Artificial intelligence, why it struggles in business
Generative tools struggle to understand context and thus to integrate into systems and organisations
3' min read
Key points
3' min read
The news bounced out of the United States at the end of August, breaking the balance of common thought that had been created for months around artificial intelligence and the beneficial effects of its application in business. And, instead, the one produced by the Mit Media Lab ('The GenAI Divide: State of AI in Business 2025') is a critical report on the goodness of the investments made in this technology.
Mit: "Low returns on investment"
.The reason? Quickly explained: despite the tens of billions of dollars spent by large companies, only 5% of pilot projects based on Gen AI tools generate rapid and measurable revenue acceleration. Most (the remaining 95 per cent) therefore remain stuck and do not produce returns on the income statement in terms of operating margin. MIT researchers have circled the real impact of generative artificial intelligence in red and with the definition of the 'Gen AI Divide' highlight the risk of minimal transformation in the face of a high diffusion of solutions, with a few sectors showing signs of structural disruption while most experiment a lot without concrete results.
Low contextual learning
.Where, then, does the criticality accompanying the adoption of generative tools in companies lie? In the transition from the pilot phase to the 'putting into production' of the projects, and in this case in causes attributable to the lack of contextual learning and the limited ability to integrate these solutions into operational processes (and to the concomitant difficulty of adapting systems and organisation to the AI verb). The main barrier to the success of these projects, in other words, would be the lack of understanding of a technology that struggles to adapt to the context and process feedbacks, not improving (and not correcting its errors) over time.
The report by the American research university has thus shone a light on a problem that is certainly not unknown and highlights a further critical element. Which is? The informal use of Gen AI, with only a minority of companies (40 per cent according to the report) having purchased premium subscriptions to Llm models, while 90 per cent of employees say they regularly use generative tools for their own work.
Piva: 'Few skills and inadequate infrastructure'
.The latter phenomenon has also been observed in Italy, as confirmed to Il Sole 24 Ore by Alessandro Piva, director of the Artificial Intelligence Observatory of the Politecnico di Milano, according to whom 'interest in Generative AI in medium-sized companies has grown, but adoption still comes up against structural obstacles such as the lack of internal skills, the difficulty of integrating it into existing processes, inadequate infrastructures and regulatory uncertainties. In addition to these critical issues,' the expert points out, 'there is the growing risk of ungoverned use of the technology by company employees, so-called shadow AI, which can lead to unintentional exposure of sensitive data and compromise corporate security and compliance'.
