Revisiting 10 Classic AI Marketing Case Studies: From Guessing What You Like to Getting You Hands-On

A few days ago, a friend showed me his Spotify.

He typed one sentence into the AI playlist feature: "songs to listen to while riding off into the sunset."

A few seconds later, a playlist appeared. He put on his headphones, grinning like he'd just won something: "See? It gets me."

Staring at that playlist, I found myself thinking about a question that comes up all the time: how much of marketing can AI actually do?

With that question in mind, I dug out a list of case studies from around 2024, went through it again, and picked out 10 classics.

The first movers were the big names — Pepsi, Unilever, Salesforce. By around 2024, the wave had swept across every industry: Google teamed up with a group of AI marketing providers to build personalized recommendations for Tripadvisor's 400 million monthly active users; the Boston Red Sox pushed Salesforce to new heights, using data to reach fans one-on-one and turning a CRM system, effectively, into a Bill James Handbook (the data-driven sabermetrics annual that changed how baseball teams think); Meta simply opened up a generative entry point for advertisers — type in a few prompts, and images and copy come out.

After going through these 10 cases, I realized they actually line up into a clear progression.

From guessing what you like, to getting you hands-on, to running the numbers for you — and finally, to buying the team that builds the AI outright.

Each step goes deeper than the last.

What do I mean by a progression? Let me walk through them one by one.

Step 1: Guess What You Like

You could go your whole life without ever seeing the same Netflix homepage twice.

Based on your viewing habits, the content of its shows, and thousands upon thousands of user-preference profiles, it re-ranks the entire homepage for you alone.

What for? Getting you to watch one more episode, and then one more. That's how retention and stickiness get built — one episode at a time.

Spotify is the same playbook for music. Personalized playlists are just table stakes; paying users can type in a prompt and have the AI assemble a playlist on the spot. That opening "riding off into the sunset"? That's where it came from.

Amazon turned this into a product. The recommendation system it built for its own e-commerce operation, called Amazon Personalize, was later opened up so that other catalog-based businesses could use it as a web service. Why recommend this item to you? It looks at your past behavior, then at your real-time behavior; put the two together, and conversion rates and satisfaction climb.

Then there's Stitch Fix, a clothing-subscription retailer that takes an even more granular approach. It uses OpenAI's machine learning to write product descriptions and built an outfit-combination model to recommend new looks to users. Its human stylists also use GPT-4 to remember each customer's preferences — the feedback you gave last time shows up in this shipment.

Everything AI does at this step boils down to one sentence: it guesses on your behalf — what you'll want next.

The more accurate the guess, the harder it is to leave.

Step 2: Get You Hands-On

Guessing is just the price of admission. At step two, AI starts inviting customers onto the field.

The boldest thinker here is Coca-Cola.

It built a GPT-4 platform and ran a series of campaigns called "Real Magic," opening up its brand assets outright: the polar bears, Santa Claus — those iconic images, yours to use. Users could create their own ads, turn them into billboards or Christmas cards, and play however they liked.

When I first came across this case, my reaction was: good grief, a brand actually dared to hand its logo over to strangers.

But think again — this may be the smartest advertising there is. The best ad is the one a customer makes with their own hands. Put effort into something and you can't bear to part with it; and if you can't bear to part with it, you'll remember it.

At Sephora, the thing you get your hands on is your own face. Its app has a Virtual Artist feature that uses facial recognition to try makeup on your selfie directly. That row of testers at the beauty counter has been moved, intact, into your phone.

L'Oréal built the whole thing into a system. Its skin and hair diagnostics emphasize inclusivity, and it ships a generative personal beauty assistant; more striking still, content for all 37 of its beauty brands is localized and checked for brand compliance by AI — one distinct tone per brand. It also teamed up with Meta on a creator collaboration called "New Codes of Beauty," bringing 3D, AR, and AI together.

The essence of this step: customers go from audience to participants.

Step 3: Run the Numbers for You

In the first two steps, AI was still touching content and experience. At step three, it starts touching money.

Heineken — a beer brand, yet its pain point is remarkably typical: data is just too hard to manage. It was early to the game, using AI at scale to unlock that data, and the questions it answered were all worth real money: when should the ads run? Where? For offline promotion, which bars should it walk into?

Behind every question sits a budget.

BMW's move is more concrete. It has an internal tool called EKHO (Enterprise Knowledge Harmonizer and Organizer) that uses AI to generate realistic car images and videos. A customer in the showroom who wants to see a particular custom configuration no longer has to leaf through a shelf of manuals — they just look at the picture. And when the marketing team needs to analyze several rounds of campaign assets, it's just as fast.

Faster math makes for faster decisions — and every fast decision saves real money.

Step 4: Buy the AI Team Home

The first three steps are, at bottom, all about "using." The most ruthless play is "owning."

Airbnb went ahead and acquired a company outright: GamePlanner AI. Once the deal closed, co-founder Adam Cheyer stepped in as Airbnb's Vice President of AI Experience.

Who is Adam Cheyer? A co-founder of Siri who, back in the day, also worked on the iPhone's development.

Why buy this team? Generative AI features for the short-term-rental business: visual tours, pricing assistance, machine-learning recommendations.

There's a detail here I particularly love. Hosts with ideas were already taking the initiative, optimizing their listings inside every kind of chatbot. When the platform reacts a little slowly, it's the users who leave it behind.

Anyone can buy tools; bringing the team that builds the AI under your own roof is the moat.

Back to That Playlist

Having been through these 10 cases, I look at my friend's "riding off into the sunset" playlist and feel something completely different.

That playlist is a miniature of the road AI marketing has traveled: first it guesses you, then it lets you play, then it runs the numbers for you — and in the end it simply takes up residence in your life.

The UX expert Jakob Nielsen once imagined something even further out: websites of the future could reshape themselves in real time, automatically rendering accessible versions for blind and low-vision users. Follow that line of thought, and AI could even let each customer choose their own plot, turning customized content into a personal adventure.

More than two years on, these cases have become classics — and starting points.

Next time an AI-generated playlist, ad, or homepage appears in front of you, you can stop and ask: which step has it reached?

There's delight in being read correctly by AI; a guess that's a little too accurate can give you goosebumps. My friend, for his part, is still showing that playlist off to anyone who'll look.

Here's wishing you one more thought than the AI — always.