Customer research, three challenger hooks and a fourteen-day head-to-head, beating a control that had gone unbeaten since November at the first attempt.
at £13.61 from the winning challenger, against the control's 47 at £14.48
the control had gone unbeaten before the test
in AI credits to make all three hooks, instead of the price of a shoot
In July 2026 I worked with Ramzy, founder of Ivy, a food-scanning app that scores packaged food out of 100, in an attempt to beat their best-performing Meta ad.
Their current creative opened with someone scanning a well-known brand of bread in the supermarket and being surprised at the low score it achieved. Running since November, the ad had beaten everything they'd put up against it.
But as spend scaled, the cost per result climbed too, up by more than a half since they launched it. The ad itself hadn't tired. People were still watching it as long as they always had. It was just running out of new people to win over.
During our discovery call, Ramzy admitted that the winning creative didn't have much research behind it. The ad just worked better than everything else they had tried, so I began researching their niche, their ideal customer and the science behind ultra-processed food.
The research showed that when people talked about using the app, the overriding emotion wasn't surprise at the results. It was much stronger than that.
People were angry that they'd been regularly eating food that wasn't as healthy as it claimed to be.
Mums who thought they'd packed healthy lunchboxes found out the food was ultra-processed and full of hidden sugars. What came through was wounded pride, because feeding their kids well is part of how they see themselves. And there were people standing in the aisle, surrounded by similar-looking food, not knowing what was healthy and what was not.
Those three feelings, and those three scenarios, became the hooks I built my ads around.
Rather than spend time and money shooting ads on untested scenarios, I created short AI-generated hooks that I edited into the existing ad's supermarket footage to keep the test fair. These three new ads ran against the existing ad in a head-to-head challenge, on a no-win, no-fee agreement.
Each one opens on a different feeling, cut into the control's existing footage, so the angle was the only thing that changed.
50 trials at £13.61. Beat the control.
The winnerHighest click-through in the test (5.2%), worst at converting. Switched off on day 6.
Didn't grab enough attention. Switched off on day 6.
I normally run a test like this at a budget of £25 a day for a week, which is usually plenty when a result costs a few pounds. Once I had access to Ivy's Meta account, I realised that at that budget the number of conversions would be too low for Meta to work out who best to target, so the outcome wouldn't be reliable. Instead, I recommended extending the test window to 14 days at £50 per ad per day.
By day 6, two of the three ads were not performing anywhere near the account's benchmarks. The overwhelm ad didn't grab enough attention, and its cost per trial was too high to justify. The anger ad had the opposite problem: it earned the highest click-through rate in the whole test but failed to turn those clicks into trial sign-ups.
I recommended Ramzy kill both. Every extra day either one ran was more of his budget going into a test that had already told us all it was going to.
That left one challenger against the control going into week two. The control had been in front every day since the test began, until my ad took the lead. The next day the control took it straight back. But the trend was turning my way. My ad was holding its own on hook rate, and its cost per trial fell a little further every day as those early, expensive results were diluted. The control's held steady.
50 trials at £13.61 from the winning challenger, against the control's 47 at £14.48. Three trials and 87p is close, close enough to sit inside the margin of error. But it won, first time out, against an ad nothing had beaten in eight months.
"I was amazed that just changing the hook could improve the results."
Ramzy Nanou, Founder of Ivy
Digging into the data a little further showed the cost of splicing a new hook onto old footage. Even though my ad got more attention and won the challenge, its conversion stats were lower than the control's.
My hook opened on wounded pride and then cut straight to footage built for surprise, so by the time the ad asked for the sign-up the feeling had gone. That points to the upside: an ad built on wounded pride the whole way through, rather than for the first few seconds, has room to move that the control doesn't.
Those three hooks cost £2.37 in AI credits to make, instead of the price of a shoot. That's the heart of how I work. Test cheaply, then scale sensibly.
A winner you can't explain is a winner you can't repeat.
Ivy had been trying to beat that ad since November and couldn't, because they couldn't explain what made it work in the first place. My approach is to do the research, produce a hypothesis and then test it cheaply. I can't promise a winner, and I'd avoid anyone who says they can. But the ads that fail cost very little, and the ones that work tell you why, so you're not starting from scratch each time.
Then back to Find, with each winner feeding the next round of angles.
Send me your current winner. I'll build new ads to go head to head with it. Or you don't pay.
Put your best ad to the test