What is ROAS — and why does everyone measure it?
ROAS — return on ad spend — is the revenue attributable to your ads divided by what you spent on those ads. Spend $1,000 on Instagram Reels ads, generate $4,000 in tracked sales, and your ROAS is 4, often written as 4:1 or 4x. It is the most-watched number in paid social because it answers the only question a business owner actually cares about: did the money come back, and how many times over?
So what is a good ROAS? Honest answer: it depends, and anyone quoting a precise industry benchmark is usually selling something. Your margins, your product price, your repeat-purchase rate and how much of the sale your tracking actually captures all move the goalposts. As a rough qualitative guide, many performance teams consider a low single-digit ROAS healthy for e-commerce — enough to cover product cost, media and overheads and still leave profit on the table.
Two caveats before the case study. First, ROAS only counts revenue a platform can attribute to an ad, so it usually undercounts the real effect — people see a video, remember the brand, and buy three days later through Google. Second, ROAS is a ratio, not a volume. A 20x ROAS on $50 of spend is a rounding error; a 3x ROAS on $50,000 might fund your whole quarter.
There is a third point, and it is the thesis of this entire article: the biggest lever in video ad ROAS is rarely targeting or bid strategy. It is the creative itself. What follows is the campaign that proved it to us.
The campaign: 24 million views on a $700 budget
The client is a watch-strap brand serving the AP x Swatch community — the collectors who queued for the Audemars Piguet × Swatch collaboration and then wanted to make the watch their own. It is a passionate, visually literate audience that lives on Instagram, photographs wrists obsessively, and knows exactly what a cheap product shot looks like.
The brief was straightforward: sell straps. Not build brand awareness, not grow followers — move product. The budget for paid media was $700 in total. That figure is not a typo, and it shaped every creative decision we made.
The challenge was sharper than it sounds. A strap for a hype watch is a considered purchase sold to people who scroll past ordinary ads on reflex. Winning that audience with media pressure was never an option on this budget; the videos would have to be good enough that the community wanted to watch them, and share them, on their own.
We chose Instagram Reels as the only channel. The community was already there, watch content performs when it looks beautiful in motion, and Reels is the one placement where a genuinely strong video can earn distribution far beyond what the ad spend pays for. With $700, organic reach was not a nice-to-have. It was the strategy.
One week after launch, the campaign had generated roughly 24 million views, around 2,000 sales and over $115,000 in revenue. This case study unpacks how that happened, what the numbers honestly mean, and which parts of it you can actually take home.
What we actually made
Everything in the campaign came out of our AI-accelerated pipeline — the same one we use for every client at our AI video studio in Málaga. Cinematic craft sets the standard; AI removes the friction between an idea and a finished cut. Four decisions did most of the work.
Cinematic AI product shots
A strap is a small object with big emotional weight, so we treated it like a hero prop in a film. Macro-style passes over stitching and texture, light raking across the case, the strap snapping onto the watch in one satisfying motion. Our AI pipeline let us produce these cinematic product shots without booking a studio for a week — we could explore lighting setups, backgrounds and moods in hours, keep what looked expensive and discard what did not.
Hooks-first editing
Every cut was built backwards from its first second. On Reels, the hook is not an intro — it is the entire pitch, because most viewers decide to stay or swipe before the second beat lands. So each variant opened on the most arresting frame we had: the transformation of the watch, a detail you had to lean into, a claim the community would want to argue with. Logo, context and explanation all waited their turn.
Vertical-native 9:16
These were not landscape films cropped for phones. Every shot was composed for 9:16 from the start — the strap filling the frame top to bottom, text placed where thumbs and captions would not cover it. As with every Pelisson Production delivery, the client also received a finished 4K master and the full set of platform cuts in 16:9, 9:16 and 1:1, but the vertical versions led the campaign because vertical is where Reels lives.
