How Are 140 Brands Actually Using AI in Marketing? I Dug Up a Real List

A while ago, a friend asked me a question.

He said, Liu Run, look — AI is everywhere right now. The marketing world talks about "AI empowerment" and "AI reshaping" every single day. But who is actually using it? What results are they getting? Can you give me a list I can actually trust?

I thought about it and said: let me dig around.

So I started digging. Through earnings reports, product launches, official statements from the brands themselves. By the end, I had something in my hands.

147 cases. 140 brands. 12 directions.

Specific tools, specific results, specific people. Not talk. Actual work done. Today, I want to walk you through this properly.

Let's Start With the Most Basic Question

What does it mean for a brand to be "actually using AI"?

It doesn't mean issuing a press release about "embracing AI." It means using AI to do a specific thing — and that thing has a number you can point to.

Like: conversion rate went up by X percent. Costs cut by Y. Z hours saved. How many users actually engaged.

If there's no number, I basically don't buy it.

So the first rule of this list: there must be a real, named brand. There must be a verifiable source. There must be a concrete result. Miss any one of those three, and it doesn't make the cut.

Alright. Rules are set. Let's go.

Advertising: The Most Visible, and the Most Overrated

Open up any brand's AI marketing report card, and the first thing on the list is always advertising.

Why? Because Meta and Google turned this into infrastructure. The moment you spend money on ads, their AI is already working for you. You don't need to understand it. You don't even need to know it's there.

Meta's Advantage+ automated everything — audience targeting, creative, placement, budget allocation. Meta itself says this system boosted advertisers' return on ad spend by 32%.

LinkedIn's Accelerate is even more aggressive. It reads your website, your LinkedIn page, your historical ad data — then writes the creative itself and finds the audience on its own. LinkedIn says it cut cost-per-action by more than half and compressed the time to build a campaign from 15 hours down to 5 minutes.

15 hours to 5 minutes.

Think about that gap. An optimizer who used to handle two campaigns a week can now do ten in a day. The people saved in that process either move into strategy — or they get laid off.

But there's a truth here I need to be upfront about.

Numbers reported by the platforms themselves will always cherry-pick the best-looking figures.

Lululemon ran Performance Max in Canada and saw ad returns go up 8%. Eight percent — not 32%, not 50%. That's the level most brands actually experience.

Advertising platform AI genuinely works — but it's the category where the impact gets most inflated.

The numbers that actually surprise you are somewhere else.

The Real Savings Happen in Creative Production

Think about how much a single ad used to cost.

Dollar Shave Club's viral video from 2012 cost $4,000. At the time, the entire industry thought that was the limit of cost-cutting.

In 2025, they made a new campaign called "250 Years, No BS." How much?

$400.

One-tenth.

They used Higgsfield and Claude. Humans still wrote the script and the comedic beats, but the tools handled rapid iteration. From brief to finished film: one week. The core creative took just three days.

Or look at Too Faced. This beauty brand produced a mascara ad that was entirely AI-generated, using Adobe's Firefly Video Model trained on their own product images. Post-production editing compressed from 4 days to 1. Where did the savings go? Into shooting more model content and hiring influencers.

This is AI's real value in creative.

What it replaces isn't the creative person — it's the production pipeline underneath them. Illustration, video editing, making variants, resizing, translating copy. All that repetitive labor gets absorbed. What remains is the person who defines the problem, writes the brief, and judges which idea is better.

Kalshi, a prediction market platform, went even more extreme. They brought in a creator who used Google's Veo 3 to produce a national TV ad — aired during the YouTube TV livestream of Game 3 of the 2025 NBA Finals.

Cost: about $2,000. Done in two days.

A traditional production of the same scale? Seven figures. Several months.

300 to 400 generations, culled down to 15 usable clips. That's the reality. AI-generated content has a high failure rate — but even so, the cost got compressed to one-hundredth of what it used to be.

That's the power of scale.

Personalization: It's Not a Feature, It's the Product Itself

In this category, I want to start with a story.

Spotify has a feature called DJ. It picks songs, talks between tracks like a radio host, and — crucially — speaks to your specific taste. It launched in 2023, and by May 2026 it had expanded to over 75 markets, adding French, German, Italian, and Brazilian Portuguese.

It shaped the listening habits of 94 million Premium subscribers.

Ninety-four million. Think about that number. A major Chinese megacity has maybe twenty to thirty million people. This DJ influences a population larger than many countries.

