A 20-minute weekly retention review
A 20-minute agenda that turns last week's curves into one edit decision, one packaging decision, and three numbers you log so the trend is visible.
A retention review that does not produce a decision is a mood. Twenty minutes is enough to read last week's curves, write two decisions, and log three numbers. It is not enough to rebuild the channel, which is why the cap is useful: when the timer ends, you stop, even if you still have opinions.
The output is deliberately small. One edit decision (a change to a video that already exists, or to the next script's structure). One packaging decision (title, thumbnail, or a test). If you leave with seven "we should" notes, you had a conversation, not a review.
Paddy Galloway's framing, via Colin and Samir, is the reason this split exists: top creators spend about 30 percent of their time on ideation and packaging; small creators spend about 5 percent, and the rest on filming and editing. A weekly slot that must emit a packaging decision is how you stop living in the 5 percent. The edit decision stops the other failure mode, which is endlessly recutting thumbnails while minute six keeps leaking.
The agenda, with the clock on
Do this on the same weekday. Monday morning after a Sunday publish, or Friday before you lock next week's outline. Open Studio (and any other platform you posted to) before you start the timer. Phone face down.
Minutes 0–3: pick the set. Last week's uploads, same format kept together. If you posted a 12-minute tutorial and three Shorts, the tutorial is the review; the Shorts get a 5-minute sidecar at the end or a separate week. Comparing a Short's viewed-versus-swiped number to a long-form average percentage viewed is how you invent a crisis. Note CTR, first-minute hold, and average percentage viewed for each. You are not diagnosing yet. You are making a pile.
Minutes 3–8: name one problem per video, maximum. Open the retention graph. Write a timestamp and a cause in five words. "2:14 chapter ended as a summary." "0:08 logo sting." "6:10 recap started." If you cannot point at a timestamp, write "slow bleed, no event" and move on. Shape names help if you already use them; audience retention analysis is the short list. Do not open a second tab to reread a taxonomy. The graph in front of you is the input.
Minutes 8–13: the edit decision. Choose one. The test is: if we only change one thing in production this week, what is it? Legal options:
- Cut a range on an existing video that is still getting impressions (dead segment, recap, pre-value ask).
- Move a beat in the next outline (reset 15 seconds earlier, promise check before context).
- Kill a runtime habit ("no more stretching 6-minute ideas to 10").
- Do nothing structural, because the miss is packaging. Write "no edit" explicitly. Blank is not "no edit"; blank is skipped.
If the video still has distribution, a range delete on an EDL timeline is a real repair: check the join on the free 480p preview pass (it carries a short per-user cooldown), and a single charge applies to the final export only. A one-beat offender is a segment retake, not a new shoot. If the miss is the opening, do not recut minute six.
Minutes 13–18: the packaging decision. Choose one. Legal options:
- Replace the thumbnail, or the title, not both at once unless you are using YouTube's Test & Compare.
- Start a Test & Compare on a video that qualifies. YouTube Help excludes Shorts, Made for Kids, mature, private, and age-restricted videos, plus scheduled livestreams and Premieres until they convert. Up to three title and thumbnail variants; the winner is highest watch time, not CTR, and tests complete within two weeks.
- Leave packaging alone, because the curve says the opening or the middle failed after a fair click. Write "no packaging change."
- For the next upload only: lock a thumbnail description before the script, so minute one has something to confirm.
A packaging decision that is actually an edit in disguise ("the thumbnail is fine, we should remake the intro") belongs in the previous slot. Send it back.
Minutes 18–20: log three numbers. Then stop. The log is the only reason week 12 will be smarter than week 2.
If you already use a creative analytics dashboard, this review does not replace it. It is the weekly pass that uses it. Pulling performance inside the agent is a way to see which posts even deserve a graph; the second-by-second curve still lives on the platform.
The three numbers, and nothing else
You can track fifty metrics. You will not look at them. These three, same definitions every week, same format:
| # | Number | Definition you must not change | Why it earns a column |
|---|---|---|---|
| 1 | First-minute hold | Median percentage remaining at 0:60 across this week's long-form in the format you reviewed | Tells you whether Block 1 (promise check) is holding |
| 2 | Body hold | Median percentage remaining at 0:60 minus median remaining at 80% of runtime | Tells you whether the middle is leaking, independent of the opening |
| 3 | CTR | Median click-through on the same set | Tells you whether packaging is feeding the curve a fair audience |
Write them as numbers, not as moods. "61 / 18 / 5.4" for a week is a row. "felt worse" is not.
Body hold is a difference on purpose. If first-minute hold falls and body hold is stable, the middle is not the problem. If first-minute hold is stable and body hold grows (more lost between 1:00 and 80%), the middle is the problem. Average percentage viewed alone cannot separate those.
Rules for the log:
- Same length band. If this week was all 8–12 minutes and last week was a 24-minute sit-down, do not draw a trend line through them. Add a band column or skip the comparison.
- Median, not mean, when you have three or more videos. One breakout will fake a recovery.
- Shorts on a separate three-number row if you insist on reviewing them in the same sitting: viewed versus swiped, average watch time at 24 hours, and whatever you use for the first-frame test. Do not park them in columns 1–3.
A month of rows is the artefact. Week-over-week noise is expected. A four-week slide in body hold, with CTR flat, is a script problem you can finally see. A four-week slide in CTR with first-minute hold up on the people who did click is a thumbnail problem, and recutting the video will not help. That is the whole point of forcing both decisions to coexist: the numbers tell you which slot was lying.
For the packaging slot, the thumbnail craft and Test & Compare mechanics are the two how-tos. Do not open them during the 20 minutes. Open them after, if the decision was "run a test."
What "one decision" looks like written down
The review notes for a week should fit on an index card.
Set: 3 tutorials, 9–11 min. Graphs: #2 drops at 2:20 (chapter summary). #1 and #3 slow bleed. Edit: Next outline, end section 1 on a question. No recut of #2 (impressions already dead). Packaging: Test & Compare on #3, two thumbnails, leave title. Current CTR 7.1 with a 0:40 cliff: overpromise. Log: 58 / 21 / 6.0
That is a completed review. A completed review is allowed to say "no edit" or "no packaging change." It is not allowed to say "think about pacing."
If two videos want two different edits, pick the one that will ship this week. The other goes on a parking list. Parking lists that are longer than three items are a backlog you are not running; delete from the bottom.
FAQ
What if I only uploaded one video this week?
Run the same agenda on that one video. The log gets a sample size of 1, which you should not trend until you have four weeks. The decisions still count. A single graph is enough to name a timestamp and to decide whether the next thumbnail is a test.
Do Shorts belong in the same 20 minutes?
Not in the same three columns. If Shorts are the product, run the 20 minutes on Shorts and park long-form. If long-form is the product, give Shorts five minutes after the timer, or a fortnightly slot. Mixing them in one median is how both look "fine" while each is slipping.
What if the edit and the packaging point at different videos?
Allowed. The edit decision can be about next week's outline (informed by video A) while the packaging decision is a Test & Compare on video B, which is still getting impressions. What is not allowed is two edits, or two packaging changes, to celebrate thoroughness.
How long do I keep the log?
Until the trend is visible, then keep going. Twelve rows is a quarter of weekly uploads and the first time a seasonal dip is obvious. A spreadsheet with three columns survives; a dashboard with twenty does not. If you have not looked at week-minus-eight in a month, the log is too wide.