Three Levels of AI Transformation: The Path to Measurable Payoff

The race is on. Every leadership team is trying to figure out how to bring artificial intelligence into the heart of their business. Companies are greenlighting multimillion-dollar technology overhauls that, in the past, would have been deferred for years. They’re rethinking operating models to make room for agentic workflows, not just tinkering at the edges. And they’re accelerating initiatives that would otherwise be considered long-term bets.

Bank of America has announced roughly four billion dollars in annual AI and technology investment, embedding intelligent systems across global operations and training. Procter & Gamble is scaling its end-to-end “AI Factory,” tying AI into product innovation, manufacturing, and supply chain planning, with projected savings in the hundreds of millions. Walmart has rebuilt major portions of its supply chain around real-time AI systems, accelerating a years-long modernization effort into the present and reshaping how goods move through its vast retail network.

Their urgency is understandable. AI is the kind of general-purpose capability that will reshape competition. Many leaders still have memories of what happened when they moved too slowly through a previous technology shift. When the internet arrived, a surprising number of execs dismissed it as a fad, a toy, or at best a novelty for nerds. When e-commerce followed, some brick-and-mortar operators waited too long to revamp their payment and fulfillment systems. As a result, what passed for an online presence amounted to little more than digital brochures. Mobile technology and the move to omnichannel experiences forced companies to pay attention sooner, though many still struggled to know where to begin. This time, no one wants to get left behind.

An Uncertain Payoff

Make no mistake, though. The investment required is massive, and the return is sometimes unclear. A recent MIT study found that early AI initiatives delivered little measurable payoff. The report noted that, despite billions invested in enterprise AI, roughly ninety-five percent of generative-AI pilots failed to produce measurable business value.

The latest Wharton–GBK study paints a rosier picture. Surveying large U.S. enterprises over a three-year period, researchers found that more than seventy percent now formally track AI’s contribution to output and profitability. Those firms report that generative AI has produced measurable financial returns, with steady gains in productivity, speed, and throughput.

The contrast in payback isn’t a contradiction—it’s a reflection of varying states of AI maturity. If you wanted to measure the benefits of running, you wouldn’t lump Kenyan Olympians, weekend club runners, and couch potatoes like me in the same study. Our inputs are different, and so are the results.

What we’re seeing is that AI transformation demands shifts in culture, process, and strategy—not just technology. Culture is about helping people build the intuition, trust, and confidence needed to work alongside intelligent tools. Process is the rewiring of workflows, so AI doesn’t sit on top of old ways of doing things. Strategy is the work of determining where AI can change the trajectory of the business. Technology is the foundation that makes everything else possible, from clean data to the infrastructure that lets systems learn in real time. Given those four factors, companies fall into three stages of AI transformation maturity.

Stage One: Screwing Around

The earliest stage is exactly what it sounds like: experimentation without expectation. People are trying a bunch of things. Execs are playing around with ChatGPT. Marketing teams are generating videos with Gemini 3 Pro to get people excited at their next all-hands meeting. Customer-service teams are putting up chatbots on their sites.

All of this is healthy. It builds intuition for what’s possible with the technology. It also helps teams learn when they can trust AI and when they can’t. Companies can help their people become better users of the tools through training and recommendations. For instance, there are a host of different AI apps to help transcribe your Zoom meetings. Which one should you use? Help teams pick the tools that are best for your organization.

Some organizations, unfortunately, are responding by trying to clamp things down. They create governance boards to prevent experimentation. They require vendors to certify that they won’t use AI anywhere in the workflow, as if that were even possible. It’s short-sighted.

But, here, the goal shouldn’t be results—it’s culture change. The output is greater literacy and a clearer sense of where value might reside. At this stage, it’s good to let a thousand flowers bloom. Just don’t expect a big harvest.

Stage Two: Use Cases

When the experimentation phase starts to feel too loose, leaders go hunting for something more tangible. They look for discrete “use cases” where AI might make things faster or cheaper or reduce human intervention.

That shift, from screwing around to solving problems, is the primary difference between companies that are seeing a payback or not. The MIT study describes a widening “GenAI Divide” between companies that dabble in front-office experiments and those that embed AI deeply into the processes where money is actually made or saved.

Most use cases are focused on efficiency. A property and casualty insurer like Allstate might use AI to parse the hundred-page “demand packet” that’s attached to an auto claim. A bank like Citi might pursue AI to strengthen regulatory compliance, especially after living through the pain of a consent decree. T-Mobile has used predictive models to identify which customers are likely to leave before they actually do.

There’s now a rush of firms trying to capture this work, each promising clarity in a moment defined mostly by confusion. Legacy management consultants are publishing surveys of what their clients are doing, hoping that passes for thought leadership. Large technology service companies are racing to stand up credible AI implementation teams. These firms are struggling to attract talent. They simply can’t match the compensation that Google, Microsoft, or Palantir offers top engineers. Those tech giants are showing up with possible use cases that a client can implement immediately, backed by talent most enterprises can only dream of attracting. That talent gap is why it’s so important to focus. You need to pick a few critical use cases where AI can actually move the needle and then commit enough resources to get results quickly.

