The AI Riddle of the Marketing World

The other day I scrolled past a number that made me put down my phone.

88% of marketers use AI tools every day.

Sounds like a lot, right?

But the small number behind it is the alarming one: in a 2025 AI report from a top consulting firm, only 6% are actually making money from AI.

88% use it. 6% earn from it.

Sit with those two sentences for a second.

What's the Gap Between Them?

I couldn't figure it out at first. You buy a tool, you use it — how do you still not make money?

Then I stared at a few companies that actually do make money from AI, and I noticed something fascinating.

This crowd isn't using AI to write copy at all.

They're using AI to tell fortunes — their nickname for predictive analytics that answers "what happens next," not "what happened last quarter."

What do I mean by "tell fortunes"?

This: which existing customer is about to leave next quarter? Which leads, if you follow up today, will actually close? Which product is going to blow up next month? They want tomorrow's answers, not yesterday's reports.

A report tells you "we dropped 15% last quarter." Fortune-telling tells you "next quarter you'll keep dropping — but here are 12 people you can win back today if you act."

One looks backward. One looks forward.

That one breath — that's the whole gap.

Let Me Tell You About 10 Companies

I spent a few weeks digging through every case I could find of companies that have squeezed real money out of AI. I picked 10, each with measurable ROI.

Let me tell you — by the time I finished, even I couldn't sit still.

Company 1: a coffee chain.

It has an internal AI platform. It has zero intention of helping you write social posts. It does exactly one thing: feeds in your purchase history, your location, today's weather, the time of day — and calculates the one coffee you're most likely to buy right this second, then pushes it to you.

Result? Members who got these AI pushes spent 3x what non-recipients spent. Overall ROI rose 30%, and the supply chain saves $125M a year.

Wow.

Company 2: a beauty retailer.

They rolled "customer lifetime value prediction," "virtual makeup try-on," and "AI recommendations" all into one system. Customers who tried the virtual try-on converted at 3x the rate of regular users. Customer lifetime value rose 29%.

What surprised me most: their marketing-content production costs dropped 38%.

Cost down, revenue up — they didn't trade one for the other.

Company 3: the streaming giant.

Everyone knows this one. Its recommendation engine shapes 80% of what users watch, and the money it saves from reduced churn clears $1B every year.

$1B. Just from "stuffing you a show you can't pull away from right before you hit cancel."

Its success rate on original content hit 93%. The industry average is 35%. This isn't marketing anymore — this is using AI to redefine the product itself.

Sharp.

Company 4: a writing-assistant tool.

They have 30 million daily active users on the free tier. The problem: with that many free users, the sales team couldn't tell who to chase and who to drop.

The old way: hand sales 400-plus leads a month, half of them bots, barely any genuinely worth talking to.

After AI took over, it picked out just 200 high-scoring leads. Account upgrades jumped 80%, and the sales cycle compressed from 60-90 days to 30. The email unsubscribe rate was 0.04% — the industry average is 2%.

The head of their marketing ops said something I copied down word for word: "We shifted from chasing volume to chasing precision."

Well put.

Company 5: an insurance company.

Its AI model was trained on 10 billion miles of driving data. It can pinpoint the exact second — right now — when a user is most ready to buy insurance.

If yes, it pops a "Buy Now" button inside their mobile app.

That one move drives $2B in new premium revenue a year.

See the pattern?

Prediction alone is worthless. Prediction plus "press the button for the right person, on the right channel, at the right moment" — that's worth $2B.

Company 6: a commercial bank.

It uses AI to score and rank all leads across the bank in real time, and shares them across departments. Lead conversion rose 2.35x. Manual lead-screening time was cut by more than 60%.

The most interesting part is its churn prediction. It doesn't just give you a churn score — it tells you why this person is about to leave.

The "why" matters.

Give me only a score and I can only stare helplessly. Give me a "why" and I know what discount to offer, what to throw in, who to put on the phone.

Company 7: a subscription fashion e-commerce.

Its core is machine learning. Calling it a clothing company undersells it — it's an algorithm company that happens to sell clothes.

