20 Case Studies Later: AI Marketing — Who's Actually Doing It, and Who's Just Pretending

A while back, a friend who teaches marketing at a university asked me out for tea. He had worry written all over his face.

I asked him what was wrong.

He said his classes were getting hard to teach. He'd be up there lecturing about positioning and communications, and a student would raise a hand: Professor, AI can already write copy, produce ads, and cut video. Does any of what you're teaching still hold up?

He didn't have an answer.

I told him: not having an answer is exactly right. A question like that can't be answered with theory. You have to show students cases. Real companies, real budgets, real results.

So the two of us dug through every generative AI marketing case we could find from the past couple of years — big companies and small, the hits and the ones that got roasted. Twenty in all.

First, let's get one thing straight. What counts as generative AI marketing?

Not rule-based automated ad placement, and not reporting software. It's tools that genuinely generate content for you: writing copy, drawing images, producing video, assembling an entire campaign. Generated, not automated.

I'll walk you through these 20 cases in three groups. The giants, the small shops, and the shovel sellers.

1. The Giants: Efficiency First, Nerve Second

Microsoft first.

Microsoft once made an ad for the Surface. Script, storyboards, backgrounds — all of it done with generative AI. The result: 90% saved on both time and cost.

What does 90% mean? The cycle that used to produce one ad can now produce ten.

And here's the more interesting part: the audience never noticed.

Think about that. The fact that viewers couldn't tell is itself worth chewing on. It means the quality of AI generation has cleared the pass mark. It also means that "couldn't tell" is bound to become an ethics question sooner or later. More on that later.

Now, Nutella.

Nutella used an algorithm to design labels for its jam jars. From the same brand foundation, it generated 7 million one-of-a-kind patterns. Every single jar, different.

The result: sold out in a month.

Seven million. Hire designers to draw those — you'd be drawing until the end of time. This is the first principle of generative AI: the marginal cost of content approaches zero.

Only at zero cost can you do what you never dared to before. Like personalizing down to every single jar.

Then there's Heinz. When I finished this one, all I could say was: this is scary good.

Heinz ran an experiment with DALL·E 2: it asked the AI to draw "ketchup." No matter how the prompt was written, the bottle that came out looked like Heinz.

The AI has seen millions of ketchup images, and the default bottle in its head is Heinz.

That's more ruthless than any consumer research. A brand taken to its extreme becomes the mental default for the category itself. Heinz turned the finding into an ad, and it promptly took over everyone's feeds.

Next, Nike.

For Serena Williams's retirement, Nike made an AI video: the Serena of 1999 versus the Serena of 2017, facing off across the net. AI analyzed her game from both eras and simulated a match of "self vs. self."

Is the idea good? Yes. But the controversy is just as big: if you use AI to recreate a real person, how do you account for likeness rights? She consented — but what about other athletes? What about those who didn't?

Coca-Cola's story has to be told twice.

The first time: a holiday campaign. They used AI to generate multiple versions targeted at different American cities — thousands of assets, rolled out fast. No arguing with the efficiency. Then they got roasted. The criticism zeroed in on the visual quality and the "uncanny" rendering.

They got the efficiency, and the reputation face-planted. That's lesson one of AI marketing: fast does not mean good.

The second time is called Create Real Magic. Coca-Cola built a platform that combined GPT-4 with DALL·E 2, opened up its brand assets, and let consumers and artists create their own Coca-Cola-themed works. The best pieces went up on the Times Square billboards.

This time the wind turned. Same AI-generated content — when people make it themselves, everyone applauds; when the brand shoves it in their face, everyone piles on.

Funny, when you think about it.

2. The Small Shops: Counting Every Penny

That's the giants. You think AI marketing is a rich man's game?

Wrong. Small shops may be the biggest winners.

In Indiana there's an online sticker shop called Otto's Grotto. The owner, Therese Waechter. One person.

She uses Jasper and ChatGPT to write product descriptions and generate labels — even her Shopify storefront was built with AI's help. Shopify runs on a language called Liquid (the platform's template language), so she had the AI generate it section by section and revise it section by section. She gave this a name: vibe coding. No code written — just following her instincts in conversation with AI until the store stood up.

In 2024, her revenue more than doubled. Without hiring a single programmer.