But Spotify isn't a story about "a marketing team using AI." AI is the product. Without this recommendation system, Spotify doesn't survive.

Netflix is the same. Their recommendation engine goes beyond the front-page slots — it's woven into personalized messaging and push notifications. Netflix has published papers on recommendation systems, contextual bandit algorithms, reinforcement learning, foundation models, and causal inference. Their goal is crystal clear: make you spend less time searching and more time watching.

When something goes from being a "feature" to being "the product itself," it's no longer a competitive advantage — it's a survival baseline.

Most brands haven't reached that point. But they're working toward it.

Sephora's Virtual Artist uses your phone camera to detect your face and overlay lipstick, blush, and contour. Over 1,000 shades. There's also AI shade matching — snap a photo, it tells you what color that is and recommends a similar Sephora product.

Financial Times did something even smarter. They replaced their paywall.

It used to be simple: read 5 articles free, the 6th one triggers a subscription prompt. Now it's a machine learning model — it decides who to ask for a subscription and who to let through. Launched in January 2025, FT reported a 290% increase in conversion rate, and lifetime value rose 7% to 10% among users exposed to the model.

The same wall — now a wall that thinks.

Sainsbury's, the British supermarket chain, invested £70 million in machine learning for personalized pricing and recommendations. It covers 18 million Nectar loyalty members and generates 260 million personalized offers per week. The result?

Fruit and vegetable sales went up by 130 million portions.

130 million portions. A supermarket used AI recommendations to get customers to buy 130 million more portions of produce. That's what personalization can do.

Customer Service: The Category With the Biggest Lesson

In this category, I want to focus on one brand.

Klarna.

Klarna is a payments company. In 2024, they deployed an AI customer service agent and loudly announced: the AI handles 2.3 million conversations per month, equivalent to 700 full-time agents. Repeat inquiries dropped 25%. Profit improved by $40 million.

Seven hundred jobs. Gone.

At the time, every headline in the industry was writing about it. The benchmark case of AI replacing human labor.

And then?

Then Klarna quietly walked it back. Customer satisfaction couldn't hold up, and they had to return to a hybrid "AI + human" model. AI handles the first layer of simple volume. Humans carry the complex cases — the ones requiring judgment, the ones requiring warmth.

That's the most important lesson in this category.

AI in customer service works best as a filter layer, not a replacement. It catches the repetitive, scriptable questions and frees up real humans to handle what genuinely needs a human.

That's exactly what Verizon did. They equipped 28,000 customer service representatives with a Gemini-based assistant trained on 15,000 internal documents. Reps can pull up answers in real time during calls. After deployment, the service team's sales went up nearly 40%. Verizon didn't lay anyone off — they retrained their customer service reps into sales.

Bank of America's Erica surpassed 3 billion customer interactions by August 2025, with monthly interactions exceeding 58 million, serving nearly 50 million users. Allstate sends out about 50,000 claims communication emails per day — almost entirely drafted by AI, reviewed by humans.

The pattern is the same: AI drafts, humans guard the gate.

Lowe's deployed an OpenAI-powered assistant across more than 1,700 US stores, reaching 300,000 employees. Staff on the floor use handheld devices to ask it about product details, project recommendations, and inventory. Lowe's says this is the first time a tool of this kind has been deployed at this scale in retail.

Scale matters. One AI assistant is easy. An AI assistant that 300,000 people actually use is a completely different thing.

Inside the Enterprise: The Area Leaders Care About Most But Hear About Least

This is the category I get asked about the most.

"Should our company build an internal AI platform? Is it worth it?"

Let me give you some numbers.

JPMorgan. They built an internal LLM Suite and rolled it out to 140,000 employees in September 2024. At the time, President Daniel Pinto raised the expected value of AI to nearly $2 billion, with most of that tied to anti-fraud.

$2 billion. One bank.

IBM applied AI agents across more than 70 of its own business areas, covering 270,000 employees. Over two years, they reported $3.5 billion in productivity impact. Their AskHR agent handled 94% of simple requests. AskIT cut IT support team calls and chats by 70%.

Cisco built its own internal AI assistant — by November 2025 it was serving over 100,000 users, processing 45 million interactions, averaging 156,000 per day. Users reported saving an average of 5 hours per week, with 73% saying their productivity had improved.

5 hours saved per week. Over 52 weeks, that's 260 hours. For 100,000 people, that's 26 million hours.