Enough variants to find a winner
This is the quiet advantage of AI video ads and the part most brands underestimate. Instead of betting the budget on one or two polished films, we produced a volume of distinct variants — different hooks, different pacing, different first frames — at a production cost per concept that traditional shooting cannot approach. Nobody, including us, knows in advance which video ad creative will connect. Volume is how you find out quickly instead of guessing expensively. It is the same volume-first approach that powers our AI UGC ads at scale service.
How did a $700 campaign reach 24 million views?
It didn't — not on paid distribution. The overwhelming majority of those 24 million views were organic. The $700 seeded the creative in front of an initial audience; the videos then earned the rest of their reach by behaving exactly the way Instagram's recommendation system rewards.
Reels distribution runs on a simple loop. A video is shown to a small test audience. If enough people watch past the opening seconds, rewatch it, share it or send it to a friend, the platform shows it to a larger circle, then a larger one again. Strong signals compound; weak signals kill a video within hours. Nothing about that loop asks how much you paid.
Our videos happened to feed that loop on every signal at once. Hooks-first editing held retention through the decisive first seconds. The cinematic product shots gave people a reason to watch to the end — the payoff of seeing the finished watch. And the subject did the social work for us: the AP x Swatch community is exactly the kind of audience that sends a Reel straight to the group chat, and shares and sends are among the strongest distribution signals a Reel can earn.
What the paid spend actually did
The $700 had one job: guarantee the first audience. Paid placement removed the cold-start problem, putting each variant in front of enough of the right viewers for the platform to gather real signals quickly. From there, every video lived or died on its own retention and share rate — and the strongest ones simply never stopped being passed along.
The $700 didn't buy 24 million views. It bought the first few thousand. The creative earned every view after that.
That is the honest mechanics of a viral video ad, and it is why we keep saying that creative quality is a distribution strategy. Media budget determines who sees your video first. Creative determines whether anyone sees it second.
The numbers
Here is the full picture from one week of the campaign, in one place.
| Metric | Result | Why it matters |
|---|---|---|
| Views | ~24 million | Overwhelmingly organic reach — the creative earned its own distribution |
| Sales | ~2,000 orders | Attention converted into transactions, not just watch time |
| Revenue | Over $115,000 | Real money through the store, not projected value |
| Paid ad spend | $700 | The seed budget — the only media cost in the campaign |
| Timeframe | One week | The compounding happened in days, not months |
| Platform | Instagram Reels only | One channel, with formats built natively for it |
Now the maths everyone asks about. Revenue divided by paid ad spend: $115,000 ÷ $700 comes to roughly 164. A triple-digit paid ROAS.
We are proud of that number, and we will also be the first to tell you it is misleading if you read it as an advertising result. The $700 did not generate $115,000 of demand; it lit a fuse. The revenue was driven by organic reach that the ad spend seeded but never paid for. Treating 164x as a media-buying benchmark would be like crediting the match for the size of the bonfire.
The right reading is this: when the creative is strong enough to earn its own distribution, the ratio between ad spend and revenue stops behaving like a normal paid campaign — and that is precisely the argument for investing in the creative.
So… can AI video ads deliver real ROAS?
In our experience, yes — this campaign put over $115,000 of revenue against $700 of paid spend, and every video in it came out of an AI-accelerated pipeline. But the interesting question is not whether AI video ads can perform. It is why AI changes the economics of performance in the first place.
Three mechanisms, all of them boringly practical. First, variant volume: AI collapses the production cost per concept, so the same budget produces many distinct creatives instead of one or two. More genuine attempts means more chances of finding the video that moves — and in paid social, finding the winner is most of the game.
Second, iteration speed: when a variant shows signs of life, we can cut new hooks, alternative openings and fresh angles on the winning idea in days rather than re-booking a shoot. The algorithm rewards momentum, and AI lets you feed it while the moment is still warm.