A sticker seller and a trillion-dollar company, using the same set of tools. The tools have been equalized. Now the gap lives in the head.

Then there's Amarra, a formal-dress wholesaler in New Jersey. They use ChatGPT to write product descriptions and cut content production time by 60%. They added an AI inventory system that forecasts stocking from historical data and seasonal trends, and overstock dropped 40%. On the customer-service side, a chatbot now handles 70% of inquiries.

Three numbers — and every one of them is profit earned by saving.

In Kent, England, there's an e-commerce company, Must Have Ideas, that pulled off something even bolder. They wanted into TV shopping, but traditional TV shopping comes with production costs steep enough to scare anyone off.

So they built their own AI system, called Spark. Pre-recorded footage gets spliced, arranged, and scheduled by AI — and on the Sky platform they spun up a TV shopping channel running 24 hours a day without interruption.

No hosts working three shifts around the clock, no live team running nonstop. Sales surged, and they built up a fiercely loyal base of repeat customers.

Two more.

Headway, a Ukrainian edtech company, uses Midjourney and HeyGen to produce video ads in a "shot-by-real-users" style — in the feed they look like genuine human UGC (user-generated content). Video ad ROI rose 40%, and in the first half of 2024 they racked up 3.3 billion impressions.

Misela, a handbag brand, composites real product photos into AI-generated scenes from around the world. The model stands on a street corner in Istanbul, in an alley in Tokyo — without ever stepping out the door. Travel costs saved, production cycles shortened, and the output looks like a shoot that circled the globe.

3. The Shovel Sellers: Who Gets Paid in Every Gold Rush

When a gold rush starts, the shovel sellers get rich first.

Omneky in San Francisco helps brands generate ad creative at scale, then feeds placement data back in — whichever version performs gets amplified immediately. Anyword in New York specializes in generating and optimizing marketing copy — web, social, email, ads, the whole set. Neuroflash in Hamburg goes a bit more mystical: it kneads neuropsychological insight into content generation, claiming its copy can scratch an itch users never knew they had. Colossyan makes AI-avatar training videos — one script, many languages, a finished film straight out the other end.

RTB House in Poland uses deep learning to predict user interests for personalized placement, lifting ad effectiveness by 41% to 50%.

And there are two new species worth a longer look.

One is OneOff, an AI fashion search engine: a user says they want "that Hailey Bieber look," and it uses large AI models to dig out the identical pieces and the dupes. The creator's path from wearing a look to monetizing it just got straightened out.

The other is Peec AI, a Berlin company. Their bet: the search entry point itself is being taken over by AI. Soon, when users ask an AI assistant "what should I buy," whether your brand shows up becomes an existential question. They built a SaaS that monitors your visibility and ranking across the AI search engines. The space is very early — in their early days they'd raised only a little over $200,000. But the direction may matter more than the money.

When the search box becomes a chat box, SEO has to be rewritten from scratch.

4. Three Questions for the Classroom

Finally, back to Toys 'R' Us.

The legacy toy retailer wanted to come back from the dead, and had someone make a 66-second brand film telling the story of its founder's early days — entirely AI-generated, history blended with fantasy.

When the film dropped, opinions split down the middle. Some saw new possibilities for brand storytelling; others said that if you use AI to tell a brand story about childhood and warmth, that layer of emotional skin is fake.

I won't hand down a verdict. But there are three questions I don't think anyone who studies or practices marketing can dodge:

First: when AI-generated content is indistinguishable from human-made, how much value is left in "authenticity"? And who gets credited with the consumer's trust?

Second: who collects the gains from efficiency? Microsoft saved 90% of its costs — does that become lower prices, profit, or yet more ads bombarding you?

Third: at what point does personalization cross the line? Seven million labels are a delight; push notifications that understand you a little more each time are a nuisance.

There are no standard answers. But the answers will decide how value gets divided up across the marketing industry over the next decade.

When I finished walking him through all 20 cases, my professor friend was quiet for a moment. Then he said: sounds like my course needs to be torn up and rebuilt.

I said, no need to tear it down. Just flip the order: it used to be theory first, examples after. From now on, it'll probably have to be 20 true stories first — then one question for the students:

If it were you, what would you do?

AI won't replace marketing. But marketers who use AI are already replacing the ones who don't.

That line is for the person at the podium — and for the students in the seats asking the questions.