From a product page to a video ad you would actually run
AI video ads for ecommerce: what to hand over, what to ask for, and the two places most attempts fall apart.
What do you actually need to start?
Less than people expect, and different from what they usually bring.
Three or four honest photographs of the product beat one polished hero render, because the render has already thrown away the detail — the seam, the texture, the way light catches the cap. Add whatever else you have: the guidelines, a previous ad that worked, a voice clip if the brand has a voice.
What does 'write the outcome' mean?
Say what the ad is for and who it is talking to, not which camera move you want.
'Fifteen seconds, vertical, cold traffic on paid social, opens on the problem, product by second four, one line of on-screen text' is a brief. 'Cinematic dolly shot, 8k, hyperrealistic' is a wish. The first can be judged against something. The second can only be admired or not.
Why does the format belong in the brief and not at the end?
Because a vertical ad and a square ad are different ads, not the same ad cropped.
If the framing was decided for sixteen-by-nine, the square version loses the product or the text, and no amount of careful cropping puts it back. Decide the shapes you need before the first frame exists.
Where does it usually fall apart?
Two places, and they are both upstream of anything you can see.
The product changes between shots, so the ad quietly stops being about your product. And the crops are made last, so the placement you cut for loses the thing the ad was selling. Both are fixable and both are decided before generation, not after.
What about the first three seconds?
They are most of the ad, and they are the cheapest thing to test.
Brief them explicitly. What is on screen at second zero, what the viewer understands by second two, when the product arrives. Vagueness here is where most of the performance goes, and it is the part people leave to chance because it feels like a small detail.
Does it need sound?
Most social video is watched muted first and unmuted second, so it has to work both ways.
That means the on-screen text carries the argument and the sound carries the feeling. An ad that only works with sound is an ad most people never saw.
What does a good run give you back?
The ad, plus the versions you will genuinely need: the vertical, the square, the cutdown, and stills pulled from it for static placements.
Asked for together from one brief they are the same ad in different shapes. Asked for separately, later, they are four ads that resemble each other — which is what most accounts are actually running without realising it.
How do you get it approved without a meeting?
Send the finished cut on one link. Whoever is reviewing pins a comment to the exact frame they mean, approves it or asks for a change.
Then you steer the production from what they said rather than starting again. The argument happens on the frame instead of in a thread, which is the difference between fifteen minutes and two days.
What should you test first?
The hook, always, and one variable at a time.
Changing the opening and the product shot and the voice at once tells you which of three ads won and nothing about why. The point of volume is learning, and you only learn from variations you can attribute.
What does the product page give you and what does it not?
It gives you the claims, the audience and usually the objections — most product pages are the most-argued-over copy a company owns.
It does not give you usable footage. The imagery on a product page is composed for a page, often heavily retouched, and frequently the wrong aspect for anything else. Take the words from the page and the pictures from the library.
What should you brief for each placement?
The shape, the length, and whether sound is available. Those three change the ad more than anything aesthetic.
A vertical fifteen with no sound is a different piece of work from a square thirty with sound, and briefing them as one job with crops is how the second one ends up mute and confusing.
How do you know it is working?
Watch-through in the first three seconds before anything else. Everything downstream — clicks, cost, conversion — is gated on people still being there.
If the hook holds and the numbers still disappoint, the problem is the offer or the audience, and no amount of new creative fixes either. That is a useful thing to be able to rule out quickly, and volume is what lets you rule it out.