Third, cost-per-result: every euro not consumed by production overhead is a euro available for media or margin. When the creative costs less to make and performs better per impression, your cost-per-result falls at both ends of the equation at once.
None of this replaces craft — a hundred mediocre variants lose to three excellent ones. The AI does not make the taste decisions; it makes the taste decisions affordable to execute at volume. That combination — cinematic standards, AI speed — is what made this campaign possible on this budget.
Is a result like this repeatable?
No — not this exact result, and we would rather lose a sale than pretend otherwise. A 164x paid ROAS is an outlier. It required strong creative, a passionate niche community, a platform algorithm in the right mood and a measure of timing luck, all landing in the same week. Anyone who promises to reproduce those numbers on demand is describing a lottery ticket, not a service, and no honest case study will tell you different.
It is worth saying plainly because this industry rarely does: most campaigns, including good ones, do not go viral. We have run strong creative that earned a solid, unremarkable cost-per-result and nothing more. That is the normal outcome, and any studio showing you its best week owes you that context.
What is repeatable is the process that made the outlier possible — and the process is worth having even when nothing goes viral.
- Hooks-first creative, built backwards from the first second, lifts retention on every video — viral or not.
- Platform-native formats consistently outperform repurposed crops, because they respect how people actually watch.
- Producing enough variants turns finding a winner from a gamble into a search — a numbers problem you can afford to solve when AI keeps the cost per concept low.
- Fast iteration on early signals compounds whatever traction appears, instead of letting it cool while a re-shoot gets scheduled.
Run that process and the realistic, unglamorous outcome is a better cost-per-result than you are getting today: cheaper attention, stronger retention, more shots at outperformance for the same budget. Sometimes that shows up as a modest, steady lift in video ad ROAS. Occasionally the algorithm hands you a week like this one. You cannot schedule the second outcome — but you only receive it if the first is already in place.
What this means for your ad budget
The uncomfortable lesson of this case study is that most e-commerce video ads have the budget upside down: heavy media spend pushing thin creative. Paying for distribution is how you compensate for a video that cannot earn any. This campaign argues for inverting that.
Creative-first budgeting looks like this in practice:
- Put the first money into video ad creative — enough variants to genuinely test different hooks and angles, not one film you are emotionally committed to.
- Seed with modest spend. A few hundred euros is enough to put the creative in front of the platform and read the signals: retention, shares, saves, cost-per-result.
- Hold the main media budget back until a winner declares itself. Scaling a proven video is investment; scaling an unproven one is hope.
- Iterate on the winner immediately — new hooks, new openings, same core idea — while the algorithm is still paying attention.
- Deliver in every platform format from day one. Our masters ship in 4K with 16:9, 9:16 and 1:1 cuts, so a winning idea can move from Reels to feed to YouTube without a re-edit standing in the way.
Notice what this does to risk. In a media-first budget, a weak video quietly burns the whole spend. In a creative-first budget, weak variants cost you a test, and the budget concentrates itself on whatever proves it can perform. You spend less to learn more — and when something takes off, you are funding a fire that is already burning.
Our watch-strap client did not out-spend anyone that week. On $700, they could not have. They out-crafted the feed, and the distribution followed.
For the full pricing picture, see how AI video costs compare to traditional production in our FAQ.
The short version
- One week on Instagram Reels: roughly 24 million views, around 2,000 sales and over $115,000 in revenue — on $700 of paid ad spend.
- The triple-digit paid ROAS is real maths but an outlier: the reach was overwhelmingly organic. The creative earned the distribution; the $700 only seeded it.
- The repeatable part is the process — hooks-first creative, vertical-native 9:16, enough variants to find a winner, fast iteration once one moves.
- AI video ads change the economics: more concepts per budget, faster cycles, lower cost-per-result — which is exactly what makes that process affordable.
- Budget creative-first: prove the video with small spend, then scale media behind the winner, never in front of it.
- No one can promise virality. A strong process just makes sure you are ready when the algorithm says yes.