Keep rate is the share of the ads you produce that you actually publish: published divided by produced. If you publish 60 percent of what you produce, shipping 40 ads means producing 67, and every cost in your production chain scales off 67 rather than 40. Almost nobody tracks it, which is why so many teams have a cost per ad figure that is wrong by a factor of two.
It is a production metric, not a media metric: it says nothing about whether your ads work. What it sets is the multiplier on everything you spend before an ad reaches a platform, which is what decides whether cheap generation is actually cheap.
The definition and the formula
keep rate = ads published / ads produced
cost per published ad = (brief + production + review cost) / keep rate
Three definitional decisions matter more than the arithmetic:
Count an output as produced the moment it exists as a reviewable asset. Not when it is exported, not when it is uploaded. If a human had to look at it, you paid for it.
Count it as published when it goes live with spend behind it. An ad sitting approved in a folder is not published, and a "saved for later" pile is the most common way teams flatter this number.
Measure per batch and report per month. Per batch is where you can act (this brief produced 12 outputs and 5 went live, so what was wrong with the brief). Per month is what you put in the budget.
One refinement worth keeping: report first-pass keep rate (excluding regenerations) next to all-in keep rate, because the gap between them is the cost of your revision loop.
Why it dominates cost per published ad
Because it is a divisor. Every other lever in production is a subtraction from one term, and this one multiplies all of them at once.
Take a team publishing 40 ads a month, in-house, one hour of editing per ad, and a loaded hourly cost of about EUR 45 (salary plus employer overhead, not the salary line alone). Review is five minutes per output. Nothing changes across these columns except the keep rate.
| Keep rate 35% | Keep rate 60% | Keep rate 80% | |
|---|---|---|---|
| Produced to publish 40 | 115 | 67 | 50 |
| Production hours | 115 | 67 | 50 |
| Production cost | about EUR 5,175 | about EUR 3,015 | about EUR 2,250 |
| Review hours at 5 min | 9.6 | 5.6 | 4.2 |
| Review cost | about EUR 432 | about EUR 252 | about EUR 189 |
| Monthly total | about EUR 5,607 | about EUR 3,267 | about EUR 2,439 |
| Cost per published ad | about EUR 140 | about EUR 82 | about EUR 61 |
The first and third columns differ by more than EUR 3,000 a month for identical output. No rate was negotiated and no tool was bought. The only difference is how much of the work survived review.
Now compare that with the levers teams usually reach for. Cutting editing time by 20 percent saves about EUR 600 a month at a 60 percent keep rate. Moving keep rate from 60 to 80 percent saves about EUR 830 and improves the work at the same time, because the mechanism is a clearer brief rather than a faster hand. Our video ad cost calculator runs this with your own hourly figure.
There is a second-order effect worth naming. A low keep rate does not only cost money, it costs cadence. If review rejects half of everything, you either ship half as many ads or spend twice as long producing, and both outcomes push against creative decay. Our 2026 ad fatigue benchmark puts CTR decline at 15 to 20 percent in a creative's first two weeks, with week three landing 45 to 70 percent below the launch baseline. A poor keep rate is therefore a media cost as well as a production one.
Typical ranges, and what each one means
These are ranges from running and advising creative teams, not survey findings.
| Keep rate | What is usually happening | The fix |
|---|---|---|
| Under 40% | No written brief, or constraints discovered during review. Reviewers rejecting on taste with no shared standard | Write the five-field brief. Move brand and legal rules into it |
| 40 to 55% | Brief exists but the offer is vague, often "learn more", so outputs have nothing specific to say | Fix the offer. This is a commercial decision, not a creative one |
| 55 to 75% | Normal for a competent team with a written brief. Rejections are mostly legitimate variety pruning | Tighten the constraints block, cap the revision loop |
| 75 to 85% | Good. The brief specifies what publishable means before production starts | Watch for the failure mode below |
| Above 90% | Suspicious. Either the review is a rubber stamp, or the brief is so narrow that every output is a near-duplicate | Widen the brief. Near-duplicates raise frequency and produce no learning |
That last row is the one people misread. Keep rate is not a metric to maximise: a team publishing everything it produces is either not reviewing or not exploring, and the second shows up within a month as eight variations of one ad and frequency climbing past 3.
The target I would set is 70 to 80 percent, with a deliberate share of exploratory outputs you expect to bin.
