In September we did something slightly embarrassing. We put 400 of our own ads on a wall of screens, turned the sound off and watched them the way people actually meet them: half paying attention, thumb already moving. The ads ran between January and August 2026 across eleven clients, and for each one we had the hold rate, meaning the share of people still watching at three seconds.
The spread was wider than any of us expected. The top quarter held 41% of viewers to the third second. The bottom quarter held 17%. Same brands, same budgets, often the same shoot day.
What the good ones had in common
We expected to find a trick. A colour, a cut speed, a face. What we found was simpler: every ad in the top quarter showed the product doing its job before the third second. Not the logo, not a person about to speak, not a title card. The serum going onto skin. The shoe hitting wet road. The can being opened.
The weak ads were not ugly. Many were the ones we were proudest of in the edit. They opened with mood: a slow push in, a sunrise, a line of text setting up a joke that paid off at second nine. On a cinema screen that is craft. In a feed it is a request for patience from someone who has not agreed to give you any.
Nobody scrolls past an ad because it started too quickly.
Sound is a bonus, not a plan
Roughly seven in ten of the views in our sample began muted (platform reporting, same period). So an opening that depends on a voice is an opening most people never get. We now write the first three seconds as a silent film: one action, one object, and no more than five words on screen. If a script needs the voiceover to make sense of the first shot, the first shot is wrong.
The text matters more than we like to admit. The best opening lines in the sample were plain claims a stranger could check with their eyes: “Dry in forty seconds.” The worst were questions. A question asks the viewer to do work, and they came here to do none.
How this changed one ad
The Lumo Run feed ad shown here first opened on a runner tying her laces in a hallway. It was a lovely shot. It held 19% to three seconds. We recut it to open on the shoe landing in a puddle, with the headline already on screen, and moved the hallway to second four. Nothing else changed: same footage, same music, same audience. The hold rate went to 38%, and over the following six weeks that version brought in new customers at a cost 27% lower than the original (Lumo Run store data, May to June 2026).
What we do differently now
Three rules came out of the wall of screens, and they are now printed above the edit desks.
First, the product works before second three. Second, the opening must make sense with the sound off. Third, every ad is cut with two different openings and both are tested, because we are still wrong about which one will win roughly half the time.
That last rule is the humbling one. Fifteen years of making ads and I cannot reliably pick the better opening from two. What I can do is make sure there are always two to pick from, and that the feed, not the creative director, makes the call.
The other twenty-seven seconds still matter. They are where the price, the proof and the reason to act live. But they are only ever seen by the people the first three let through.




