How Many Revisions Does a Video Ad Take? Data From 999 Projects

Across 742 Filmito projects with numbered cuts, the median video ad was approved on cut 3: a first cut plus two rounds of revisions. Fewer than one in five was approved on the first cut, 45.6% were done by cut 2, 66.2% by cut 3, and 3.6% needed nine cuts or more. We pulled those numbers from our own archive at the end of August 2026 and published the full set on filmito.io/en/ai-ads-report, so what follows is not an industry estimate. It is what actually happened on our projects, including the ones that took a long time.

The full distribution, cut by cut

First, what we counted. Our archive holds 999 client projects with video deliverables and 9,721 finished video files, about 9.7 files per project. Of those 999, 742 used numbered versions in the file names (cut 1.0, cut 2.0 and so on), so those are the projects we can say something honest about. The rest were single-file deliveries or named in ways we cannot reconstruct, and we left them out rather than guess.

• Cut 1: 19.0% approved
• Cut 2: 26.5%
• Cut 3: 20.6%
• Cut 4: 12.4%
• Cut 5: 8.2%
• Cut 6: 4.2%
• Cut 7: 3.4%
• Cut 8: 2.0%
• Cut 9 or more: 3.6%

Cumulatively, 45.6% of projects were approved by cut 2 and 66.2% by cut 3. The median is cut 3.0. Read the shape, not just the peak: it climbs to cut 2, then falls steadily for a long way. There is no cliff. Every extra round moves some projects to approval and carries the rest forward.

One caution about what a cut is. A numbered version is any file we sent with a new number, and some of those are tiny: a typo in a super, a logo swap, a different end card. We add on-screen text in post rather than inside generated footage, so a copy fix is a fast new version, not a new film. That means the counts above are generous. The real number of sit-down-and-rethink rounds is lower than the number of cuts.

Why two included rounds is the honest number

Our Launch ads on filmito.io/en/pricing come with a first cut in 3 days and two revision rounds included. People sometimes ask why not one, or why not unlimited. The data answers both.

If we included one round, only 45.6% of projects would finish inside the included work. More than half of our clients would hit a paid round on a perfectly normal project, and a normal project should not feel like an overage.

If we included unlimited rounds, we would have to price the tail into everyone's ad. The 3.6% of projects that need nine or more cuts would be quietly paid for by the 19.0% that approve on the first one. That is not fair, and it also removes the small, healthy pressure that helps a project converge.

Two rounds cover 66.2% of projects. Two-thirds of the people who buy an ad from us never need to think about revision pricing at all, and the median client uses exactly what is included. That is the line where the policy matches the data instead of the marketing. Anything beyond it is a conversation before we cut, not a line on the invoice after.

“A revision is not the first cut failing. It is the moment a client sees the idea on screen and finally knows what they wanted.”

What the nine-plus tail looks like

3.6% is small, but in a sample this size it is real projects with real people on the other end, and every one of them was finished and approved. The pattern is rarely a bad first cut. It is usually one of three things.

• The brief changed while the film was in progress: a new price, a renamed product, a feature that got dropped the week before launch.
• More than one person was approving, and they did not agree with each other. Round 4 undoes round 3, round 5 undoes round 4.
• Regulated categories. In pharma work every round can carry MLR compliance edits, and those come from a review committee, not from taste. We build MLR-aware scripts and supers into the masters, but a legal note is still a new cut.

Why do we keep cutting instead of stopping at a limit? Because the film carries the client's name, not ours, and an ad that is 90% right is still an ad that cannot run. Our Launch tier carries a money-back promise, and we would rather do cut 9 than return money on a film that was two notes from finished. AI production also makes this less painful than it sounds: single generated takes hold for roughly 5 to 8 seconds and our films are cut from short takes, so a late change usually means regenerating one shot, not a day back on set.

What we will say plainly: if your approval process has five sign-offs and no owner, two included rounds will not be enough for you, and a fixed-price studio is probably not the right fit. A retainer with an in-house editor, or a tool your own team operates, will hurt less. We would rather tell you that here than discover it together on cut 6.

A checklist for getting to yes by cut 2

45.6% of projects get there. Here is what the ones that do tend to have in common, seen from our side of the timeline.

Brief quality

• One sentence on who the ad is for and what they should do after watching it.
• The must-haves: logo, product name, price, legal line, end card. The must-nevers: colours, claims, competitor names.
• Two or three references, even bad ones. Not like this is useful information.
• Lock the script before we generate. In AI production a script change is new footage, not a re-edit.

A short brief is not a weak brief. The entire client input for filmito.io/en/case-studies/be-the-kaptaan was one photograph of the can, and it became a 72-second brand epic plus three retargeting films, 8 finished files in one month. It worked because everything that mattered was already decided: the name, the strength, the heritage line printed on the can. What slows a project is not a thin brief. It is an undecided one.

One feedback channel

• Name one person who says yes. Others can comment, but their notes go through that person.
• Give feedback in one place: one email thread or one shared document, not chat plus email plus a call.
• Watch the cut on the device it will run on before writing notes. A vertical edition viewed on a laptop lies to you.

Consolidated notes

• Send one set of notes per round, with timecodes, after everyone who needs to watch has watched.
• Separate must change from nice to have. We will do both, but knowing which is which keeps one round from becoming two.
• Copy fixes are cheap and fast; scene changes mean new generations. If a note can be solved with text, say so.

What this data cannot tell you

A few honest limits. This is one studio's archive, so it describes our clients and our process, not the industry. Timestamps before 2021 are unreliable, so we do not publish a per-year trend and we will not guess at one. And the sample counts numbered files, so it cannot tell a hard creative round from a one-word super fix. If someone quotes you a global average for video ad revisions, ask where the files came from.

If you want to see how your own project would run, the quote flow at filmito.io/en/see-your-ad gets you the idea, the price, and the date within one business day, and the full report with every chart is on filmito.io/en/ai-ads-report. Bring a decided brief and one person who can say yes, and cut 2 is a reasonable place to aim.