You've Invested in So Many AI Projects — Why Aren't You Making Money?

A while ago, I came across a number that made me stop and stare.

PwC surveyed 1,217 companies across 25 industries worldwide. The finding: 20% of those companies captured 74% of all AI returns.

What does that mean?

Out of 100 companies doing AI, the top 20 made nearly three times what the remaining 80 made combined.

You might wonder: so what are those other 80 doing?

The answer: they're in meetings.

A Scene You've Probably Lived Through

From New York to Singapore, the same scene plays out in the conference rooms of nearly every big company.

Someone pulls up a slick slide deck, neatly listing the AI pilots: a chatbot here, a decision engine there. Everyone nods along.

Then someone asks: which of these grew revenue? Which one cut costs? Which decisions got faster, or better?

Silence.

Behind that silence hides an uncomfortable truth: plenty of companies are "doing AI," but very few are "making money from AI."

PwC calls that 20% who actually make money the "AI leaders." What did they get right?

That's what I want to dig into today.

7.2x

Let's start with a number that made me do a double take.

PwC scored every company's "AI financial performance." The method isn't complicated: add up the revenue growth and efficiency gains AI delivered, then compare against the industry median.

The result? The companies with the strongest AI capabilities scored 7.2 times higher than everyone else.

7.2x.

Think about it. If your competitors make a dollar a year from AI, that top tier makes seven twenty. And that gap keeps widening.

What's their edge?

PwC's answer: these companies have built a kind of "AI fitness."

What's AI fitness?

It's a company's ability to aim AI at what actually matters, lay the right foundation, and then thread AI into every capillary of the organization.

Sounds simple. In practice, it's anything but.

Stop Staring at Cost-Cutting — Stare at Growth

When most people talk about AI, they think about saving money. Insurance companies use AI to speed up claims. Software companies have AI write code. Efficiency goes up, sure — but that's only one slice of AI's value.

AI leaders use AI to save money too. They just don't stop there.

They treat AI as an engine to reinvent the business.

PwC's data shows these companies are 2.6 times more likely than others to use AI to reshape their business models.

2.6x. That number matters.

Even more interesting: PwC found that one variable drives AI financial returns more than any other. They call it "growth capture in industry convergence."

In plain English: AI helps you step outside your own industry and make money on someone else's turf.

For example. Picture an automaker teaming up with a healthcare provider to pack a car with sensors that monitor the driver's health, with AI analyzing the data in real time and tailoring a prevention plan. The car is no longer just a car — it's a mobile health-management terminal.

That's the new value pool that industry convergence creates.

John Deere is doing something similar. The company famous for building tractors built an AI system called See & Spray, mounted on the booms of sprayers, with cameras that recognize weeds and hit only the right spots with herbicide.

During the 2024 planting season, the system ran across more than a million acres, saving farmers roughly 8 million gallons of herbicide — cutting average herbicide use for corn, soybeans, and cotton by 59%.

But the really smart move: John Deere didn't pitch See & Spray as a one-off hardware selling point. They turned it into a service. Farmers pay for "verified outcomes" — which means the business shifted from selling sheet metal to selling a service, from one-off revenue to recurring revenue.

The biggest returns from AI come from changing what you sell and how you create value. Doing what you already do, only faster, is just the appetizer.

Run AI with the Discipline of an Investor

Okay — so you want to use AI for growth, for reinvention. What makes you think you can pull it off?

PwC's research is clear: you have to build the foundation.

But that doesn't mean spending two years on "full digital transformation" first. AI leaders build foundations in a sharply focused way — only the parts that support high-value AI use cases.

That foundation has six parts: strategy, investment, data and technology, talent, governance and risk, and innovation. Let me pick a few of the most counterintuitive ones.

First, run your AI budget like an investor.

AI leaders spend a share of revenue on AI that's 2.5 times higher than other companies. Software, banking, and media are the most aggressive — plowing about 5% of revenue into AI.

But money alone isn't enough. They're also more willing than other companies to kill projects. When strategic priorities shift, they're 1.3 times more likely to pull people and money out of low-value efforts and redeploy them.

It's the same logic as investing: buy the right stock, and you double down when you should; realize you bought the wrong one, and you cut your losses. The worst thing you can do is leave your portfolio sitting there untouched.

PwC offers a very practical suggestion: run a "scale or kill" review every month. Only projects showing real movement on well-defined business metrics get more money.

Second, don't let AI live as "everyone's side gig."

The moment AI becomes "an extra task for everyone," it's basically dead.

PwC found that AI leaders combine two practices with particularly strong results: providing dedicated experimentation infrastructure (like a sandbox isolated from the production environment) and embedding full-time innovation owners inside business units. Companies doing both are 1.5 times more likely than others to pull it off.

Put simply: give people, give them room to work, give them ownership.

