Is the Money You're Pouring into AI Marketing Actually Worth It? Let Me Show You Ten Companies' Books

A while back, a friend of mine who works as a CMO invited me out for tea.

He said his budget had been cut, yet the money his boss allocated to AI tools actually went up. The reasoning came down to one sentence: everyone else is using it, and if we don't, we lose.

He asked me: tell me, is this money actually well spent?

I said, stories won't answer that question. Slide decks certainly won't.

You have to look at the books.

As it happens, I've recently gone through the books of ten companies. Retailers, consumer goods, software, banking, film — plus one Singaporean small business. All kinds of industries, but every single entry on the ledger is verifiable.

Today, I'll walk you through these ten ledgers, one by one.

First, a Baseline Number

Before we open the books, let me give you a reference point.

McKinsey published a State of AI report in 2023. It noted that companies that genuinely weave AI into their marketing function see revenue gains of 3% to 15%, and sales ROI improvements of 10% to 20%.

3% to 15% — what does that mean?

Think about it. For a company with $1 billion in annual revenue, 3% is $30 million, and 15% is $150 million. That is no longer the magnitude of "trimming a few expenses." That is a change of league.

Why didn't people believe it a few years ago? Because back then, AI marketing was still stuck in the pilot stage. Big brands played in closed-door sandboxes while everyone else watched through the glass.

Now the glass is gone.

All right. Let's open the books.

Sephora: One 11%, One Hundred Million Dollars

Let's start with Sephora.

Sephora installed an AI personalization engine in its app and on its website. What you've bought, your skin tone data, what you've been browsing lately — it remembers all of it, then makes recommendations like a personal beauty advisor.

Customers who used its Virtual Artist virtual try-on converted 11% better than those who didn't. Even more striking: over one fiscal year, Sephora attributed more than $100 million in incremental revenue to the personalization engine.

$100 million.

One recommendation engine produced the annual revenue of a mid-sized company.

There's a detail here worth chewing on. Sephora didn't push the same wave of "summer mega-sale" promotions to everyone. It served every customer as an audience of one. That's the difference between AI personalization and traditional segmentation: every time the model gets used, it gets a little smarter.

The more it's used, the more accurate it gets. The more accurate, the more money it makes.

That's compounding.

Coca-Cola: Cutting Content Production Time in Half

The second ledger: Coca-Cola.

Coca-Cola was an early big brand to put generative AI on the table, teaming up with OpenAI and Bain to do it together. In 2023 it ran a campaign that let consumers directly create branded creative content with GPT-4 and DALL·E.

The consumers had fun, and the company's books looked good too: content production cycles were shortened by up to 50%, and creative iteration costs dropped sharply across global markets.

What does 50% mean?

For a company running thousands of localized campaigns across dozens of markets at the same time, that doubles content capacity — without hiring a single additional person.

That's what AI does at the execution layer: it removes the human bottleneck, but keeps human judgment at the strategy layer.

Alibaba and JPMorgan: Can Machine-Written Copy Really Outperform?

At this point, you might say: hold on, can machines really write as well as humans?

Good question. Let me show you two more ledgers.

Alibaba's AI copywriting tool, Luban, can generate twenty thousand product descriptions for platform merchants in one second. Twenty thousand, in a second. No human team could ever come close.

But fast doesn't equal good, right? The key is the A/B testing that followed: AI-written product descriptions got click-through rates on average 8% higher than human-written ones.

It's not an isolated case. JPMorgan took AI-generated ad copy and ran controlled experiments against copy written by its internal marketing team, using Persado's natural language generation technology. The AI version won every time — in some scenarios its click-through rate was more than double the human version's. JPMorgan then signed a five-year enterprise contract with Persado.

A bank. The industry with the strictest compliance rules and the most conservative culture. Even in shackles, AI outdanced the professional copy team.

Unbelievable.

What does that tell us? It tells us that "humans definitely write better" is an assumption, not a fact. When AI has been trained on high-converting samples and is optimized toward an explicit conversion intent, it wins on scale — and it can win on a single point too.

Unilever and Netflix: Whatever Gets Saved Is Pure Profit

Now two more ledgers, the "money-saving" kind.

Unilever rolled AI out across its global programmatic advertising, using machine learning to adjust bids, audiences, and creative in real time. The result: customer acquisition costs fell by 25%, without missing a single reach target. Given the scale of its ad spend, that's hundreds of millions saved every year.

Where does the saved money go? To put it bluntly, straight into profit.

Netflix is even more extreme. Its recommendation system governs every poster you see and every title that gets pushed to your homepage. Netflix has disclosed that this system saves it roughly $1 billion a year in customer retention value. Without it, churn would rise noticeably, and Netflix would have to spend far more buying content to fill the hole.

