The AI Boom Is Producing a Very Expensive Illusion

May 11

Picture a familiar meeting. The chief executive asks for an update on AI. The technology leader lists the copilots now available to employees, the pilots underway in three departments, and the agents scheduled for the next quarter. A consultant presents a maturity model. Someone uses the phrase “enterprise transformation.” Then the finance chief asks a smaller, sharper question: What, exactly, has improved?

The room changes.

That moment is the black dot in the chart. It is the point when enthusiasm meets arithmetic, when a company discovers that it has purchased a great deal of artificial intelligence without building much organizational intelligence around it. This is the central mistake of the current AI boom. Companies are investing in the visible part of transformation while neglecting the part that makes transformation possible. Tools are visible. Licenses arrive. Demos sparkle. Press releases travel. Foundations are different. Cleaning data, rebuilding integrations, clarifying ownership, redesigning workflows, training employees, and establishing governance are slow, political, and largely invisible. One produces excitement. The other produces capability.

Guess which one gets funded first?

According to BCG’s 2026 AI Radar, companies expect AI spending to rise from roughly 0.8% of revenue in 2025 to about 1.7% in 2026. That is not a cautious experiment. It is a major reallocation of corporate capital. I am not opposed to the spending. I am opposed to pretending that spending itself is evidence of progress. A purchase order is not a strategy, and a portfolio of pilots is not a transformation.

The warning signs are already visible beneath the enthusiasm. Flexential’s 2025 infrastructure survey found that 44% of IT leaders viewed infrastructure constraints as the leading barrier to AI expansion. 59% reported bandwidth problems, while 53% reported latency challenges. Cloudera’s 2026 Data Readiness Index found that 96% of organizations said AI was integrated into core business processes, yet nearly four in five said limited access to data across environments was constraining their AI and data initiatives. Only 18% said their data was fully governed. NetApp’s IDC-backed study found that 84% of organizations said their storage was not fully optimized for AI. These are vendor-sponsored studies, so they deserve a raised eyebrow. But when different surveys keep revealing the same fracture, dismissing the pattern becomes its own form of denial.

Give AI clean data, a coherent process, clear decision rights, and accountable owners, and it can create astonishing leverage. Give it duplicated records, contradictory policies, five approval layers, and nobody responsible for the final result, and it will make the confusion faster, cheaper to reproduce, and harder to detect. A chatbot attached to a broken customer journey does not fix the journey. It merely gives the dysfunction a conversational interface.

This helps explain the strange gap between AI activity and AI value. McKinsey’s 2025 global survey found that 88% of respondents said their organizations regularly used AI in at least one business function. Yet nearly two-thirds said their organizations had not begun scaling AI across the enterprise, and only 39% reported any enterprise-level EBIT impact. The lesson is not that AI has failed. The lesson is that access, adoption, scale, and value are four different things, although many executives speak about them as if they were synonyms.

The better approach begins with a business problem, not an AI product. Choose a workflow with economic weight: claims processing, service resolution, proposal development, inventory planning, software testing, financial reconciliation. Measure its current cost, speed, error rate, and customer impact. Then ask where AI could alter the work itself, not merely decorate it.

I would go further…

The unit of AI transformation is not the model. It is the workflow.

I love that phrase because it captures a lesson I’ve had to learn the hard way over and over again: the impressive stuff rarely matters if it does not change the actual work. It is also a reminder to myself, because I can get excited about the technology too, but the part I care about most is whether people can finally do something better, faster, smarter, or with less frustration. If the workflow remains unchanged, the company is probably automating fragments and calling the result innovation. This does not require waiting three years for perfect data or flawless architecture. Perfection is often procrastination wearing a responsible-looking suit. Companies need a minimum viable foundation for each priority workflow: trusted data, sufficient infrastructure, secure integration, a named business owner, clear human oversight, employee training, adoption measures, and a date when leaders will decide to scale, revise, or stop. Build the foundation alongside the use case, then expand both as evidence accumulates.

Boards should therefore ask a question that is less glamorous than “What is our AI strategy?” but far more revealing: For every dollar committed to AI, how much supports the conditions required to make it work? If leaders cannot separate spending on tools from spending on data, infrastructure, process redesign, governance, and people, then the AI budget is not a strategy. It is a shopping receipt.

I am bullish on AI. I am bearish on magical thinking. The companies that win will not be those that buy the most tools or announce the most pilots. They will be those willing to do the difficult, unphotogenic work that turns technical possibility into organizational capability. AI can transform a business. But first, the business has to become transformable.