Ironically, many of the most valuable use cases don’t need frontier models like Claude or GPT-5. T-Mobile’s customer retention challenge requires significant machine learning and data analytics, not generative AI. Generative tools are powerful, but they sit alongside predictive models, optimization, vision systems, and other forms of machine learning that often drive just as much value. But LLMs have captured enough executive imagination that budgets are suddenly available for the “boring” infrastructure required to make any of this work. The winners treat AI like a portfolio, not a single bet.

The biggest challenge in Stage Two is culture and process. Stage Two succeeds only when companies open up their core workflows, not just their edge experiments. Use cases often sit within the turf of a particular business unit leader, who’s perfectly happy to let the innovation team tinker around the edges. Just don’t touch anything vital. As long as that dynamic persists, the payoff will always be constrained. It’s like building a beautiful e-commerce site without reinventing your supply chain. Nothing meaningful happens until the organization rewires itself around the capability.

The companies that are seeing the greatest payoff are those that are opening up their core business processes to new thinking. Those companies are redesigning workflows so systems can learn from context and operate at scale. Less mature organizations are trying to bolt AI onto existing processes and hope for transformation. Without integration, alignment, and a willingness to rewire how work gets done, AI remains an expensive experiment rather than a source of meaningful change.

Stage Three: Fail Cases

At the third stage, companies are moving beyond optimization to reinvention. The most transformative applications of AI aren’t the ones that mimic existing processes or shave minutes off routine tasks. They’re the ones that tackle problems nobody is solving today, or that surface threats that didn’t exist a year ago. Stage Three is all about bringing technology to bear on two kinds of critical fail cases.

The first kind asks a blunt question: what are the three or four reasons your company might not exist in five or ten years? Then ask how AI might help prevent that outcome. The second kind flips the perspective: how might AI itself erase the competitive advantage you believe you have? Which parts of your business model might the technology make irrelevant?

One of OpenAI’s earliest partners was Morgan Stanley. CEO Andy Sapperstein saw a looming cliff: client demand was rising, but the supply of new wealth managers wasn’t—and a wave of advisor retirements was about to make the gap worse. Rather than accept that constraint, the firm turned to AI as a way to dramatically increase the number of clients each advisor could serve, shifting the question from how Morgan Stanley could keep up to how it could scale its expertise. Because, make no mistake, the cliff is coming.

Every company has those kinds of fail cases — the uncomfortable truths that rarely make it into the annual plan but shape the future all the same.

Consider that insurance company wrestling with hundred-page demand packets. Right now, AI can help the company to process those claims faster. But what if the job got a lot harder? Right now, accident attorneys are getting rolled up by private equity. Very soon, claimants may be sending in five thousand pages instead of a hundred, with the goal of overwhelming a claims department. If your systems aren’t ready for that, you’ll feel like you’ve brought a knife to a gunfight.

Or imagine a food company whose entire identity is built around effective brand marketing. What happens when buying decisions are mediated by agents and algorithms? The warm glow of your thirty-second spot may never reach the person who’s supposedly making the choice, because a person isn’t making the choice at all.

The challenge in Stage Three isn’t just process. It’s strategy. It’s the ability to envision how things might play out over the next five or ten years and come up with a plan to win. You can’t save Stage Three for later, because the futures that can break you don’t wait for your maturity timeline.

The Real Challenge is Leadership

Across all three stages, technology is necessary but not sufficient. Enterprises don’t transform because they buy models. They transform because they rebuild workflows, incentives, and decision systems. That way, learning can compound over time.

Of course, beneath all the challenges of strategy, culture, process, and technology lies something deeply human. That’s leadership. Most of us are present-focused, even when the future is barreling toward us. Our brains aren’t wired to imagine disruptive futures. We default to what’s familiar. We imagine that the world will be roughly similar to how it is today. Maybe that will be true. But let’s not bet on it.

The real work of leadership is expanding our own capacity to think beyond the present, and then guiding hundreds or thousands of people through a stretch of profound ambiguity. The hardest truth is that AI isn’t waiting for us to get comfortable. It’s moving faster than our instincts, our habits, and in many cases, our assumptions about how business works. The leaders who rise to this moment won’t be the ones who polish the old playbook. They’ll be the ones willing to question it. They’ll treat uncertainty as a place to lead from, not a problem to eliminate. They’ll push themselves to imagine futures their organizations can’t yet see—and prepare their teams to operate inside those futures before anyone else does.

That’s the real frontier—the moment when leadership stops being about protecting what worked yesterday and starts being about equipping people for what’s coming tomorrow. AI isn’t just transforming business. It’s transforming the kind of leader the future will require.