When Does AI Marketing Finally Start Paying for Itself?

Last week, a friend who works in marketing took me out to lunch.

The moment he sat down, he asked: "You study business all day, so let me ask you a practical question — does AI marketing actually make money?"

I told him to hold on and not rush the answer. "First tell me: has your company's AI budget gone up over the past two years?"

He thought for a moment. "More than doubled."

"And what about your ROI reports?" I asked. "Has anything gone up there?"

He paused for three seconds. "I never calculated it."

I put down my chopsticks and told him a blunt truth: that "never calculated it" of his is the most common answer across the entire industry over the past two or three years.

What Is ROI, Anyway? Money Goes In, and It's Supposed to Come Back Out

What is ROI? You put one dollar in, and after a while it should come back to your account with principal and interest. That payback is your return on investment.

The story AI marketing told from the very start, everyone has heard: efficiency doubles, personalization goes to the max, and decisions get smarter every day.

Visions that big raised expectations sky-high. But once real money was put in, people soon found that the payback came back much slower than imagined.

In 2025, Boston Consulting Group (BCG) published a survey with very blunt numbers:

Only about a quarter of companies made it through the pilot stage and captured visible value from AI.

In other words, the remaining three-quarters sent their money out, and their ROI is still on its way. You might think that is the story of some unlucky company. No — that is the current reality for most companies, and the thought alone sends a chill down the spine.

And this is not a single opinion. Put the same kind of surveys side by side, and the picture is basically the same:

• About 47% of companies say their AI projects have started turning a profit; nearly a third are only breaking even, and the remaining 14% are losing money.
• When IBM asked companies that had not yet seen ROI, more than half said they would have to wait another one to two years before seeing decent savings.
• Deloitte's numbers are in a similar vein: 74% of enterprises have not taken enough value from AI yet.
• Going further back, Gartner was already running the numbers in 2023: that year, only 54% of AI projects survived the pilot stage, and plenty of the survivors still failed to deliver the financial or operating impact they had promised.

So the conclusion is actually fairly clear:

Everyone is investing in AI, but most companies still have not seen the money come in.

Why Can't the Account Be Settled? Four Pitfalls, Each One Fatal

AI deployment is this hard — where exactly does it get blocked? Experts keep going back and forth, and the same four pitfalls keep recurring.

The first pitfall: the data is not clean. No matter how fast your AI runs, your data weighs you down just as heavily. There is an industry saying: data problems can consume around 80% of an AI project's working hours. In a survey by the data service platform Informatica, 43% of companies also listed data quality as their number one stumbling block.

The second pitfall: the account was never really settled. The tech is bought, the pilot is run, but few people have seriously calculated exactly which revenue this feature is supposed to earn back. In Deloitte's data, 39% of companies get stuck with an unclear strategy, no real usage, and no way to scale. And about half of all AI projects quietly shut down before even reaching production deployment.

The third pitfall: the people have not kept up. In Informatica's data, 35% of companies say what they lack is never the tools, but people who understand data and know how to use AI. Once the projects are rolled out and nobody runs them, they naturally never make it to production.

The fourth pitfall: risk is underestimated. Some estimates put it as high as 85% of AI projects delivering skewed or even misleading results, because of poor expectation management and landmines buried under the data. Deployment costs and infrastructure get underestimated by as much as ten times. And then there is compliance. In a survey by Accenture, fewer than 1% of enterprises are "fully prepared" for the new AI regulations expected over the next five years, and 45% of executives believe the probability of a major AI incident in the next 12 months is greater than one in four.

Data, accounting, people, and risk — tread through all four potholes in a row, and your ROI naturally stays out of reach for a long while.

Some Companies Really Did Settle the Account. What Did They Do Right?

Don't swipe away just yet. Beyond the numbers, there is a group of winners, and their endings are completely different.

I have looked into this group, and what they do right, over and over, comes down to three things:

First, they teach "knowing how to use AI" as a real skill. Companies that trained their employees in AI achieved project deployment success rates a full 43 percentage points higher than those that barely trained at all. Tools are everywhere on the market; what actually separates the leaders is people.

Second, they dare to push deeper. Going by McKinsey's statistics, on the marketing and sales front, once AI is applied at "depth," average sales ROI runs 10% to 20% higher than peers'.

Third, they are willing to keep a three-year book. BCG pulled out the companies that consistently stayed ahead and compared them: over three years, their revenue growth came out at 1.5 times that of peers, and return on capital at 1.4 times.

The closest thing to a formula is still BCG's 10-20-70 principle:

10% of resources go to algorithms; 20% go to technology and data; the remaining 70% go to "people" and "processes."

Read the ratio carefully, and do not mix it up. Mix it up, and you lose half the money before you even start out.

Where Do You Start? First, Get Clear on What "AI Marketing" Means

So do not rush to ask "should we even go with AI?" first. The real question is "how."

What is AI marketing? It means letting an AI model serve as your advisor — running the marketing strategy and hunting for users. In a mountain of data, it can dig out patterns that people cannot see, and then tell you in plain language: who is most likely to buy.

To make money from it, McKinsey says, pick these three most pragmatic tasks first:

1. Lead identification. Pull the people most likely to place an order out of the crowd first;
2. Marketing optimization. Let the system decide what time and which channel, instead of you making a snap call;
3. Personalized outreach. Let the bot chat one-on-one with every user on your behalf.

Then run the ROI math again. The companies that do well basically all get three things done: on strategy, they write one sentence that clearly states what AI is here for; on budget, they raise the share of the digital budget going to core AI technology above 20%; on the team, besides buying tools, they staff data scientists and strategists who understand the models.

When Choosing an AI Tool, I Keep My Eyes on One Thing: Whether It Can Settle the Account

We finally reach the last step: how to pick the tool.

Do not let the parameters dazzle you. Fix on one thing first: explainability.

What does that mean? When the tool reaches a conclusion, can it tell you "on what grounds I judged so"? Some AI is like a black box — whatever it says goes, and all you can do is trust it. The better type is a glass box: it opens up and shows you the basis of the judgment.

Iterable, which builds for marketers, takes the latter path. It embeds explainable AI into two modules of AI Suite — Brand Affinity and Predictive Goals — so it can both grasp what customers like and explain clearly where a judgment came from. What it gives you is a ledger you can open up and inspect, not a "trust me."

In Closing

Back to the friend from the opening.

The last thing I told him: "Wanting to get all the money back within three months is not realistic. But fill in the data potholes, build up your employees' AI skills, put seven tenths of your energy into people, and pick an AI tool that can explain its reasoning. A year or two from now, invite me to lunch again — and bring that ROI ledger with you."

He smiled: "In that case, this meal is on me."

I said: "When that day comes, Mr. Ledger, you will be treating me to something expensive."

Here's to you, the one who settles accounts a little late: may what you finally calculate be confidence, not anxiety.