Five barriers that prevent businesses from earning from AI

While some business leaders proudly report double-digit efficiency gains thanks to artificial intelligence, others invest millions – and see no tangible results. The gap between those who win and those who fall behind is determined not so much by the choice of models or the size of the budget, but by whether management is willing to answer five uncomfortable questions honestly.

The first question is this: do your employees truly know how to work with AI, or are they only creating the illusion of engagement? The shortage of skills is now described as the main тормоз to AI adoption worldwide. At the same time, companies often mix up two very different categories of competence.

On the one hand, there is a lack of technical specialists. Data engineers, machine learning experts, and cloud solution architects remain in clear deficit. The WTW report on digital talent for 2025 notes that in the European financial sector, data analysts, visualization specialists, predictive analytics experts, and data storytelling professionals are particularly in demand.

But technical talent is only part of the picture. As Cisco CHRO Kelly Jones rightly noted, soft skills are becoming the new hard skills. The World Economic Forum, in its “Future of Jobs – 2025” report, highlighted the key competencies of the AI era: analytical thinking, resilience, empathetic leadership, creativity, and self-awareness. If a company develops only engineers but does not change its managerial culture, it risks placing a powerful tool in the hands of people who do not understand how or why to use it.

The second question concerns data: are they your asset or your weakness? Many organizations mistakenly assume that data is something self-evident. Yet the quality and accessibility of information become a critical obstacle. According to the PEX report for 2025-2026, 52% of companies name data as the main barrier to effective AI.

The problem is rarely that there is too little data. More often there is too much, but it is fragmented, poorly structured, and controlled by different departments. Add regulatory requirements, cybersecurity concerns, and ownership issues – and you get digital chaos in which even the most advanced model becomes useless.

Data is sometimes called the new oil, but it is more accurate to compare it to a garden: to get a harvest, you need constant care, cleaning, and organization.

The third question is: are you truly investing in AI, or are you simply shifting expenses around? Goldman Sachs analysts predict that global spending on artificial intelligence will reach 2 trillion dollars as early as 2026, with AI services accounting for around 325 billion.

However, big numbers alone do not guarantee success. In 2025, many companies reduced staff, explaining it as automation. But often it turned out that it was not about replacing people with algorithms, but about redirecting saved payroll funds into financing AI projects. This may look tactically reasonable, but strategically it does not create new value.

Companies that achieve real ROI treat AI as a long-term investment portfolio over a three-to-five-year horizon, managing it as seriously as they once managed innovation or major deals.

The fourth question is unexpectedly practical: where will you get the electricity? Conversations about AI often sound like purely digital transformation, but in reality everything depends on physical infrastructure. Goldman Sachs forecasts that by 2030, data center electricity consumption will grow by 175% compared to 2023.

Training and running large language models require enormous computing power operating around the clock. Where energy is expensive or unstable, AI development becomes objectively limited. WTW emphasizes that data centers and energy resources should be viewed as part of a company’s critical infrastructure, not as a secondary technical issue.

The fifth question is the most difficult: did you embed AI into old processes, or did you rethink the processes themselves? Most failures happen именно here. Many companies take existing workflows and simply attach a neural network to them. The outcome is almost always the same: expensive, complex, and with no meaningful effect.

Successful leaders approached it differently. They asked not how AI could help do familiar things faster, but what processes should be rebuilt entirely in light of new capabilities. This requires revisiting roles, abandoning old metrics, and sometimes temporarily reducing efficiency in exchange for future advantage.

The strengths of machines are obvious: automating routine tasks, detecting patterns in large datasets, forecasting. The strengths of humans are context, ethics, empathy, and the ability to act under uncertainty. The greatest impact is achieved where these capabilities are combined rather than replacing one another.

The five barriers – skills, data, capital, energy, and processes – have been known for a long time. But the key point is that companies that overcome all five simultaneously and systematically begin to pull ahead at such speed that catching up in a few years will be nearly impossible. The question is no longer whether AI is needed. The question is whether businesses have time to delay the decision.

From the perspective of machine analysis, these barriers are not equally significant. The history of technological revolutions shows that infrastructure constraints – especially energy – often become the main bottleneck. This was true during industrial electrification and during the rise of the internet. Back then, the problems also seemed insurmountable until the market found unexpected solutions.

Interestingly, the energy factor is the least responsive to managerial effort: you can hire specialists and redesign processes, but you cannot accelerate the construction of power plants through corporate will alone.

The data statistics also reveal многое. If 52% of companies name data quality as the key obstacle, it means that almost half have either solved the problem or are simply underestimating it. The second option appears more likely. And the question of how honest corporate self-assessment is in the AI readiness debate remains open.