Want to Make Money From AI? First, Figure Out How to Grow

A while back, I had dinner with some friends who run companies.

Over the meal, one of them pulled out his phone and showed me his company's AI pilot list for the year: a customer-service bot, document summarization, a coding assistant, sales forecasting… a whole screen, densely packed. He was pretty pleased with himself.

I asked him one question: "Which of these actually made you more money?"

He was stumped.

This, in fact, is the awkward moment almost every company has been living through for the past two years. In the meeting room, everyone nods at a handsome AI-pilot spreadsheet. Then someone asks: Did revenue go up? Did costs come down? Are decisions faster and sharper?

Silence.

PwC did something fairly ruthless. They found 1,217 companies around the world — mostly $1-billion-plus in annual revenue, director-level executives — and asked them, one by one: How much is your AI actually producing? The survey was run in October and November 2025, across 25 industries and six continents.

The results sting.

A Set of Numbers That Makes You Sit Up

PwC found that the real money AI brings in is highly concentrated.

20% of the companies captured 74% of the AI upside.

And the remaining 80%? They split that 26%.

The harder numbers were yet to come. The companies with the "best AI fitness" — PwC calls them AI leaders — saw financial returns from AI that were 7.2 times those of other companies.

7.2 times.

Not 7.2%. 7.2 times. Same world, same wave of AI, and the gap can stretch this wide.

My first reaction after reading this report was: Why them?

What Does "Good AI Fitness" Mean?

PwC coined a term: AI fitness （PwC's term）.

What does fitness mean, in plain terms? It's this: can you wield the AI blade where it most needs to cut; is your foundation solid; has AI truly grown into your business processes?

They broke "fitness" down into 9 factors. Six are foundational: strategy, investment, data & technology, talent, governance, and innovation. Three are about usage: how widely and deeply AI is deployed, how advanced the use is, and whether it's aimed at new industry-convergence opportunities.

Let me tell you why this set of things is worth 7.2 times.

Start With the Foundation. It's the Converter That Turns "Activity" Into "Outcomes"

Many companies are stuck like this: AI projects launch one after another, but each is a one-off. When it's done, the experience can't be carried forward, the components can't be reused, and the data still lives in separate silos. The next project starts from zero again.

It's as if every time you cook, you have to rebuild the kitchen from scratch.

Hidden in PwC's data is a finding that stuck with me: when a company builds a solid foundation, every additional unit of AI usage it adds brings in roughly 2 times the return of a company with a weak foundation.

The foundation, at its core, does one thing. It lifts AI's conversion rate.

Good data and platforms let AI deploy faster; redesigned processes and trusted employees mean more people use it; the more it's used, the richer the data, and the smarter the system gets. This is a compounding cycle.

And the reverse? A rotten foundation means every deployment reinvents the wheel.

So where exactly is the AI leaders' foundation stronger? A few points in PwC's report really hit me.

First, they dare to spend — and they're willing to move money. These companies invest in AI at, on average, 2.5 times the rate of others. Software, banking, and entertainment are the most aggressive, putting in around 5% of annual revenue. But spending alone isn't enough — they're also better at "rebalancing": the moment business priorities shift, people and money immediately flow to higher-value AI projects. They're 1.3 times more likely to do this kind of dynamic reallocation. In plain terms, they manage AI projects the way investors manage a portfolio.

Second, they give innovation a "sandbox" and an "owner." Don't let AI become everybody's side project. Everybody's side project eventually becomes nobody's main job. AI leaders deliberately set up experimental environments (sandboxes) so people can try things freely, and they assign an innovation owner inside every business unit. After the experiments? Regular reviews decide what to scale and what to cut. They're 1.5 times more likely than other companies to have this mechanism in place.

Third, they take "trust" seriously — seriously enough to be almost counterintuitive.

I need to say a bit more on this one.

Trust Isn't a Soft Metric. It's a Throughput Bottleneck

Think about it: AI's value only gets realized when someone actually uses it.

If employees don't trust it, they won't dare use it; if they won't use it, the system sits idle; if the system sits idle, data stops flowing, and the whole compounding cycle seizes up.

PwC puts it bluntly: insufficient trust is a throughput constraint. Not some soft "change-management" line item.

The data backs this up. Employees at AI leaders are 2.1 times more likely to trust AI insights and actually use them at work than employees at other companies.

How do you build trust? It takes a system.

Get the business, data, and AI sides co-creating together from the design stage. Don't build something and then toss it over the wall to users — that kind of handoff is the easiest way to kill adoption. Give employees clear incentives to experiment. Offer continuous, role-specific training, and have executives themselves lead by taking the classes and using the tools. Then add clear guardrails: what AI can do, what must escalate to a human, who's responsible. When the boundaries are clear, people dare to let go.

The Wyndham story (yes, the global hotel franchise group) is especially telling.

It used to take them, on average, 30 days of manual work to change a single brand standard. PwC helped them deploy AI agents with human oversight — automated prompts, human-machine co-creation, real-time monitoring — with framework and training rolled out together. The result? The review time for brand-standard changes dropped 94%; AI review ran 20 times faster than humans, saving 40 to 80 hours each time.

