How Far Have Brands Taken AI in Marketing?

A few days ago, a friend who works in marketing vented to me.

He said his company bought all kinds of AI tools last year. At this year's budget meeting, they ran the numbers: efficiency was definitely up, but revenue hadn't grown by a single cent.

I had to laugh when I heard that. Not because his situation was funny, but because it's not just his company's problem. Over the past couple of years, nearly every marketing department has been doing the same thing: rolling out AI. But what happens after the rollout?

Some companies made real money back. Others ended up with nothing but a stack of "AI implementation" presentation decks.

What's the difference?

I recently went through a batch of real-world cases from global brands — from Coca-Cola to Procter & Gamble, from L'Oréal to a UK price-comparison site — and it turns out there really is a method to it. Let me tell you a few of the stories.

First, a Story About a "Granny"

O2, a UK telecom operator, did something genuinely funny.

They trained an AI granny named Daisy. What's she for? Answering scam calls.

Think about it: a scammer calls, thinking he's hooked a tech-illiterate old lady — but on the other end is an AI that chats with him for hours, until he starts questioning his life choices. Every minute of a scammer's time she burns is a minute he can't spend conning real people.

Brilliant!

But catching scammers isn't really O2's main point. As a side effect, it tells every customer: this is the level of our anti-scam technology. And just like that, the brand's image locks into place.

Which brings me to my first point: the most valuable use of AI in marketing is often not "efficiency" — it's doing something that simply couldn't be done before.

Sounds abstract? No problem, keep reading.

The Price-Comparison Site's Mascot Came to Life

Here's another one.

There's a UK insurance price-comparison company called Compare the Market. They have a mascot called AutoSergei — the persona of an extremely reliable "advisor."

What did the company do? They spent nine months turning AutoSergei from an ad character into an AI tool that actually works. It watches insurance prices across providers for you, nudges you when it's time to renew, and offers advice when you need it.

The result? They sent out fewer communications than before, but better-targeted ones. CRM revenue rose 55%.

Here's the counterintuitive part — let that sink in: the fewer messages you send, the more you earn.

The old marketing playbook was more reach, more impressions, more coupons. AI makes "fewer but sharper" calculable for the first time. Fewer messages means people actually read them; when they read them, conversion follows.

What is precision marketing, really? Simply put: saying exactly the right thing, at exactly the right moment. It used to be impossible because no one could do the math. Now AI does it for you.

Coca-Cola's "Dual Velocity" Play

So you might ask: what do the big brands do?

Coca-Cola's marketing team coined a phrase: "dual velocity." What does it mean?

One leg sprints: use generative AI to produce assets fast, test fast, iterate fast — dozens of ad concepts tried within a single day.

The other leg walks slowly: brand-level things — values, aesthetics, long-term equity — still rest with humans, never carelessly handed to a machine to generate.

I especially love one of their stated principles: dive in and experiment actively, but never use AI for the sake of using AI.

That may sound like PR-speak, but compare it with the norm and you'll see how rare it is. How many companies' AI strategy boils down to "the boss demanded AI, so we used AI"?

Procter & Gamble (P&G) walks a different version of the same path. They use AI for ad testing and optimization — testing cycles that used to take weeks compressed into days, at one-tenth the cost. Note: the money and time saved still get plowed back into figuring out which creative actually works.

Tools getting faster is meant to make judgment sharper — not to let you spam ten times more junk.

The Overlooked Half: Keeping People

At this point you might assume I'm about to say "AI will fully replace marketers."

Quite the opposite.

An executive in charge of data and media at the consumer goods company Reckitt once said something I think deserves to be painted on the wall of every marketing department: generative AI should "replace tasks, not jobs."

What does that mean? Writing first drafts, translating, assembling assets, running tests — those "tasks" go to AI. But judging which insight is right, and which line fits the brand — that's where the value of the "job" actually grows.

There's an even more uncomfortable warning: if you use AI to cut every junior role, the money you save buys out your own future talent pipeline. Today's junior staff are tomorrow's marketing directors, ten years on. Freeing them from drudge work to do things that actually build real skill — that's the right answer.

Then there's L'Oréal. Asmita Dubey, their Global Chief Digital & Marketing Officer, takes it a step further: for marketers today, the most valuable skill is being a "critic" of AI output, not a "conductor" of it.

AI can hand you a hundred versions of a plan. Your job is to be capable of asking the one question that punctures it.

What a line.

What's Blocking Most Companies Isn't the Technology

Finally, something low-key but possibly most important.

Research looking at companies with the highest AI ROI found a pattern: they aren't the ones buying the most tools — they're the ones with the clearest governance.

What's governance? It's whether anyone in the company can answer these questions: What is our AI strategy? Who can use it, and for what? Who manages the data? Who's accountable when something goes wrong?

Companies that can't answer mostly stay stuck at the "pilot" stage. One department runs a trial, another department runs a trial, a busy year goes by — and not a single business process company-wide has actually changed.

The counterexample is an insurance company called YuLife. They rolled out AI across the entire company in six weeks. Why so fast? Because it was led by the product and technology teams, run like a product launch — with goals, a timeline, and an owner.

Rolling out AI should be like launching a product, not sending a memo. Memo-senders are done when they hit send; product-launchers have only just begun.

One more caution: the starting point of consumer purchases is quietly moving into AI search and AI assistants. Users ask, compare, and decide entirely within a conversation — never reaching your website. While you're still agonizing over "acquisition vs. retention," that may be the wrong question. The real risk is that customers slip away where you can't see them, and you never even know.

Back to the Beginning

That friend asked me: so what should we actually do?

Having gone through all these cases, my answer boils down to three sentences.

First, don't treat AI as a tailor who only saves you money — treat it as a hired hand that can do new work. O2's granny and the comparison site's AutoSergei are both "new work."

Second, hand the fast parts to machines and keep the slow parts for humans. Fast is volume; slow is judgment.

Third, figure out "who's responsible" before debating "which tools to buy."

There will always be more tools to buy, and the questions keep changing. May you never be the one holding a pile of AI tools who, at the budget meeting, can't answer the question "how much money did we make?"