Clip yield: a throughput baseline per upload

Most repurposing dashboards count posts per week. That number hides the input that actually constrains the system: how much source you put in. A team publishing 12 clips a week from four hours of interviews is doing different work from a team publishing 12 clips from 40 minutes of talking-head. The first is a distribution habit. The second is a yield problem, and you cannot see it until you divide clips by source hours.

Clips per source hour is the metric. Most teams have never computed theirs. This post is how to measure the current number, what a realistic target looks like from vendor-analysed pipelines, and which steps move it without turning the long-form into a clip farm.

Two numbers, not one

Yield and labour are easy to mash into "we should be faster." They answer different questions.

Metric	Formula	What it diagnoses
Clip yield	Published clips ÷ hours of source video, in a fixed window	Whether the asset is being mined
Labour throughput	Published clips ÷ hours of editing and packaging	Whether the process is slow

A 45-minute interview that produces two Shorts is a yield of about 2.7 clips per source hour. A 45-minute interview that produces eight Shorts is about 10.7. Same source, different mining. If both took three hours of editing, labour throughput is 0.7 versus 2.7 clips per labour hour, and that is a separate conversation about tools.

Vendor-analysed pipelines in 2026 often cite a working range of 6 to 10 clips from one 30-minute long-form, in about 90 minutes of work, with steady-state creators publishing 5 to 7 short-form posts a week across platforms from 2 to 3 originals. Treat those as observed ranges from vendor datasets, not as a quota. Converted to yield: 6 to 10 clips from 0.5 source hours is 12 to 20 clips per source hour if you actually ship that many. Converted to labour: 6 to 10 clips in 1.5 hours is roughly 4 to 7 clips per labour hour.

Most operations sit well below the yield number, because they clip by chapter, re-edit each cut from scratch, and never go back to the same source. The first job is to find out which of the two numbers is the bottleneck.

Measure current yield in one sitting

Pick a window you can defend: the last 8 weeks, or the last 12 long-form uploads, whichever is smaller. Mixing a quiet month with a launch week will lie.

1. List every long-form source that shipped in the window: YouTube videos, podcast episodes, webinars, keynotes. Write duration in hours, not minutes. A 42-minute episode is 0.7 hours.
2. Count every short-form post that was cut from those sources, across platforms. A Reel and a Short of the same cut count as two posts for distribution, but one clip for yield. Yield measures unique cuts, not unique URLs. Keep a second column for unique-cut-to-post if you care about cross-posting.
3. Divide unique cuts by source hours. That is current yield.
4. Divide unique cuts by hours of people-time you actually spent pulling, captioning, and packaging. That is labour throughput. If you do not have hours, stop estimating. Track the next four sources with a timer.
5. Split the yield by source type if you have more than one. Interviews, monologues, and screen demos do not yield the same, and a blended number will send you to fix the wrong show.

Write the result as a single line: "Last 12 sources, 9.4 hours, 31 unique cuts, yield 3.3 clips/source-hour; labour not timed." Ugly and true beats a round number you cannot reproduce.

If the number is 2 to 4, you are almost certainly clipping chapters and leaving the rest of the hour on the table. If the number is already in the teens and labour is the pain, stop hunting for more cuts and fix packaging. Content repurposing fails in both directions: under-mining the source, and over-processing each clip.

What a realistic target is

Use the vendor range as a ceiling to test against, not as a KPI for a junior editor.

A defensible near-term target is 6 unique cuts per source hour, which is 3 cuts from a 30-minute video or 6 from a 60-minute one. That is below the 6-10-from-30-minutes cluster, on purpose. High enough that chapter titles alone will not get you there, low enough that you can still reject beats that are not self-contained.

Then set a labour budget. Ninety minutes for 6 to 10 cuts is the vendor observation. If your current labour is four hours for three cuts, the yield target will not save you; the process will. The long-form-to-Shorts pipeline exists so selection, vertical conversion, and captions are one pass. Converting existing footage to Shorts from a master is a yield activity. Rebuilding each clip as a new project is a labour activity wearing a yield costume.

Do not raise weekly cadence until both numbers are visible. Cadence without yield is more days in which the same two clips come out of a one-hour recording. Cadence without labour tracking is how a team "does not have time to repurpose" while spending the time inside a social app.

A separate output of the same measurement: cuts per source that you actually published, versus cuts you exported and sat on. Inventory that never ships is not yield. It is a folder.

The steps that move the number

Yield moves when you change which moments you take, not when you buy a faster encoder.

1. Clip from retention spikes, not from chapters. A rewatch bump on the long-form is a moment people already replayed. Chapter marks are for viewers who are already in the video. Pull the spike, then expand left and right until the beat is self-contained. Audience retention analysis is the input to selection. If you do not have a retention graph (audio-only, or a webinar hosted elsewhere), use the transcript and score for standalone claims instead.

2. Require a self-contained beat. Setup, claim, payoff, inside the cut. If the guest says "as I mentioned earlier," it is not a clip until you pull the earlier mention in or you rewrite a one-line setup on screen. This filter lowers raw count and raises usable yield, which is the number you should be protecting.

3. Transcribe once, select many. One transcript of the source, with timestamps, is the map for every later cut. Getting a transcript from the master is cheaper than watching the hour back at 1.5x for the third time. A selection prompt that earns its place:

"From this transcript, list 8 ranges between 30 and 55 seconds that work as standalone short-form. For each, return start and end timestamps, the setup sentence, the claim sentence, the payoff, and a reject reason if the range needs context from outside itself. Do not propose chapter-length ranges."

Watch the top of that list. Reject anything that only works if you already heard the episode.

4. Cut on one timeline, from one master. Each clip is an in/out on the source, not a new project. Trim and agent-side trim exist so the mechanical cut is not the rate limiter. Vertical conversion, captions, and a hook card happen on the cut.

5. Stop at the first good set, then go back later. Mining a source to exhaustion on day one produces near-duplicates and a long-form that has already given away every turn. Ship the 6, measure, and leave unused spikes for week three. One brand video into many assets is a real pattern. Thirty assets from one video is a different operation (stills, carousels, quotes) from thirty near-identical 40-second cuts.

6. Keep labour in the same pass as yield. Caption style, safe-zone text, and a clean export belong in the cut session. A second session "to make them native" is how 90 minutes becomes an afternoon.

Recompute yield every quarter against the same definitions. If unique cuts per source hour rises and completion on those cuts falls, you have started shipping fragments. Yield is only a good number when the clips still work as clips.

FAQ

Should a clip posted to three platforms count as three toward yield?

No. Count unique cuts for yield, unique posts for distribution. Otherwise a team that cross-posts looks three times as productive as a team that mines more of the hour. Track both. Do not add them.

What if the long-form is a 12-minute video, not an hour?

The formula does not care. Twelve minutes is 0.2 hours. Two unique cuts is a yield of 10. Very short long-form often has fewer self-contained beats, so a lower yield can still be correct. Compare like with like: 12-minute sources against other 12-minute sources, not against podcasts.

Is there a point where more clips per hour starts hurting the original video?

Yes. If every turn in the long-form is already circulating as a Short, the remaining reason to click the original is the connective tissue, and some viewers will not. Ship the spikes that stand alone, and leave the through-line in the full video.

Do generated B-roll inserts count as extra yield from the same source?

They count as extra posts if they are new videos. They do not count as extra yield from the recorded hour unless they are cut from that hour. Keep generated pieces on a separate line.