The key is that employees actually felt safe using it. Because the guardrails were there, the training was there, and the human was in the loop.

Aim AI at "Growing," Not Just at "Saving Effort"

That's the foundation. Now let me get to the point in this report that most overturned my assumptions.

What do most companies use AI for? Efficiency. Insurance companies use it to speed up claims processing; software firms use it to write code. They do what they're already doing — just faster and cheaper.

AI leaders also pursue efficiency, but they don't stop there.

They treat AI as a "growth engine" and a "reinvention engine," not just an "efficiency tool."

There's a line in the report I read three times over: Of all the AI fitness factors, the one with the biggest impact on financial return is "seizing the growth opportunities created by industry convergence." Not efficiency. Not data. Not talent. Growth.

What does industry convergence mean?

An example. An automaker and a healthcare company team up: health sensors in the car monitor the driver's body data, feed it to AI, and then tailor a preventive care plan for you. One sells cars, one treats patients, and in between, a brand-new business grows up.

AI leaders are 2 to 3 times more likely to engage in this kind of "cross-industry collaboration and cross-ecosystem competition." They're 2.6 times more likely to use AI to reinvent their own business models.

The John Deere story (yes, the tractor maker) is a textbook case.

Their See & Spray system puts cameras and on-board computing on sprayers to identify weeds and spray only where it should be sprayed. During the 2024 planting season, this was used on more than a million acres, saving farmers roughly 8 million gallons of herbicide — an average 59% reduction across corn, soybean, and cotton fields.

But the truly clever part is that John Deere packaged this capability as a "pay-for-outcome" service. They're not selling you a more expensive machine; they're selling you the result of "how much chemical you saved." A one-time hardware margin turned into a sustainable stream of service revenue.

This is aiming AI at "growing," not just at "saving effort."

The Final Blow: Let AI Run Itself, and Roll It Out Across the Board

There's another set of data in the report that's pretty interesting.

The companies getting the strongest financial returns from AI are nearly twice as likely as others to have genuinely rolled AI out across the main links of their value chain. Not just piloting in one or two departments — but deploying from strategy to supply chain, from back office to front office.

And they use it more aggressively.

What does that mean? PwC ranked AI maturity into several tiers. The lowest is summarization and drafting; up a level is analysis, prediction, and recommendations; higher still is executing a routine task for you (but a human approves); above that is running multiple tasks autonomously inside guardrails; at the top, it's fully autonomous and even self-optimizing.

AI leaders are roughly twice as likely as others to sit in the top two tiers. They're 2.8 times more likely to let AI make decisions without human intervention.

At this point, you might get nervous. Is this about replacing people?

The report offers a measured, and very real, judgment: full autonomy is still the exception for now. Only 15% of AI leaders say their most mature AI use cases are fully autonomous. And while 48% of leaders expect AI to reduce headcount by at least 5%, an equal 49% expect headcount to stay about the same or even grow.

What's really changing isn't "driving people out" — it's "driving out latency."

AI, inside guardrails, handles the repeatable judgment calls; people handle the exceptions, make the trade-offs, and steer decisions toward strategic goals.

The Lucid story (yes, the EV maker) is compelling. They started in the finance department. PwC helped them use operational data, AI models, and agent tools to AI-ify forecasting, reconciliation, analysis, and monitoring. The result: the end-to-end forecasting cycle went from weeks down to under a minute. Within ten weeks, they designed and began rolling out 14 AI-driven use cases, and they're now spreading from finance into procurement and operations — plus an AI assistant for executives that shows the movement of more than $1 billion in investments in real time.

Finance first, draw the blueprint, then copy it across the whole company.

This Is a Compounding Game. You Can't Afford to Wait

Reading through this report, my strongest feeling is this: AI is turning into a brutal compounding game.

The frontrunners have richer data, smoother processes, higher trust, and faster deployment — every new AI use case they launch returns more than anyone else's. Then they learn faster, copy solutions faster, and dare more to let AI run on its own.

The faster they run, the wider the gap grows.

PwC says it themselves: the companies that want to catch up really can't wait any longer.

How do you catch up? The report keeps coming back to one thing: stop counting how many AI pilots you have.

Pick one or two of your most valuable business goals and aim AI at them. Fix the one or two foundational shortcomings that keep blocking your repeatable rollouts. Give key projects clear owners and metrics, and run a monthly "scale or stop" review. Only the projects showing real movement on a real business metric keep getting funded.

Treat "industry-convergence growth" as its own AI portfolio — with executive backing, a roadmap, and hard choices.

Shift from a "save money" mindset to a "make money" mindset.

This playbook isn't, in itself, that hard. What's hard is whether you're willing to step out of the comfort zone of "the pilot list looks great."

The day after that dinner, my friend messaged me: he'd cut that pilot list by two-thirds. Behind every remaining item, he'd written a number — revenue, cost, or a specific decision-quality metric.

He said: "Only after I cut it did I realize — I had no idea why I was doing most of those before."

That's the first step from "activity" to "outcomes."

Take that step, and only then does the 7.2x gap start to have a chance of closing.