Why this matters
When views stall, teams change the thumbnail, the posting time and the topic at once. The retention report says which is the problem, often none of them. The retention curves guide gives the vocabulary; this exercise asks you to use it across several videos, where the useful findings repeat, then test the strongest one with paid money rather than opinion.
The school and its stalled views
Grainline Woodworking School is an invented school selling online courses in hand-tool joinery, with a free taster lesson as its main sign-up. Its channel posts long tutorials and Shorts, and it runs a paid video ad that opens on its logo and sends viewers to the taster lesson.
Grainline Woodworking School: four recent videos at seven days
| Video | Length | Click-through rate | Still watching at 0:30 | At half-way | At the end |
|---|---|---|---|---|---|
| A: Cutting dovetails by hand | 22:00 | 6.1% | 58% | 33% | 21% |
| B: Sharpening chisels properly | 16:00 | 3.2% | 78% | 44% | 30% |
| C: Short, dovetail fit reveal | 0:38 | Viewed against swiped: 71% | Average viewed 104% | ||
| D: Short, workshop tour | 0:52 | Viewed against swiped: 44% | Average viewed 52% |
- A opens with a logo animation, a welcome and a spoken tool list until 0:50. Subscribers still watching at 0:30: 74%; suggested viewers: 51%.
- A has a sharp dip from 6:10 to 7:00 during a tool-shop sponsor read that recovers, and a spike at 14:20 where the joint slides together.
- B has a sharp dip at 9:30, where the host says "that's the main method done, the rest is extras", and it does not recover. B's relative retention is above average throughout; A's is below average in the first minute.
- D opens on a door swinging open onto an empty workshop.
- Paid ad: £24 per taster sign-up, about 40 sign-ups a week, at a CPM of £6.
Reading drops, dips and spikes
Each shape has a small set of causes, and the filters decide between them. A steep early drop with a healthy CTRShare of impressions that result in a click. points at the opening, not the packaging. A weak click-through rate with strong retention points the other way. A dip that recovers is a skipped segment; one that does not recover is a reason to leave. A spike is a moment worth rewatching, and usually the seed of the next Short.
Task
- Diagnose the intro drop on A and the click-through and retention pattern on B, using the traffic source split.
- Explain each dip and the spike, and say what each one means for the edit.
- Compare A and B with relative retention, and compare the two Shorts.
- Write the script and edit changes as instructions an editor can follow.
- Design a hook test for a paid version of the best clip.
- Set a weekly production rhythm that feeds the findings into the next videos.
Worked answer
A's intro. A strong click-through rate followed by a steep fall is a broken or delayed promise: the title sells dovetails and the first fifty seconds show a logo and a list. Subscribers forgive the slow start; suggested viewers, the ones the channel needs, do not. Open on the finished joint and the promise, and move the tool list to the screen and the description.
B's packaging. B holds viewers better than A on relative retention, so the content is sound and the problem is upstream: few people choose to click. Rework the title and thumbnail around the visible result, a mirror-polished edge, before touching the edit.
Dips and the spike. A's sponsor dip recovers, so viewers skip it: shorten it and move it after the first payoff. B's dip is an announced ending; never tell viewers the useful part is over, and move the extras into their own chaptered video. A's spike is the joint fitting together, which viewers rewatch: slow it down, chapter it, and build the next tutorial's opening around that kind of moment.
Shorts. C opens on its payoff and loops, so its first second works; D's opening frame shows nothing worth stopping for. Short-form retention is read for the loop and the first moment, as the Shorts and long-form comparison explains.
Edit instructions. Open every tutorial on the finished result within the first seconds. No spoken tool lists. Sponsor reads after the first payoff and under a set length. Never announce the end before it arrives. End within seconds of the last step, with the end screen over content. Every Short opens on the payoff.
Hook test. Following the ad creative hook test design, cut C into a paid ad and change only the opening: the joint sliding home against the line "Gappy dovetails?" over a poor joint, with every later frame identical. Use the platform's split test, equal budgets, the same audience and the taster sign-up as the event. Judge on cost per sign-up, with post-click conversion and frequency as guardrails, and read only at the planned end.
Hook test budget
- The team's sample size calculation calls for about 120 sign-ups per arm to detect the difference worth acting on, a cost per sign-up around £20 or lower in the challenger.
- At £24 per sign-up, that is about £2,880 per arm, or £5,760 in all.
- At the current rate of about 40 sign-ups a week across both arms, that takes six weeks. Doubling weekly spend for the test brings it to three, if the audience can absorb it without frequency climbing.
Rhythm. Following the creative production playbook: on Mondays, review last week's curves at seven days with the same traffic filters and annotate them; on Tuesdays, turn findings into changes to the script template and edit checklist and brief the next tutorial and the Shorts cut from its spikes; film and edit midweek; publish at the end of the week. Every fortnight, read live A/B TestControlled comparison of two variants. results and record them in the creative library under one naming convention; every month, refresh the concept backlog.
Common mistakes
- Fixing the thumbnail when the opening is the problem. Read click-through against the early drop.
- Comparing raw watch time across lengths. Use relative retention for craft, absolute to find the second.
- Testing two different ads. Only the opening may change.
Test this in the content diagnosis checkpoint, or revisit platform playbooks at the intermediate social media level.
