156 Brands Actually Using AI in Marketing: I Read the Whole Database, and Three Things Surprised Me

A while back, I came across something interesting online.

Someone had organized cases of 156 brands worldwide using AI in marketing into a searchable database. Each card has three lines: what tool they used, what result they got, and where the original source is. Updated through July 2026.

I figured it would be yet another "AI will disrupt marketing" lip-service roundup. I clicked through a few pages. It wasn't.

It doesn't predict the future. It doesn't shout slogans. It just lays out the facts.

Coca-Cola used AI to remake that "Holidays Are Coming" spot — an interactive AI Santa avatar that, over 60 days, racked up more than 1 million interactions across 43 markets. Adidas used generative AI for personalized email, lifting sales 37% and cutting cost 91%. Sephora's virtual try-on has a purchase completion rate three times that of the regular flow.

Think about it: 156 names, each one backed by a string of real-money numbers.

Isn't this exactly the "who's actually using it, and how's it going" we've all been waiting to see?

I flipped through the whole database, and I want to tell you three things that probably aren't what you'd expect.

First: The Ad-Platform Gains Are Real — Just Don't Deify Them

Let's start with the easiest category to oversell. Ads.

Meta's Advantage+ and Google's Performance Max — these two systems are things you're basically already using the moment you spend money on ads. What they do, stripped down, is one thing: the bidding, audience, and placement combinations a media optimizer used to babysit, the algorithm now calculates for you.

And the results? The numbers in the database are remarkably consistent.

Meta Advantage+ lifts ROAS by an average of 22%. Google Performance Max lifts conversions by an average of 13%, with CPA roughly flat. Lululemon ran Performance Max in Canada, paired with a measurement approach built on attribution and incrementality testing, and lifted ROAS 8%.

8% to 22%.

Do these numbers matter? They do. In an industry where profit is counted in percentages, a ten-plus-point lift is real money.

But it's not the kind of revolution that makes you go "wow." What's it more like? It's like you used to employ a decent optimizer — and now that optimizer works 24 hours a day, never takes a break, and doesn't ask for overtime.

The numbers that actually make you sit up aren't on the ad side.

Second: Where Cost Gets Slashed by 90% Is Creative Production

This is what I most wanted to tell you about after reading through the whole database.

Adidas' personalized email creative: cost down 91%. Nestlé's internal tool, NesGPT: 150,000 employees using it, content-production time cut 60%. Unilever built an AI content factory where the output of a single campaign is 17x what it used to be.

91%. 60%. 17x.

These aren't numbers about "a modest improvement." They're numbers that upended the cost structure of the entire production line.

But notice — I'm not saying "AI replaced creativity."

Go look at that Too Faced mascara ad. It was made with Adobe's Firefly Video Model, five production stages, took about two weeks. The editing step alone went from four days down to one. Where did the saved money go? Into hiring more creators for collaborations, and shooting more real-model footage.

Humans set the brief; AI runs the production line.

That's what these cases are really saying. Designers, copywriters, strategists weren't eliminated. What got eliminated was the repetitive, assembly-style work — the kind where you take one banner and resize it into 40 versions.

Let me give you an even starker example. Kalshi, a prediction-market company, found a creator who used Google's Veo 3 to shoot a national ad, aired as a YouTube TV stream during Game 3 of the NBA Finals. Cost: about $2,000, done in two days. A traditional shoot of the same caliber? Seven figures, several months.

Dollar Shave Club went even further — they used Higgsfield and Claude to make an ad for $400. Their breakout 2012 hit cost $4,000.

When the cost of producing an ad drops from the hundred-thousands to the thousands, the whole industry's capacity logic changes.

This isn't a prediction. It's already happened.

Third: The One That Grabbed the Headlines Quietly Reversed Course

This section, I want to spend on Klarna alone.

In early 2024, Klarna deployed an OpenAI-powered customer-service chatbot, covering 23 markets and more than 35 languages. In its first month, it took on two-thirds of all customer-service conversations, and response time dropped from 15 minutes to under 2 minutes. The company publicly said this was equivalent to the work of 700 full-time customer-service agents, contributing $40 million in profit improvement.

Sounds like a textbook win for AI customer service, right?

That's exactly how the media reported it, too.

But by 2026, Klarna had shifted to a hybrid model and brought human agents back. The reason? Satisfaction on complex questions dropped.

I really want you to remember this reversal.

The customer-service category in the database has 20 cases. Bank of America's Erica has logged 2 billion cumulative interactions. Wendy's voice ordering hits 86% accuracy. Verizon equipped 28,000 customer-service reps with a Gemini assistant, and sales rose nearly 40%. These are all hard, real results.

