ordinary: people leave as they find what they need |
| Flat stretch | A part that held viewers well |
| Spike | A part that people rewatched or skipped to |
| Sudden dip | A part that lost people |
These are clues and not proofs. A steep drop at the start can also mean that many viewers arrived from a suggestion and left at once, which is normal, and the later part of the curve is about the people who stayed. Treat a graph as a question: why did they leave there? Then watch that moment of your own video and see what you would do as a viewer.
Fixing an opening
- Check the promise. Does the first line say what the title and thumbnail promised? If not, change the first line or the packaging.
- Cut the warm-up. Greetings, thanks and long introductions belong later, if at all.
- Show the result first. A quick look at the finished thing earns the time for the explanation.
- Keep the pace in the first minute. Short shots and clear words, as in lesson 5.
- Change one thing and compare the intro share of the next few videos, not a single one.
Judging a video by its average alone. Two videos with the same average can have opposite problems. Look at the curve and the intro share.
The curves and video are made up. Real curves are bumpier. The straight-line reading and the trapezoid average are simple approximations for teaching, and Studio’s own numbers may differ slightly.
Practice for this lesson
Check what you learned
Five quick questions. Pick an answer and you will see straight away whether it is right and why.
What does the intro share in the retention report measure?
- The share who subscribed
- The number of likes
- The share of the audience still watching after the first 30 seconds
- The length of the intro in seconds
A high value could mean the start matched the promise.
Why can two videos with the same average viewing have different problems?
- Averages are always wrong
- The videos are the same length
- Retention is random
- An average hides where viewers leave
Video B lost more at the start but held its viewers later.
What should you do after finding a steep early drop?
- Delete the video
- Check that the opening matches the title and thumbnail promise
- Add a longer intro
- Ignore it
The opening must keep its promise.
Why compare retention across several videos and not one?
- The site requires five videos
- Curves are identical
- A single video may be unusual
- Videos have no shape
Look at patterns over several videos of similar length.
What is a good way to use a retention graph?
- As a question: why did they leave at that moment?
- As a score to maximise at any cost
- As proof of a cause
- As a replacement for watching the video
Then watch that moment as a viewer would.
Summary
- A retention curve shows where viewers leave.
- Two videos with the same average can have different shapes.
- The intro share is the share still watching after 30 seconds.
- Treat a drop as a question, fix one thing, and compare several videos.
Questions
Where do I find the retention report?
In YouTube Studio, open a video's analytics and look for audience retention, usually under Engagement. Menus change, so search YouTube Help for the report name if it has moved.
What is a good retention number?
There is no universal number. YouTube Help suggests comparing with your own recent videos of similar length. Videos differ a lot by topic and length, so compare like with like.
Should I make videos shorter to keep retention up?
Not by default. A video should be as long as it needs. Cut anything that does not help, but do not cut useful steps just to improve a percentage.
