If AI were delivering even a fraction of what companies promised their boards, we would not be having this conversation. Yet study after study shows the same uncomfortable truth that most organizations are pouring millions into AI and getting little or nothing back.
In just a few years, AI has gone from experimental curiosity to corporate fixation. Nearly every senior leader now claims their AI investments are generating positive ROI. Confidence is sky-high. AI sits at the top of CEO agendas, even as governance, alignment, and execution continue to lag. The narrative of widespread success is compelling, but it is also fragile, because when you look past the rhetoric, the data tells a different story.
Deloitte's 2025 analysis reveals that only 6% of organizations see AI deliver ROI within a year, while most projects take two to four years to break even, if they ever do. That gap between expectation and reality is quickly becoming a career risk. Boards are no longer asking whether AI is strategic. They are asking why the returns are missing and who is accountable.
The problem is that AI works, but our approach doesn't. In most organizations, AI is treated as a centralized, technical initiative. It lives in IT. It is managed by a small group of specialists. It is rolled out through long, expensive programs designed to "transform the enterprise." Frontline employees, the people who actually sell, operate, analyze, and serve customers, are largely spectators. That approach almost guarantees disappointment.
I see another dynamic at play as well: anxiety. People are perfectly comfortable using generative AI at home. They will ask the AI questions, plan trips, or explore ideas without hesitation. At work, that confidence evaporates. Employees worry about policies, compliance, making mistakes, or "doing AI wrong." So, they step back and wait for permission. Or worse, they wait for IT to solve everything for them.
The result is predictable. Projects balloon in scope. Timelines stretch to nine or 12 months. Budgets hit seven figures. And when the promised savings don't materialize, the initiative is labeled a failure. What makes this particularly frustrating is that, for the most part, the value of AI does not require moonshots.
One reason organizations overspend is that they lump all AI into a single, mysterious category. It helps to separate it into three distinct flavors.
First is machine learning. This is pattern recognition based on historical data, and by that I mean predictive maintenance, demand forecasting, and anomaly detection. If you already have data, these solutions can often be deployed in weeks, not years, for hundreds or thousands of dollars.
Second is generative AI or natural language processing. These tools work with text, ideas, images, and audio. They summarize documents, draft content, analyze proposals, generate reports, and automate knowledge work. Many of the highest-impact use cases I see cost less than a hundred dollars a month and save hours per employee every week.
Third is agentic AI. This is where systems plan and act autonomously. It is powerful, complex, and expensive. It requires technical expertise, careful rules, and real investment. This is the domain of IT and advanced teams.
The mistake is starting with the third flavor and ignoring the first two. Machine learning and generative AI belong in the hands of everyday employees. These are low-cost, low-risk tools with immediate returns. Saving even six or eight hours a week for a salesperson, analyst, or operations manager is real money. Time saved becomes time reinvested in customers, revenue, and better decisions. Multiply that across a workforce, and the ROI becomes obvious.
Agentic AI, meanwhile, should remain the playground for technical teams pursuing high-risk, high-reward innovation. Let IT chase the five percent opportunities. But do not bet the company on them.
In my time at Process Innovation, Inc. (PPI), I have watched organizations waste staggering amounts of money by assuming they need the Cadillac version of AI when a Schwinn would do just fine. Leaders chase massive savings targets, declare victory only if millions appear on the balance sheet, and overlook hundreds of small wins that compound into real advantage.
There is a counterargument, of course. Some believe that decentralizing AI is risky, that governance, security, and consistency demand tight control. Those concerns are valid. But control does not have to mean paralysis. Paid, secure tools trained on internal data are far safer and more effective than pretending employees won't experiment anyway.
The bigger risk is doing nothing or doing everything at once. Right now, many companies are effectively gambling on their slim chance of success. They are buying Powerball tickets and calling it a strategy. A far smarter approach is to fund dozens or hundreds of small experiments, each tied to a real business problem, each with clear ownership and fast feedback.
If you want a practical starting point, give every employee a modest AI budget and a simple mandate: identify one task you do regularly and use AI to make it better. Support experimentation, reward results, and measure time saved and decisions improved, not just dollars spent.
The money being wasted on AI today is not inevitable. It is the consequence of poor framing, misplaced control, and unrealistic expectations. The organizations that fix this now, by decentralizing AI, empowering their people, and focusing on steady, measurable wins, will be the ones leading their industries tomorrow. The rest will be left explaining to their boards why billions were spent for pennies in return.
About the Author
Raymond Sheen is President of Product and Process Innovation, Inc. (PPI), a training, consulting, and advisory firm that helps organizations translate strategy into execution through digital transformation, project management, and operational improvement. With over three decades of experience spanning engineering, manufacturing, systems development, and corporate consulting, Sheen has worked with companies across industries to improve efficiency, reduce costs, and deliver measurable business results. At PPI, he leads a consortium of seasoned industry professionals who design practical, customized approaches to complex business challenges, eschewing one-size-fits-all solutions in favor of methods that reflect how organizations actually operate.