75% of the selection in each box is decided by AI. Average order value rose year-over-year for 7 straight quarters, with the most recent quarter up 9%.

It also acquired one company for CLV prediction and another for demand sensing, then jammed both into its recommendation engine. Layer on layer.

This taught me something: the companies that make real money from AI never rely on a single model. They stack multiple predictive capabilities and braid them into one rope.

Company 8: another beauty giant.

Its virtual try-on was used more than 100M times in a year, and users who tried it converted at 3x the rate of regular users. Marketing campaigns with AR experiences converted 20% to 80% above baseline.

Even wilder: it has a tool that scans 3,500+ sources to predict beauty trends 6 to 18 months out.

By the time you've just noticed "this year's color is X," they stocked the shelves half a year ago.

This isn't prediction — it's clairvoyance.

Company 9: a fast-fashion brand.

In this industry, most companies sell only 60% of inventory at full price. This one sells 85%.

That extra twenty-something percent is worth billions.

Its AI system pulls real-time sales data back from 6,000+ stores worldwide, down to how many units of each SKU every single store should stock. From design to shelf: 10-15 days. Industry average: 3 to 6 months.

Inventory-management costs fell 20%, forecast error dropped from 21% to 17%, and potential stockout losses fell 24%.

It even knows what you'll want to buy next month before you do.

Company 10: the largest e-commerce player.

Its recommendation engine processes 150 billion user-behavior signals a day across 600M+ products. Revenue driven by recommendations is 35% of total revenue.

35%. At its scale, that's more than $70B.

It tells you a plain truth: once you can accurately guess "what does the customer want next," everything else gets simple.

What Did These 10 Companies Get Right?

I pulled out the common threads, and there are four of them.

One: they all use AI to look forward, not backward.

Ordinary companies use AI to write copy, pick images, auto-send emails. Honestly, all of that is just "doing yesterday's work a little faster."

The top companies use AI to answer: what will happen next month, next quarter, next second?

Two: they all wired "prediction" to "action."

A churn score sitting on a dashboard won't keep a single customer. You need a button, a workflow, and someone to press that button.

The best AI companies, every single one, have welded prediction to action.

Three: they all start from the business problem — business first, tech second.

Not one of them said "let me evaluate AI tools" and then went looking for "what can I solve with it." They all started by asking —

How do I keep more customers? Who should sales chase? What will demand look like next quarter?

Get the question clear, then bring the tech onstage.

Four: they all redesigned their workflows.

This is the observation from the same consulting firm: the top-performing 6% are 3x more likely to have redesigned their workflows around what AI outputs.

They tore the old workflow down and rebuilt around what AI can do. Bolting AI onto an old workflow doesn't count.

Sharp.

So What Should Ordinary Companies Do?

Honestly, when I saw these numbers, I felt a little anxious.

Not for myself — for the friends who "just bought an AI tool and figured they were done."

Industry reports predict the AI-in-marketing market will grow from $20B+ in 2025 to $82B in 2030.

But what's actually valuable isn't "did you use AI" — it's "what question did you ask AI."

The 10 companies I just walked through built in-house AI teams and spent years and tens of millions of dollars grinding these capabilities out. For most companies, that bar is simply too high.

But here's the good news: these capabilities are becoming purchasable.

A few predictive-AI companies already let marketing teams ask in plain language: "Which customers will churn next month?" "Which leads are most likely to close this week?" You ask, you get back predictions you can act on.

It's the difference between fortune-telling being a temple abbot's secret and a street-corner trade — anyone can get a reading.

On average, tools like these cut churn by 12%, lift customer lifetime value by 10%, and lift ad-spend return by 15%.

Not earth-shattering numbers. But for the vast majority of marketing teams, that's already enough to recoup the cost.

In Closing

When I finished this piece and closed my laptop, only one sentence was left in my head.

88% use it. 6% earn from it.

What the 82% in the middle are missing was never the tool. What they're missing is: are you using AI to tell fortunes, or to do homework.

Anyone can do homework. Fortune-telling — for that, you first have to figure out what to ask.

Maybe that's the one thing in this AI wave most worth thinking hard about.

Here's hoping — you ask the right one.