Every September, half the brands in the country run the same ad. Back to school.
Oddbox, the fruit and veg box company, did too. It didn't work. The ads underperformed and nobody could quite say why.
Gastón Tourn, Oddbox's Chief Growth Officer, told the story at aiblLIVE in London on 20 October. He'd been under pressure from his board to use AI to make ads faster and cheaper. Instead, he used it for something more useful.
"Don't use AI to make the ad," his slide said. "Use it to understand the ad."
The question marketers never answer
Gastón made a confession most marketers will recognise. For around fifteen years, he said, "we have really struggled to understand why certain ads perform better."
We know which ad won. We can see the clicks and the costs. What we've rarely known is why. Was it the hook, the picture, the time of year or the offer? Usually we guess, then make another ad based on the guess.
Oddbox fed its ads into creative analysis tools, MotionAI and Meta's own, that can read hundreds of ads, the comments on them and what competitors are running. The tools spotted something the team hadn't. A lot of other brands weren't saying "back to school" at all. They were saying "back to routine".
Separately, the data showed that most of the people actually buying from those ads weren't parents.
The human bit
Here's the part I like best. The AI laid out the pieces. It took a person to put them together.
A member of Gastón's team called Charlie looked at those pieces side by side and connected them. If most buyers aren't parents, why are we talking about school? He suggested shifting the message to "back to routine": September as a fresh start for anyone, kids or no kids.
Oddbox says the change cut its customer acquisition cost by 17% and lifted conversion by 21%. That's their own test, so it's their claim, but it's a big swing from a few words.
Gastón summed it up neatly: "AI will always find the pattern, but gut feel will find the idea."
Why this is good news for small businesses
The usual assumption is that AI's job in marketing is to churn out more content. More ads, more posts, more versions. That's the obvious use, and everyone else is already doing it.
This is a quieter, cleverer one. Use AI as the analyst you could never afford. Let it read your last fifty posts, your last twenty ads, the comments, the reviews, your competitors' pages. Ask it what's different about the ones that worked. And use your own judgement, the thing you actually know about your customers, to decide what to do about it.
You don't need special tools to start. Copy your last ten posts into ChatGPT or Claude along with how each one did, and ask it to spot the patterns and what it would test next. Treat the answer as a lead and see what your gut says.
What would it mean for your marketing if you knew why your best post worked?
You'd stop guessing. Every new ad, email or post would build on something real. And your budget would go further, because you'd be doing more of what works and less of what only looked good in the planning meeting.
Maybe it's worth trying with whatever you've posted lately. You might find your own "back to routine" hiding in there.
