The five pitfalls of artificial intelligence for Italian SMEs

The third critical area concerns data, the essential fuel for AI. Even before considering which tool to adopt, it is necessary to come to terms with a reality that many SMEs tend to underestimate: data is essential for systems to function, but is often lacking or unusable. Incomplete master data, inconsistent documents, information scattered across systems that do not communicate with one another, and know-how that exists only in people’s minds and has never been transferred to a shared repository. A wealth of information that exists on paper but which, in practice, cannot fuel any form of intelligence, be it artificial or human. Investing in an advanced AI model without first organising this wealth of information is a costly mistake: it is like fitting a Formula 1 engine onto a toy car body. The power is there, but the system cannot cope. Mapping out what you have, in what format it is held and who can access it is a less glamorous task than a tech demo, but it is essential for building a functioning and reliable infrastructure. Without this step, technical debt accumulates before you’ve even started, and you’re making an investment doomed to fail.

Another key issue concerns control. Entrusting AI with the development of code or the automation of operational processes without structured supervision exposes the company to a risk that is often invisible until it becomes an emergency: technological lock-in. When critical workflows are built and managed by automated systems, without anyone in the company understanding the underlying logic, a sort of ‘alien code’ is created: it works – until it stops, and nobody knows how to fix it. This happens when human developers have been reduced in number or eliminated to cut costs, resulting in blind reliance on technology. Who will get to grips with the company’s cybernetic heart when something breaks? Who knows how and why that specific change was made? Maintaining a degree of human involvement in automated processes – what is known in English as human-in-the-loop – is a form of oversight and safeguards that no company should shirk.

The fifth dimension of risk is perhaps the one most overlooked in public debate, and it concerns people, particularly younger workers. Replacing junior staff with AI is one of the choices that appears to be the most efficient in the short term, but it is also one of the most counterproductive in the long term. The artificial intelligence models available today have been trained on skills, processes and know-how accumulated by generations of scholars and professionals who came before them. By disrupting the chain of practical training, we must ask ourselves who tomorrow’s senior staff will be. Meanwhile, the world will be changing, so on which skills will the AI of the future be trained? If ‘world models’ come into play, from whom and how will they learn? A company that stops training also stops knowing, because it loses the ability to critically evaluate the outputs of the tools it uses; it cannot break down the results to make corrections; it erodes organisational memory and, consequently, its culture.

We might consider ‘enhancing’ junior staff’s profiles with AI, but here too there is a major point to bear in mind. In the absence of strong, structured domain expertise, the hallucinations and contextual errors typical of LLMs go unnoticed. Furthermore, young people might be tempted to perfect their work at an early stage by delegating tasks to AI, thereby shielding themselves from negative feedback, particularly in contexts where psychological safety is entirely lacking. Finally, we must consider the biological damage to which millions of people are already being subjected: the cognitive atrophy caused by the use of AI, which has been the subject of recent research by the MIT Media Lab. A cognitively atrophied older person still has decades of cognitive development behind them to compensate for the damage; we should start asking ourselves what might happen to younger people, so as to avoid burning out entire generations and only realising it when it is too late.

When we bring these five areas together, a picture emerges that is by no means anti-technology. The question is not whether to adopt AI, but how to do so in a considered manner: with greater attention to data quality, human oversight of critical processes, the value of people – which remains indispensable – an honest assessment of the real costs, and a focus on what makes a business unique in the market. Artificial intelligence without governance may seem like a highly efficient tool for growth, but it ultimately becomes a technical, cultural and organisational liability. The true value of AI for an SME does not lie in replacing people to save money today, but in the ability to understand whether and how the tool should be adopted and managed in a way that serves the company’s objectives. To do this, we need strategy; we need to stop and think – as only we humans know how to do.

*Founder and Co-CEO of Mudra