Does AI Marketing Actually Work? I Went Digging Through a Stack of Ledgers with Real Numbers

A while back, a friend of mine who runs an e-commerce shop asked me: "Run, be honest — is AI marketing actually useful, or are we all just hyping each other up?"

I said: good question. But armchair debate gets us nowhere. Let's look at the books.

I've recently gone through a pile of numbers companies have published themselves. Not feelings — ledgers. How much conversion rates rose, how much time was saved, how much more money was made. After reading through them, here's my honest take: some of these numbers made me sit up straight.

Let me walk you through a few.

First, a Bra Seller

Adore Me, a DTC lingerie brand. They had one chronically painful task: writing product descriptions.

One batch of descriptions used to take 20 hours. Twenty hours — nearly three working days — burned on "This bra features…"

Then they adopted Writer's AI tool, and specifically trained it on their own brand voice. The result?

Twenty hours per batch became 20 minutes.

You read that right. Not 20 hours down to 15 — down to 20 minutes. Time-to-launch in a new market dropped from months to 10 days. Non-branded SEO traffic rose 40%.

And there's a detail here worth pausing on: they didn't just bluff their way through with generic AI. They put real effort into teaching the AI their brand voice. That's the gap between taught and untaught.

The same tool only becomes your tool once you feed it what's yours.

Next, a Tamale Seller

You might assume this is a big-company game. Wrong.

There's a small family business in the US called The Original Tamale Company, selling tamales (a Mexican corn-dough dish). No marketing department, no budget — the owner runs it all herself.

She writes scripts with ChatGPT, edits with a cheap video tool, and jumps on trends with meme-style content. Each video takes roughly 10 minutes to produce.

The result: 22 million views, 1.2 million likes in three weeks, and visibly more foot traffic in the physical store.

Ten minutes. Competing with big brands for the same pool of attention.

So stop asking "can small companies afford this?" When the barrier collapses, it doesn't send anyone a warning first.

Numbers Talk. Here Are a Few Entries from the Ledger

Stories alone aren't satisfying — let's read the numbers. All of these are publicly verifiable:

AS Watson Group deployed Revieve's AI skin analysis, moving that in-store "your skin type would really suit…" consult online. Customers who used the AI advisor converted at 396% higher rates than those who didn't, spent 4x more per person, and had average order values 29% higher.

Heinz — yes, the ketchup company. They used DALL-E 2 to generate images and turned "AI draws Heinz" into the campaign itself, inviting the public to submit prompts, and actually shipped limited-edition bottles. 850 million organic impressions, social engagement 38% higher than their previous campaign, and 25x media ROI.

HubSpot applied AI to their internal emails, shifting from "which segment does this person belong to" to "what does this person want right now," using AI to predict each individual's intent. Conversions up 82%, open rates up 30%, click rates up 50%.

Verizon uses generative AI to predict why customers are calling support, correctly guessing the reason for 80% of incoming calls ahead of time. Each in-store interaction saves 7 minutes, and they estimate avoiding roughly 100,000 customer losses. Note: they didn't use AI to replace support staff — they gave support better intelligence.

L'Oréal's virtual try-on has been used over 1 billion times, delivered more than 20 million personalized skin diagnostics, and converts at 3x.

And Cadbury's "Not a Cadbury Ad" campaign used generative AI to produce over 2,500 localized video ads, starring Bollywood megastar Shah Rukh Khan — except in each ad, he was "endorsing" a different small shop. It reached 140 million people, with engagement soaring 32%.

Good grief. Over 2,500 ads. A production volume nobody would have dared imagine before.

The Big Tech Report Card

Google recently (April 2026) compiled a collection of 1,048 real-world deployment cases. I picked a few entries from it too:

Etsy, the handmade-goods marketplace, uses Vertex AI and Gemini to process a catalog of 130 million items — product understanding is 80x more efficient, 90 million shoppers get personalized recommendations, and search relevance improved 3–5%.

Auto parts company Valeo gave Gemini Code Assist to all 100,000 employees. Now 35% of their code is AI-generated.

WPP, the advertising giant, runs creative production with over 100,000 Gemini agents. The value created works out to 2.5x, and they ship campaigns at a pace of one every 4 days.

Gazelle, a Brazilian real-estate content company, cut a single property description from 4 hours to 10 seconds.

Swarovski personalizes across more than 140 markets: email open rates 17% higher, localization efficiency 10x.

I could keep reading entries off this ledger. Radisson: productivity up 50%, revenue up 20%. Sojern: customer acquisition costs down 20–50%. Game company Square Enix: personalized email open rates up 20%…

But reciting numbers isn't the point. The patterns behind the numbers are.

Three Patterns Hidden in the Ledger

Laying all these cases out, I noticed the winners' moves fall into roughly three categories.

Pattern one: train AI like one of your own, instead of treating it like an outsourcer.

Adore Me taught AI its brand voice. Virgin Holidays used Phrasee to write email subject lines — also feeding it their own tone — and gained a 2% lift in open rates. And remember, at their scale, 2% is millions of pounds in incremental revenue. Then there's Vector, a B2B software company that took the CEO's own best two or three dozen posts to train an AI, then fed it interview transcripts weekly to keep things fresh. LinkedIn followers went from 7,000 to 11,000, inbound demo requests quadrupled, and they ship 4–5 pieces of high-quality content a week with just 15 minutes of human review.

Generic, it's a tool. Trained on your voice, it's a colleague.

Pattern two: make people stronger, instead of replacing them.

Verizon is the clearest example. That 80% call prediction wasn't used to cut support headcount — it was used so support agents know why you're calling before they pick up. HubSpot is the same: from segmenting to mind-reading, their sales and ops people get sharper intelligence.

The companies thinking "use AI to cut people" post mediocre numbers; the ones thinking "use AI to make people stronger" post numbers that are frankly scary.

Pattern three: dare to put AI on stage.

Heinz made the AI painting process the campaign itself. Cadbury had AI-generated superstar ads cheer for little shops. Moving-and-storage company PODS went further, using Gemini to turn its New York trucks into "the world's smartest billboards" — copy swapped in real time by neighborhood, covering all 299 neighborhoods in 29 hours, dynamically generating over 6,000 distinct headlines.

Hide AI in the back office and it's just a cost-cutting tool. Put AI on stage and it becomes the story itself.

The best ad space for AI might just be AI itself.

Finally, a Bucket of Cold Water

By this point you may be fired up, ready to go all-in on AI starting tomorrow.

Slow down.

These ledgers share one common flaw: they only record the wins. Implementation cost? Not mentioned. Time to results? Not mentioned. How many versions failed before it worked? Definitely not mentioned. And whether you'll need to keep feeding it, maintaining data, tuning models forever? Also not mentioned.

So treat these numbers as direction, not as guarantees. Someone else's 396% is someone else's. Your share has to be built, entry by entry, by you.

But the direction itself, I'd argue, is already clear. Just like the tamale lady at the beginning — she won't debate you on "will AI replace marketing." She only cares about her next 10-minute video.

And maybe the most valuable pattern in this whole ledger is this: the people who start while everyone else is still debating.

Here's to finding your own ledger soon.