Does AI Marketing Actually Work? One Study Measured 3.8 Billion Interactions — and Put a Number on It

A while back, I had dinner with a friend who works in brand marketing. He said something that has stuck with me ever since.

Here's what he told me: "AI marketing? We've tried all of it. Does it save time? Absolutely. Does it make money? Sorry — I can't show you a single number."

And to my own surprise, I couldn't argue with him.

Think about it: haven't you been hearing the same lines everywhere for years now? AI brings unprecedented productivity, unprecedented conversion rates, unprecedented speed and scale. The word "unprecedented" has been worn smooth from overuse.

But where are the numbers?

Nobody hands you numbers. AI marketing is a black box. Everyone says there's gold inside, but no one has actually opened it up and looked. So more and more people are quietly starting to wonder: does this thing actually work, or is everyone just shouting along with the crowd?

Someone Finally Opened the Box

Blueshift did exactly that. They produced a benchmark study, ROI of AI Marketing: 4 Levers for Cross-Channel Success, that set out — for the first time — to turn AI's real value in marketing into numbers you can actually see.

They analyzed 3.8 billion marketing interactions. Note: not surveys, not interviews, not "respondents say" — real marketing campaigns that ran in the real world, across every channel and every industry.

The results made me do a double take the first time I saw them:

AI-driven marketing campaigns lift customer engagement by up to 7x and revenue by 3x.

7x. 3x.

This isn't one customer's feel-good success story. It's the average across 3.8 billion interactions. Broken down, the key findings look like this:

• Revenue lift: wherever AI sits in the marketing mix, it delivers roughly 3x revenue in that spot;
• Engagement lift: depending on the scenario, engagement rates rise by anywhere from 3.1x to 7.2x;
• Channel differences: on push notifications, AI's engagement lift is roughly 2x what it delivers on email;
• And here's the most interesting one: the AI engine keeps learning from every user interaction. Over time, it adds roughly 50% more lift on top of its early results.

The first few findings are about starting high. That last one is about running faster the longer it goes.

That 50% Is Exactly the Part Everyone Overlooks

What does it mean for an AI engine to learn on its own?

Here's an example. The first time you use AI to pick your audience, your content, and your timing, the results are already good. But that's just the beginning. Every time a user opens, clicks, ignores, or unsubscribes, the AI takes note. The next send, it understands that user a little better. The one after that, a little better still.

It's like onboarding a new hire. In the first month, they're only about seventy percent as good as you, and you're probably thinking you'd rather just do it yourself. But six months in? They're the person who knows your customers best — except they never sleep and never ask for a raise.

That extra 50% lift is the skill that grew over those six months.

So the question that stumped my friend at dinner now has an answer: he probably only experienced "a new hire's first month" — and concluded that AI was nothing special.

So How Do You Actually Use It? Four Words: Who, What, When, and Where

The report breaks AI's value in cross-channel marketing into four levers. Put simply, they answer four questions:

Who to target (The Who). Use predictive audiences to pick out, from all your users, exactly who this particular campaign should go to. Not pestering every single user you have — just the ones worth reaching.

What to say (The What). Use predictive recommendations to decide which piece of content, which offer, which product each person should see. Marketers have been talking about true one-to-one personalization for years. AI makes it actionable for the first time.

When to say it (The When). Use predictive engagement timing to nail the exact moment each person is most likely to open. Same user, sent at 8 a.m. versus 9 p.m. — the outcomes can be worlds apart.

Where to say it (The Where). Use predictive channel preference: if someone loves email, send email; if they live in app push, go push. Your guess will never be as accurate as what their own behavior tells you.

Who, What, When, Where. The four questions marketers bring up every single day — AI turns each one into a problem you can actually compute.

Back to That Dinner

So where exactly did my friend go wrong?

Not with AI. He treated AI as an "efficiency tool" — write copy a bit faster, make creatives a bit faster. But the real leverage of AI marketing lies in decisions: who, what, when, and where. Get each of those four decisions a little more right, and the gains stack up into a gap measured in multiples.

The black box has been opened. The numbers from 3.8 billion interactions are sitting right there: 7x engagement, 3x revenue — not shouted into existence, but measured.

To everyone still on the fence, I have just one line:

The question is no longer "does AI marketing work" — it's "your competitors are already using it."

Here's to being the one who opens the box first.