You scale ad creative production by making review cheap, not by making production cheap. Production is already cheap: a video ad variation built from footage you own costs about a euro and a few minutes of machine time. What still costs real money is the human who has to watch each one and decide whether it ships, and if that person spends twelve minutes per variation, forty variations a month is a full working day of somebody's attention plus whatever the rework loop adds.
That is the whole shape of the problem once you automate generation. The queue does not back up at the render farm. It backs up at approval.
Where the bottleneck moves once production is automated
Before automation the constraint is obvious: an editor turns out a handful of finished videos a week, so the campaign gets a handful. After automation the same team generates forty before lunch, then finds it has no process for looking at forty of anything.
Do the arithmetic on your own review habit:
review hours per month = variations generated × review minutes each / 60
At 40 variations and 12 minutes each, that is 8 hours. At 3 minutes each, 2 hours. Nothing about the creative changed between those numbers. The second team stopped treating each variation as a project with a feedback round and started treating the batch as the unit.
Our 2026 ad fatigue benchmark, a synthesis of published platform and agency figures rather than our own measurement, concludes that throughput rather than talent is the bottleneck: unique concepts shipped per month predicts campaign longevity better than the quality of any single ad. It associates shipping 15 to 50 variants a month with 3 to 5 times longer campaign lifespan than quarterly refreshing. Fifty a month at twelve minutes of review each is ten hours, which is why teams quietly settle back to fifteen.
Batch review instead of per-ad review
Per-ad review means opening one variation, watching it, writing notes, sending it back, waiting. Batch review means one sitting, fixed gates, a binary outcome. Rejects are replaced rather than fixed, because a replacement costs one credit and a revision round costs a day.
The process that works for us and for the agencies on our agency creative workflow setup:
- Generate the whole batch first. Never review while generating. Sixteen at a time is a good unit.
- First gate: first two seconds, sound off, thumbnails only. Scan the opening frames as a grid. Anything with a weak or repeated opening frame is out before you watch a second of it. This removes roughly a third in about ninety seconds.
- Second gate: watch the first three seconds with sound. Does the first line say something specific? Reject vague openings without writing why.
- Third gate: claim check. Watch in full only what survived, checking for factual and legal problems rather than taste. This is the gate where a human is irreplaceable.
- Log a reject reason from a fixed list. Weak opening frame, vague first line, wrong claim, off-brand footage, duplicate of another variation in the batch. Five codes, no free text.
- Ship the survivors as one upload. One session, one batch, one set of exports.
Two rules make this hold: no revision requests, only rejects, and no individual notes. The reject codes are the notes. If eleven of sixteen come back "vague first line", that is one brief problem to fix upstream, not eleven pieces of feedback.
Keep rate is the number to track
Keep rate is kept variations divided by generated variations, per batch.
keep rate = variations shipped / variations generated
Generate 40, ship 22, keep rate is 55 percent. Track it per batch and per reject code. It is more useful than any output metric because it tells you where to spend your next hour, and unlike CTR it is available the same day.
| Keep rate | What it usually means | What to change |
|---|---|---|
| Above 80% | Your prompts and footage are too narrow, so the batch has little spread and you are approving near-duplicates | Widen the brief. Ask for hooks you expect to reject |
| 50 to 70% | Healthy. Enough spread to be testing something, enough discipline to reject | Nothing. Hold the cadence |
| 30 to 50% | Workable but the input is thin, usually not enough usable footage per concept | Upload more source footage before generating more variations |
| Below 30% | The brief or the source library is wrong, and reviewing harder will not fix it | Stop generating. Fix the brief, re-shoot or add footage |
| Below 30% with one dominant reject code | A single systematic fault, not a quality problem | Fix that one thing, regenerate the batch |
The trap is optimising keep rate upward. A team at 90 percent is generating timid variations: a batch where you reject half is a batch that explored something. We hold the 50 to 70 band and treat a rising keep rate as a warning.
What a kept variation actually costs
Production cost per variation is the number people compare, and it is the small term. The honest number is cost per kept variation, which includes review labour and divides by keep rate:
cost per kept variation = (production cost + review minutes × reviewer cost per minute) / keep rate
A reviewer on a fully loaded EUR 60 an hour costs EUR 1 a minute, which makes the arithmetic easy to follow. Four routes, same 40-variation month:
| Route | Production per variation | Review minutes | Keep rate | Cost per kept variation |
|---|---|---|---|---|
| Freelance editor, brief per video | EUR 180 | 25 | 90% | EUR 228 |
| In-house editor on salary | EUR 113 | 15 | 85% | EUR 150 |
| Credit-based footage agent (ours), per-ad review | EUR 1.52 | 12 | 55% | EUR 25 |
| Credit-based footage agent (ours), batch review | EUR 1.52 | 3 | 55% | EUR 8 |
The bottom two rows are the argument. Identical production cost, identical output, and the review process alone moves cost per kept variation by a factor of three. Automating generation and keeping a per-ad review ritual gets you most of the cost and none of the throughput.
Check the in-house row yourself: EUR 4,500 a month of loaded salary over 40 variations is EUR 112.50 each, plus 15 review minutes at EUR 1, over a 0.85 keep rate. Substitute your own salary, volume and review minutes, or let our video ad cost calculator run it.
Genyad is our product, so treat the bottom two rows as disclosure rather than a neutral benchmark. You upload a footage library once, it transcribes and tags every clip, and each variation is a fresh script with its own shot selection, voiceover and caption set drawn from that library rather than a re-cut of one timeline. One variation is 1 credit, all four export ratios included, and the 65 credits on the €99 Growth plan work out at €1.52 each, which is where the table figure comes from. Editing, re-exporting and uploading footage cost nothing, which is what makes rejecting cheap. Full detail sits on our pricing page.
We do not do several things a scaling plan might assume: no AI avatars or synthetic presenters, no static banner formats, no product-URL import, no product-feed or CSV-driven template rendering, no predicted performance scores, and no publishing directly into Meta or TikTok, so a human uploads the exports. Nothing tells you which variation will win, which is why the review process above exists.
When hiring an editor is still the right answer
If you need one hero film a quarter, a brand piece with original shooting, or anything where the footage does not exist yet, hire an editor and a crew. Volume tooling solves the next problem along, turning footage you already own into enough distinct variations to keep a rotation fed. A manual editor such as CapCut (free, $9.99 a month for Pro as publicly listed in August 2026, and prices move) is fine for a handful of variations a month, where a review process is overhead.
Frequently asked questions
What is a good keep rate for generated ad creative?
Between 50 and 70 percent per batch. Below 30 percent means the brief or the source footage is the problem and generating more will not help. Above 80 percent usually means the batch lacks spread, so you are shipping near-duplicates and testing less than you think.
How long should reviewing one video ad variation take?
About three minutes as a batch average, including the ones you reject in seconds from the thumbnail grid. Getting there means reviewing in gates, rejecting instead of requesting revisions, and logging a code rather than writing notes. Teams that spend twelve minutes per variation are usually running a per-ad approval ritual designed for expensive creative.
Does scaling creative volume mean quality drops?
Quality per variation drops and quality of what ships does not, provided you actually reject. A batch of sixteen where you ship seven gives you seven pieces chosen from a spread, rather than three you commissioned and then had to defend. Our 2026 fatigue benchmark makes the same point from the media side: unique concepts shipped per month predicts campaign longevity better than the quality of any single ad.
Can one person handle creative for several accounts this way?
Yes, and that is where batch review earns the most. One reviewer at three minutes a variation clears about 160 variations in an eight-hour week, enough for several accounts on weekly cadences. The limit is usually source footage per account, not reviewer time.