Southwest Airlines cashed in on exactly this. Their flight-attendant attendance system ran on a creaky old tech stack, with documentation full of holes, kept alive by the tacit knowledge inside veteran employees' heads.

Working with PwC, they used generative AI to reverse-engineer the source code, extract functional requirements, and turn them into a clear to-do list. A backlog that used to take ten weeks to compile was done in five — saving over 200 engineering and business hours. They produced 600+ requirements, 90% of which were rated high quality.

The significance goes beyond time saved. It proved a repeatable, modern method. The next time they hit a legacy system like this, they can run the same playbook.

Third, if employees don't trust AI, AI is decoration.

A lot of managers treat "employees not using AI" as a small change-management problem. Dead wrong.

PwC says it bluntly: trust isn't a soft topic — it's a throughput bottleneck. Employees don't trust it, they don't use it; they don't use it, there's no impact; no impact, and the money you poured in goes down the drain.

Employees at AI leaders are 2.1 times more likely than employees elsewhere to trust and act on AI insight.

How is trust built? Not with one move, but with a system:

Have business, data, and AI people co-create the solution. Don't let developers build in a vacuum and toss it over the wall to users — that handoff is what kills adoption. Give employees continuous, role-specific AI training. Leaders learn and use it themselves, visibly. Draw clear red lines so everyone knows what AI can do, what it can't, and when a human has to step in.

Wyndham's approach is worth a look. This global hotel franchisor used to spend an average of 30 days of manual work to change a single brand standard.

Working with PwC, they designed an agentic workflow with human oversight: automated prompts, collaborative editing, real-time monitoring — the team could always guide and review AI's output. They paired it with a Responsible AI framework and ongoing training.

The result? Review time for brand-standard changes dropped 94%. AI review ran 20 times faster than manual review, saving 40 to 80 hours per review.

What Wyndham got right wasn't adopting AI — it was making employees unafraid to use it, fluent in using it, and willing to use it.

Thread AI Into Every Capillary

Everything so far has been about aiming and laying the groundwork. The last step is pushing AI into every corner of the company.

AI leaders are twice as likely as other companies to embed AI into their core business processes.

How? Three directions, all at once.

Spread horizontally. At most companies, AI is still crammed into a handful of departments. AI leaders push proven use cases out to every team, region, function, and product line. For instance, once you've shown AI can speed up invoice processing in finance, the same document-processing and workflow model can roll into legal for contract review or into operations for claims handling — at near-zero marginal cost.

Lucid Motors did exactly this. They started in finance, working with PwC to fast-prototype AI-driven forecasting and reporting. Within ten weeks they'd designed and started rolling out 14 AI use cases. End-to-end forecasting cycles compressed from weeks to under a minute.

Then the method spread: procurement, operations, and even an AI executive assistant for leadership that could surface a real-time view of more than $1 billion in investments.

Go deep vertically. Don't just slap an AI button on an existing process — bake AI into the bones of standard operating procedure. Customer service is the easiest example to grasp: AI runs right inside the ticketing system, pulls customer context and knowledge-base content automatically, drafts replies, and only escalates the complex cases to humans. Instead of a chatbot hanging off to the side while agents copy-paste back and forth.

Move toward autonomy. PwC measured a pile of operational metrics and found that "letting AI make decisions on its own" correlated most strongly with AI financial returns. AI leaders are 2.8 times more likely than others to push decision autonomy to AI.

But that doesn't mean "machines replacing people." PwC's data is nuanced: 48% of AI leaders expect AI to reduce headcount by more than 5% — but 49% expect little change or even growth.

What usually happens is "de-latency, not de-headcount." AI handles large volumes of repeatable judgment calls inside guardrails, while humans manage exceptions, make trade-offs, and steer strategy.

Where do you start? Pick a handful of high-frequency, repeatable, measurable, low-to-medium-risk decisions. Classification, ranking, routing. Automate within clear boundaries, monitor decision quality, and only expand once you hit the bar.

A Few Last Words

PwC's research left me with one clear takeaway: AI isn't a question of "is it useful." It's a question of "do you have the fitness to use it well."

Why do those 20% of companies take home 74% of the returns?

Not because they bought more expensive models or hired better engineers. It's because they made the right string of management choices: aiming AI at growth and reinvention, building only the foundation that supports high-value scenarios, and pushing AI deep into business processes. Three things — none of them sexy, all of them widening the gap.

What's more sobering: they're making these choices right now, and getting faster at it. Every use case they run gives them thicker data, faster feedback, smoother deployment. The longer you wait, the bigger the gap.

If you're still counting up how many AI pilots you have,

I suggest you stop.

Pick one scenario that can genuinely move the growth needle, and build it solid — foundation, process, people — in one go. Then copy it. And copy it again.

The companies that won't catch up from a 7.2x gap probably aren't losing on technology. They're behind because they never stopped piloting — and never ran anything for real.