There's another easily overlooked detail. A sizable chunk of Netflix's AI lives in the marketing layer, not the product layer. The same film gets different thumbnails for different users — and relying on this computer-vision image selection, click-through rates can differ by 20% to 30%.

The right content, paired with the right image, delivered at the right moment to the right person.

That's what AI marketing looks like at its finest.

Spotify and Starbucks: Serving an Audience of One

These two companies' ledgers tell the same story.

Spotify's Wrapped year-end recap goes viral every December, but the personalization engine behind it runs nonstop, 365 days a year. Every email, every push notification, every in-app message is tailored to the individual user. AI-generated personalized emails get open rates two to three times those of mass-blast emails.

Starbucks takes it into everyday life. Its internal AI platform, Deep Brew, looks at your purchase history, location, time of day — even the local weather — and then pushes an offer to you at exactly the moment you're "just in the mood for a drink." After AI personalization went live, offer redemption rates hit three times the level of the old rule-based segmentation era. And the 30 million active members behind it have always been its most important revenue engine.

3x redemption. 2–3x open rates. Plus Sephora's 11%.

Have you noticed? Whoever pushes personalization to the individual level gets step-change returns — not patch-up improvements of 5%.

Why? Because no matter how fine the segments get, a segment is still "a crowd of people." Only the individual level is "you."

HubSpot and a Small Company: This Game Isn't Reserved for Giants

After so many giants, you're probably thinking: sure, but they're giants with deep pockets.

Let's look at two small ledgers.

HubSpot built AI lead scoring into its own CRM, and ran it on its internal sales team first before selling it to anyone. Sales teams using AI scoring converted 30% better than those using manual scoring, and reps' time stopped being wasted on hopeless leads.

And then comes my favorite ledger of all.

A Singaporean small business doing B2B services, leaning on a marketing platform like Hashmeta AI, ran six months of "AI content plus AI response": on one side, publishing search content at a steady rhythm — AI-assisted, but with human strategists as gatekeepers — and on the other, letting AI respond to every inbound inquiry within 90 seconds, covering WhatsApp and email alike.

Six months later: organic search traffic up 340%, qualified-lead conversion up 58%.

340%.

No huge budget. No data science team. An ordinary small business, six months.

What AI amplifies is judgment, not budget.

Three Patterns Hidden in These Ten Ledgers

All right, the books are open. Lay all ten ledgers on the table and look at them together — what do you see?

Pattern one: speed. Twenty thousand descriptions in a second; a lead answered in 90 seconds. AI has removed the human execution bottleneck. But note: human judgment is still sitting at the strategy layer — untouched.

Pattern two: precision down to the individual. Segment-level marketing already belongs to the previous era. What individual-level personalization brings is a step change in conversion and retention — not fine-tuning.

Pattern three, the sharpest one: compounding. Every interaction makes the model a little smarter; every additional piece published makes the content library a little thicker. Those who start early pull further ahead every quarter.

Latecomers want to chase? What they need to chase isn't one step — it's a gap that has been snowballing for quarters.

Still, I have to nag one more time.

Among these ten companies, not one won by letting "AI fight alone." Sephora has beauty advisors tuning the recommendation logic; JPMorgan has senior marketers reviewing AI output; Starbucks has strategists designing the mechanisms of its loyalty program.

AI handles speed and scale; humans handle direction and taste.

AI amplifies human expertise. It does not replace it.

So Where Do You Start?

At this point, the question becomes: where do I begin?

For most growth-stage companies, there are three levers with the most force.

First, search visibility. Use AI to produce content that can rank, at a steady volume, while watching two new frontiers: GEO and AEO. What are GEO and AEO? Getting your content cited by ChatGPT, Perplexity, and Google's AI overviews. Users increasingly ask AI instead of searching for links, and this frontier is expanding fast.

Second, lead response speed. Here's a number with almost no dispute around it: a lead contacted within 5 minutes is 9 times more likely to convert than one contacted an hour later. AI response systems are on duty 24/7, catching WhatsApp, email, and web chat alike. An inquiry that lands at 3 a.m. gets answered without losing a single minute.

Third, customer engagement. An AI customer-service agent that remembers conversation history, makes personalized recommendations, and knows to hand off to a human when it can't handle something is, in essence, replicating the retention playbook of Starbucks and Spotify — at small-business cost.

Push open these three doors, and the returns will snowball.

Back to That Cup of Tea

That day, at the end, I told my friend: you asked the wrong question.

You shouldn't ask whether AI is worth it. You should ask yourself when you're going to start. Because these ten ledgers tell us the winners were never the ones with the biggest budgets — they were the ones who moved earliest.

The gap widens every month.

When it comes to AI, the cost of waiting is far higher than the cost of experimenting.

I've turned all the books over to you. The next move is yours.

May you be the one rolling the snowball.