Put that number on the table, and it's impossible for a CEO not to pay attention.

Nestlé's NesGPT is built on the same technology as ChatGPT. In the first three months, over 7,000 US-based employees ran nearly 230,000 prompts. Users reported saving an average of 45 minutes per week. Forty-five minutes isn't much — but multiplied by 7,000 people, multiplied by 52 weeks, that's over 54,000 hours. And that's just a portion of the US workforce.

Why is internal enablement the topic enterprise leaders care about most? Because the returns here are the most direct, the most quantifiable, and the hardest to fake. You invest $100 million, and you can calculate exactly how much labor you saved and how many decisions you accelerated this year. Unlike "brand exposure" from advertising, this is real accounting.

A Few More Underappreciated Directions

Let me quickly run through a few more areas. They're not as flashy as advertising, but they're quietly doing work inside brands.

Supply chain. Walmart's route optimization technology eliminated 30 million miles of wasteful transportation and avoided 94 million pounds of carbon emissions. Target's inventory ledger processes 360,000 transactions per second. Zara uses AI for demand forecasting and inventory management — 2025 sales were up 7.1% year-over-year.

Fraud and security. This is the area where AI has been working the longest and most reliably. Stripe's payment foundation model helped merchants recover $6 billion in legitimate transactions that were wrongly declined in 2024. During the Black Friday to Cyber Monday window in 2025, Visa's AI intercepted 144% more suspicious fraud activity than the previous year. HSBC replaced rule-based anti-money laundering screening with AI — detecting 2 to 4 times more financial crime, with 60% fewer false positives. The time to analyze billions of transactions dropped from weeks to days.

This is AI's hidden force outside of marketing. It doesn't show up in consumers' line of sight, but it affects profit, inventory, and risk control.

So, If You're a Brand Leader, What Should You Do?

I'm not going to give you a "five-step strategy."

But there are a few things I noticed across these 147 cases that I can share with you.

First, start with whatever your team touches most every week.

A/B testing for email personalization, rapid ad creative variant generation — these are the easiest places to begin. Measurable, reversible, no system rebuild required. You're already using Meta Advantage+. You're already using Klaviyo. Squeeze every drop out of those first.

Second, internal enablement returns come faster than customer-facing AI.

Give your customer service team a document assistant. Give your marketing team an agent that can read consumer research. Give your sales team a tool that can draft RFP (Request for Proposal) responses. These aren't sexy — but the time they save is real.

Hyatt put AI to work on corporate sales. They handle 1.5 million corporate RFPs a year. AI drafts the responses; sales shifts from writing from scratch to reviewing and editing. The CEO said each salesperson saves an entire day per week. One entire day.

Third, creative is where the biggest savings live — but the brief is still written by a human.

Dollar Shave Club, Too Faced, Kalshi — these cases tell you: AI replaces the production pipeline, not the creative person. If you have someone who can write a brief, you can cut execution costs to a tenth or even a hundredth of what they were. If you don't have that person, AI can't help you.

Fourth, in customer service — remember Klarna's lesson.

AI is a filter layer, not a replacement. You can let AI handle 70% to 90% of the simple volume, but that last layer — the complex, the emotional, the judgment calls — has to be human. If you don't reserve that layer, customer satisfaction will humble you real fast.

Fifth, don't wait.

Let me be blunt about this one. It's 2026. If you're still asking "should we use AI?" — you're already late. The question hasn't been "should we" for a long time. It's where to start, how to measure, and how to scale.

Finally

As I was going through these 147 cases, one feeling really struck me.

Real change never arrives as a single thunderclap. It arrives as 140 brands, each quietly, getting one thing done.

Coca-Cola used AI to remake their classic 1990s Christmas ad. Heinz asked DALL-E to draw ketchup bottles — and every image the AI produced looked unmistakably like Heinz. Cadbury used machine learning to turn one Shah Rukh Khan ad into thousands of localized ads for small businesses. Mint Mobile had Ryan Reynolds read copy written by ChatGPT — he himself said it was "a little scary."

These things happened one by one. Added together, they amount to a migration.

From "AI is a new thing" to "AI is infrastructure." That migration crossed its tipping point in 2025. By 2026, the question had become something else entirely.

Not "who's using AI" — but "who isn't yet."

I don't know where your brand sits on that spectrum. But I hope that, after reading these 147 real stories, you can find your own starting point.

That's enough.