Spend 30% of your time on ideation
The gap between top and small creators is time spent on ideation and packaging, not kit. Track your split and reallocate without cutting weekly output.
Paddy Galloway's most useful number is not a retention target. It is a calendar. In his Colin & Samir conversation he puts top creators at roughly 30% of their time on ideation and packaging, and small creators at roughly 5%, with the remaining 95% going into filming and editing. The gap is not a better camera and it is not a newer model. It is where the hours sit.
That is why the 30% belongs on ideation and packaging as one bucket. A title, a thumbnail and a first frame are the decision the audience makes before your cut gets a vote. Spending the week inside the timeline is how people get very good at a step the viewer has not reached.
The job is not "make less video." The job is to move hours out of production waste and into the decisions that decide whether the video is opened.
The gap is not the camera
Two creators can generate the same length, on the same tools, at the same credit cost, and get different channels out of it. The difference that shows up in public work is not fidelity. It is whether the file was worth packaging.
Ideation is choosing the promise: the specific claim, the payload, the reason this exists rather than a neighbouring idea. It is a list with kills on it, not a brainstorm that ships everything.
Packaging is the first frame, the title, the thumbnail or cover, and the on-screen text that restates the promise. MrBeast's leaked production document assigns minute one a single job: confirm what the thumbnail promised. If packaging and opening disagree, you paid for a click you immediately refund.
Small-creator default is the reverse order. Film or generate first, because that feels like work. Write the title in the upload box. Grab a frame at 0:04 because it is in focus. Then wonder why a carefully edited middle never gets watched. The 95% created a file. The 5% decided whether anyone would start it.
You do not need Galloway's room to use the split. You need a week of honest labels on your own hours.
Track the four buckets for one week
Four buckets, logged in 15-minute blocks, for seven days of work you were going to do anyway. Do not run a special "measurement week" with different habits.
| Bucket | What counts | What does not |
|---|---|---|
| Ideation | Listing payloads, killing ideas, writing the promise in one sentence | Scrolling for inspiration with no list at the end |
| Packaging | Titles, thumbnails, cover frames, first-frame spec, on-screen claim | Recutting the body because the title overpromised |
| Production | Generating, filming, directing the agent, picking takes | Re-rolling a finished body because you have not chosen a title |
| Edit | Assembly, captions, sound, colour, export | Redesigning the thumbnail on the timeline |
At the end of the week, sum the hours. Ideation plus packaging is the number you compare to 30%. Production plus edit is the rest.
Most people who think they "spend a lot of time on ideas" discover that the hours were production: generating six bodies for one slot, then picking the least bad. That is not ideation. That is sampling. Ideation is the kill list before the first generation.
If you cannot log in 15-minute blocks, log per episode instead. For each published file, estimate the four buckets to the nearest half hour. Ten episodes is enough to see whether you are in the 5% or somewhere on the way to 30%.
A productivity system that still starts at "open the generator" will reproduce the 5%. The order has to change, not the tool.
Steal time without cutting the calendar
The 25 points between 5% and 30% cannot come from dropping output. They have to come from production that was not load-bearing. Four places the hours usually are.
1. Bodies generated before a promise exists. Pick the payload and write the title first. Then generate once toward that promise. A hook pack before the video exists is the extreme form of this: you test the open without building the middle. Even a paper version helps. Three candidate titles and one chosen promise should exist before a credit is spent on the body.
2. Recutting packaging problems. A video with strong retention and weak click-through did not need another afternoon in the timeline. It needed a new title and a new thumbnail. Diagnose before you recut. If the drop is in the first 30 seconds, the opening and the packaging disagree, and the cheaper fix is usually to move the payoff earlier so minute one confirms the promise. Diagnosing a weak opening joins the timestamp to what was on screen. Recutting the middle of a packaging miss is the 95% defending itself.
3. Thumbnail as an afterthought. Generate three cover concepts per upload, not one frame grab. YouTube's native test allows up to three title and thumbnail variants, picks a winner on watch time rather than click-through, and can resolve inside two weeks. Shorts are not eligible for that tool, so off-YouTube you change one packaging element at a time on consecutive days. The craft of thumbnails that earn the click is a 30% activity. A thumbnail generator makes the three-concept rule cheap enough to keep.
4. Re-rolling instead of specifying. Identity drift, a bad hand, a wrong first frame: inpaint or re-specify the shot rather than regenerating the episode. The 95% loves a full re-roll because it looks like standards. It is often a missing first-frame spec.
None of those four reduce the number of episodes this week. They stop you building the episode twice. Output holds. The mix of hours moves.
On an EDL timeline a packaging change is an instruction-list edit, not a rebuild. The 480p preview pass is free of a credit charge and carries a short per-user cooldown, so use it to check that the new first frame and the title still agree. The final export is charged once, regardless of how many times you previewed.
A weekly loop that holds the 30%
A target without a loop returns to 5% by Friday. This is the loop.
Monday, 90 minutes, no generation. Write twelve payload lines for the week. Kill nine. The three survivors get a one-sentence promise each. If you cannot kill nine, the list is too flattering. This block is ideation. Protect it the way you protect an upload slot.
Monday, next 60 minutes, still no body. For each survivor: three titles, three thumbnail sketches, one specified first frame. Pick a package before you pick a model. This block is packaging. Together with ideation that is two and a half hours, which is about 30% of an eight-hour week. If your week is twenty hours, these two blocks need to grow with it, not stay at two and a half hours while production absorbs the rest.
Then production. Generate toward the chosen promise. One body per slot unless the preview fails a spec on the bible (wrong first frame, length out of band, identity break). Failures get a local fix, not a new concept.
Then edit. Captions, sound, colour, export. If you are writing a new title during export, the Monday block failed and the 30% leaked.
Friday, 20 minutes. Sum the four buckets. If ideation plus packaging landed under 20%, name the production task that stole the hours and cut that task next week. Do not "try to be more strategic." Remove a re-roll habit.
The A/B protocol belongs inside the packaging block, not inside the edit. You are testing titles, first frames and covers. You are not testing six different middles. Hook rate is the metric for the open; watch time is what YouTube's own thumbnail test optimises for. Those can disagree. Decide which question you are asking before you look.
Thirty percent is a prior, like 70/20/10. Once you have six weeks of logs, you will know whether 25% or 35% is the number that holds output on this channel. The failure is staying at 5% and buying another light.
FAQ
Does 30% mean I should publish less?
No. The reallocation is inside the same week of output. If you currently ship three episodes, you still ship three. You spend the missing hours on the kill list and the package, and you find them by not generating the two bodies you were going to discard. If a week cannot fit both the 30% and the current cadence, the cadence was already overstuffed and the files were arriving unpackaged. Cut cadence only after the log shows production has no waste left to steal.
What counts as packaging on TikTok or Reels, where there is no thumbnail test?
The cover frame, the first frame, the first two lines of on-screen text, and the caption. That is the whole package. Change one of them at a time. Consecutive days, same clock time, pre-committed metric. There is no native three-variant tool on those surfaces, which is why the Monday block has to produce the variants before you post, not after.
I generate faceless video. Do I still need 30%?
More, not less. Faceless files compete entirely on promise and first frame, because there is no familiar face to do the recognition work. A faster generator does not replace the kill list. It makes the 95% cheaper to overproduce. The 30% is the brake.
Can the agent do the ideation block?
It can draft twelve payload lines. It cannot kill nine in your voice without you. Use it to expand a tight brief ("twelve payloads for the locked series skeleton, no presenter, length 28–34s"). Then you kill. If the agent is also choosing the title at upload time, you are back at 5% with better copy.