I dug through dozens of AI marketing cases. Only three uses are worth anything

Let me start with a story.

Nutella. That jar of chocolate spread on the supermarket shelf. One day it did this: used an algorithm to generate labels for 7 million jars, one by one. Every label different.

7 million jars, 7 million designs. The jar in your hand is the only one of its kind in the world.

The result? Sold out in a month. Fans even rushed to social media to show off their jars — more than 10,000 posts. Nobody paid them. They posted on their own.

Think about that. This is the thing every marketer dreams of: customers advertising for you.

But here's what I actually want to talk about: the algorithm Nutella used is no black magic. Open any AI image tool today and you can do it too.

So why was it Nutella — not you — who made "7 million one-of-a-kind jars" happen?

A while back, I went through every AI marketing case I could find. Fast food, cars, cosmetics, flowers, gloves... lay dozens of cases side by side, and one pattern stings:

Same tools. Different leagues.

For most companies, AI comes down to one word: save. Write copy a little faster, get images a little cheaper. Does it help? Yes. But an advantage built on saving gets erased just as fast by a competitor saving the same way.

Only three uses truly separate winners. Let's take them one at a time.

The first: selling "one of a kind" at wholesale prices

What is personalization?

Everyone can say the word. In practice, you get as far as "Dear Mr. Zhang" at the top of an email. One step further and the costs run away from you.

What did AI do? It pushed the cost of treating every person individually to nearly zero.

Netflix's recommendation system drives 80% of all viewing on the platform. When you open it tonight, every poster you see, the order it comes in — all computed around you.

Spotify goes further. Personalized playlists account for 35% of users' listening time. On the ad side, AI picks the content and matches the people for the same ad slot — and ad recall jumped 2.7x. What does that mean? Same slot, right message, more than twice as many people remember you.

Starbucks' Deep Brew engine watches the rewards app every day: what you usually order, when, what today's weather is like — then decides which drink to push and which coupon to send. It also computes next week's stocking, so even waste goes down.

Burger King flipped the logic. It ran a "Million Dollar Whopper" contest: users built the burger of their dreams, and AI generated a custom poster for every entry — plus a jingle of its own. Users happily passed their "creations" along. The buzz built itself.

See what these companies have in common?

They aren't using AI to blast ads. They're using AI to remember every single person.

The second: swapping "one bet" for "ten thousand tests"

The second use is almost embarrassingly plain: creative used to be a gamble. Now it's a test.

How did ads get decided before? The boss went with gut feel, pitches were won by whoever argued loudest, one "feels pretty good" plan got the money, and then you left it to fate. Like emptying your savings on a single lottery ticket that draws once.

Now? Let me run the numbers.

A global snack company used an AI system called Albert to test 11,340 ad variants automatically. Creative, copy, audiences, bids — every combination tried.

11,340 of them. Test by hand? One optimizer reviewing ten concepts a day would need three years.

And Albert kept moving money to whichever combinations performed while it ran. The final result: cost per lead went from $10.75 to under $3.

Before: buy one ticket, one draw, live with it. Now: grab ten thousand at once, watch the draws come in live, and pour money onto whichever wins.

A flower seller, Euroflorist, used Evolv AI to test thousands of layout-and-copy combinations on its website in real time. Conversion up 4.3%.

You might curl your lip: 4.3%? That's it?

For a flower seller, 4.3% might be the year's entire profit. And it was nearly free: no added budget, no new product — just handing the page over to a machine to run the trial and error.

Luxury e-commerce player Farfetch went even finer. Its emails can't afford to have a single word off, so it used Jacquard to let AI test subject lines, tone, and phrasing — the brand voice guarded by humans. Result: open rate up 31%, click-through rate up 38%.

Notice something? AI never made a decision for anyone. It crashed the price of "making decisions."

The ideas are still human. But the cost of verifying them has collapsed.

The game has changed. A good idea used to be precious because trial and error was expensive, so everyone could only bet on a few ideas. Now that trial and error is nearly free, guarding one idea and gambling to the end is a fight against probability.

The third: turning "impossible" into a story

The third is the one I admire most.

First, a question: Serena Williams in 1999, or Serena Williams in 2017 — who's stronger?

That question never had an answer, because it could never be tested.

Nike did something about it. It used AI to analyze Serena's footage from two eras — running, swings, explosiveness, tactics — modeled how both "hers" play, then put the 1999 Serena and the 2017 Serena on screen for a match 18 years apart.

The ball really got hit. The audience really got it. The film's title hits hard too: Never Done Evolving. Evolution, never done.

Think about that. The idea, people could always come up with — the filming, they couldn't. AI killed the problem of "can't be filmed."

Here's an even more moving one. The nonprofit Malaria No More wanted to rally the world to end malaria and brought in Beckham. But he only speaks English, and malaria's hardest-hit regions are in Africa and Southeast Asia — the languages simply don't line up.

So, using Synthesia's video synthesis, the voices of several malaria survivors were "lent" to him — and Beckham "spoke" nine languages.

The video earned 700 million impressions, drove 420,000 online searches, and helped unlock $4 billion in funding commitments — the Gates Foundation among the funders.

One man who speaks only English, speaking for survivors in nine languages. Ten years ago, that was science fiction.

Heinz got in on the fun too, letting AI "paint ketchup." The painted images ran everywhere, online and off — 850 million impressions. A sauce maker, its machine-drawn pictures lighting up screens around the world.

What do these three share?

AI didn't take over creativity. What it took over was "execution."

The ideas are all human. It's just that before, these ideas would die in the meeting room for being "unfilmable." Now, they're alive.

Now, a bucket of cold water

By now you might be fired up. Hold on.

Across these dozens of cases, the most common use is actually the other kind: saving money.

Unigloves, a glove maker, used Midjourney and Adobe Firefly to generate 250 product images covering the use cases of 5 occupations — not a single live shoot — cutting design time by 57%.

SimCorp used Synthesia to raise explainer-video output 5x — 300-plus videos without entering a studio once.

British Council localized 1,000+ ads into 7 languages. Used to take weeks; now days.

A training company called Ultima.school simply replaced its entire presales team with a ChatGPT bot: reception, qualifying leads, booking meetings — fully automated. Customer acquisition cost cut in half, conversion didn't drop a bit.

Mattress company Tomorrow Sleep used MarketMuse to fill content gaps; in one year, monthly organic traffic went from 4,000 to 400,000. 100x.

Does it work? It works all too well.

But you must get one thing straight: all of these are efficiency. Not one is differentiation.

You can cut design costs 57% — so can your competitor, tomorrow. You ship videos 5x — your competitor installs the same tool next month. Efficiency's dividend has a cruel property: it always gets competed away.

H&M built AI "digital twins" of 30 models, letting the twins carry the commercial shoots that used to mean retouching until midnight. Notice one detail: the likeness and image rights were agreed in black and white. What's saved is money; what's held is the bottom line.

So the money-saving use is the 2026 passing grade for marketing. The passing grade's job is keeping you at the table. Want to win? That takes the three above.

Back to those 7 million jars

Remember the Nutella story at the start?

Now you can read it properly: what made the 7 million jars different was the algorithm; what made those 7 million jars get snapped up was the insight. Every child, at the moment of lifting that jar, hopes the one in their hands is the only one of its kind in the world.

The algorithm is the tool you pick up afterward — once you've understood that.

Dozens of cases later, my biggest takeaway is one sentence:

AI will work for you. It won't judge for you. The more democratic the tools, the more judgment is worth.

Tools — soon everyone will have them. Insight won't.

Here's hoping you put AI to work the third way — one day sooner.