But these winners share one thing in common: AI is the layer stacked on top of a human team — not a replacement for that team.

Verizon is the textbook case. It gave customer service an AI tool not to put agents out of a job, but to move them from "answering the phone" to "selling on the phone."

Allstate sends out about 50,000 claims-communication emails every day, and now almost all of them are drafted by AI. But every single one is reviewed by a claims adjuster before it goes out.

Lowe's equipped roughly 300,000 employees across more than 1,700 stores with an assistant built on OpenAI — so on the sales floor, staff holding a handheld device can ask about product details, inventory, or installation advice. It doesn't replace the employee. It makes the employee answer more accurately, more quickly.

Where's the lesson in Klarna's story? Treat AI as a replacement, and it'll drop the ball in some corner you're not watching. Treat it as an amplifier, and it amplifies what your team can do.

That one line is worth remembering more than any percentage.

The Ones Nobody Shouts About Are Going the Deepest

After those three categories, I have to bring up one category that's easy to overlook.

Internal enterprise AI deployment.

This category isn't public-facing, doesn't make the news, doesn't flood feeds. But it's the direction enterprise leaders ask about most in private.

JPMorgan Chase took 8 months to roll out a tool called LLM Suite to about 200,000 employees. Users save 3 to 6 hours per week, and nearly half use it daily. Microsoft used itself as a testing ground for Copilot, published the deployment playbook, and saved about $500 million in a year. IBM spread AI agents across more than 70 of its own business lines — over two years, $3.5 billion in productivity impact.

Are those numbers big?

Big.

But what's more interesting is how they did it. Amgen, from 2023 to 2025, rolled out generative AI to about 20,000 employees in several waves — leaving a governance and testing buffer between each wave. Vodafone first ran a three-month trial with 300 people, measured 3 hours saved per person per week and 4 hours saved per week in the legal department, found that 90% wanted to keep using it — and only then rolled it out to 68,000 people.

Not a single one went all in at once.

They're all doing the same thing: run a small-scope test, get numbers, then use those numbers to win over the next wave.

Isn't that what any organizational change is supposed to look like? AI just makes it run faster and measure more precisely.

Where Do You Start, If You're Going to Make a Move?

The database's compilers laid out three rules. I think they're pretty solid, so I'll pass them along:

First, only use named, real brands — no anonymous "a Fortune 500 retailer" filler.

Second, every number must be traceable to an original source. Announcements, earnings reports, press releases, vendor case studies. Posts from a vendor's own blog get flagged separately, so you know who's talking.

Third, there must be a concrete result. "Improved efficiency" doesn't count. Percentages, dollar amounts, hours saved, people impacted. That counts.

I'll follow that same logic and give you a judgment of my own.

Don't start in the sexiest place.

Generative video for a brand film is cool. But your first ROI you can take into a review meeting will almost certainly come from "boring" work: subject-line testing, ad-creative A/B tests, product-detail-page copy.

What these tasks have in common: they're quantifiable, reversible, don't require a new vendor, don't need to go through procurement. Mailchimp's AI subject-line tool lifts open rates 23% on average. Klaviyo's predictive send lifts revenue per recipient 30%. These are features already sitting inside tools you're already paying for.

Use these to run a number first, then go chew on the hard stuff — personalized recommendations, audience segmentation, in-app experience customization. By then, the conversation inside your team won't be "should we try AI?" — it'll be "what did this tool deliver for us last quarter?"

The level of that conversation is completely different.

One Last Thing

This database covers retail, food and beverage, finance, mobility, telecom, pharma, hospitality, media — 156 names.

Retail and e-commerce are moving fastest, because they already have product catalogs, recommendation engines, and user data — plugging AI in is a natural fit. Food and beverage are right behind, with Coca-Cola, Heinz, Cadbury, Nestlé, and PepsiCo all on the list. Financial services are pushing hardest on customer service and fraud detection. Mastercard, Stripe, JPMorgan — AI has been carrying the load in this space for over a decade.

If you want a benchmark to learn from, from a CMO's perspective I'd recommend these: JPMorgan (internal LLM + Persado for ad copy), Coca-Cola (generative AI in advertising), Adidas (personalized email), Nestlé (content production), Sephora (visual search and recommendations), Spotify (personalization made into the product itself), Walmart (employee training + supply chain).

Behind every name is a string of numbers you can trace back to the source.

Don't deify it, don't dismiss it. After reading through all 156 cases, I don't feel AI has "disrupted" anything in marketing. What I see is something more practical. The people who know how to use it have already started pulling away from the people who don't.

That gap didn't appear yesterday. But in 2026, it's finally grown large enough that everyone can see it.

May you be on the side that's pulling ahead.