How to raise it
In order of effect, roughly the reverse of the order teams try:
- Put constraints in the brief, not in review. Banned claims, required disclaimers, logo rules, competitor mentions, ratios, language. A constraint stated up front is an instruction; the same constraint applied in review bins a finished asset. This single change usually moves keep rate from the forties to the seventies.
- Define the offer precisely. Vague offers produce ads that fill the gap with adjectives, and adjectives are what reviewers reject when they say an ad feels generic.
- Review against the brief, not against taste. Six checks: claim accuracy, disclaimer, offer wording, audio, angle match, ratio and length. Taste-based review has no floor.
- One accountable reviewer per set. Committees reject more and explain less.
- Cap the revision loop at one pass. An output needing two attempts is an output the brief failed to specify. Our regenerate or edit guide covers the choice between a fresh attempt and a small manual fix, and the fix is frequently the cheaper answer.
- Fix the footage before the briefing. If outputs get rejected for looking alike, the shortage is source material, usually in the hook and call-to-action roles. No brief solves a thin library.
How a low keep rate makes cheap generation expensive
Say generation costs about EUR 1.52 per variation, which is one credit on our EUR 99 Growth pack, and review stays at five minutes per output. Genyad is our product, so substitute your own numbers, but the shape holds for any tool where output is nearly free.
| Keep rate 35% | Keep rate 60% | Keep rate 80% | |
|---|---|---|---|
| Produced to publish 40 | 115 | 67 | 50 |
| Generation cost | about EUR 175 | about EUR 102 | about EUR 76 |
| Review hours at 5 min | 9.6 | 5.6 | 4.2 |
| Review cost at EUR 45 loaded | about EUR 432 | about EUR 252 | about EUR 189 |
| Total | about EUR 607 | about EUR 354 | about EUR 265 |
| Share of cost that is review | 71% | 71% | 71% |
Look at the last row. Once generation is nearly free, review is roughly three quarters of your cost regardless of keep rate, and the total still nearly doubles between 80 and 35 percent. Cheap production does not rescue a bad keep rate. It changes what a bad keep rate costs you from money into attention, and attention is the scarcer resource.
The trap I see most often: a team adopts a generation tool, produces four times as many outputs, keeps the same review process, and discovers that the bottleneck simply moved. They now spend fifteen hours a month reviewing instead of four, publish only slightly more, and conclude the tool did not work. The tool worked. The brief did not, and the review process was designed for six ads a month.
Measuring it, starting this week
Four columns in a spreadsheet: brief name, produced, published, date. That is the whole instrumentation. Add regenerations as a fifth column if you want the gap between first-pass and all-in keep rate.
Run it for four weeks before you change anything, because the first number is usually a surprise. Most teams guess 80 percent and find 45. Then change exactly one thing, the constraints block, and measure another four weeks. Our creative testing workflow sets out where the produced and published counts come from in a weekly cycle, and the weekly report is where this number belongs, not a quarterly review.
One caution about incentives. Make keep rate a target for the reviewer and they will simply approve more; make it a target for the briefer and briefs narrow until outputs repeat. It belongs to whoever owns both the brief and the read, and it should be read next to published volume, never on its own.
Frequently asked questions
What is keep rate in ad production?
Keep rate is the number of ads you publish divided by the number you produce, measured per batch and reported per month. An output counts as produced as soon as a human has to review it, and as published only when it goes live with spend behind it. It measures production efficiency, not ad performance.
What is a good keep rate?
Between 70 and 80 percent for a team with a written brief and a constraints block. Under 55 percent usually means brand or legal rules are being discovered in review rather than stated in the brief, or the offer is too vague for outputs to say anything specific. Above 90 percent is a warning sign, because it means either the review is a formality or every output is a near-duplicate of the last.
Why does keep rate matter more than cost per ad?
Because it divides cost per ad. Producing 115 outputs to publish 40 costs about EUR 140 per published ad on the same rates that give you about EUR 61 at an 80 percent keep rate, so the quoted per-ad figure is meaningless without the ratio. It is also the only production lever that lowers cost and improves the creative at the same time, since the mechanism is a clearer brief.
How do I raise our keep rate?
Move every constraint that could cause a rejection into the brief, make the offer specific, and review against six fixed checks rather than against taste. Give each output set one accountable reviewer and cap revisions at a single pass. If outputs keep getting rejected for looking too similar, the problem is a thin footage library rather than the brief.