The study

This is why 95% of Ai projects fail

Research by iKN Italy and Casaleggio Associati explains the real cost of artificial intelligence for companies

by Massimo De Laurentiis

24 February 2026

3' min read

Translated by AI

Versione italiana

3' min read

Translated by AI

Versione italiana

After the initial enthusiasm for the arrival of artificial intelligence, it is now time for companies to take stock: what works, what is the real impact of this technology and, above all, how much does it cost. To answer these questions, a study by iKN Italy in collaboration with Casaleggio Associati analysed the hidden costs and the real return on investment of Ai tools for Italia's business fabric.

Companies do not know how much Ai really costs

The starting data are not encouraging: around 60% of companies report marginal or below-expected benefits, and 95% of Ai projects do not make it past the pilot phase. Furthermore, less than half of the companies have a complete picture of the costs and profitability of their trials. This is because companies are often unaware of the real cost of these projects: they tend to focus only on the most visible and easily controllable items of expenditure, but underestimate certain variables that are crucial for the programmes to survive and generate a real impact in the long term.

The Value of Time

"We have found that many realities still consider artificial intelligence as a product to be purchased and not as a transformation involving the entire organisational structure," explains Laura Ghisleri, Content&Networking Director at iKN. On the contrary, to really exploit the potential of this technology "the governance of processes must be created for the application of artificial intelligence".

One of the aspects the study focuses on is the value of time when it comes to Ai. "Delay is not neutral, it is a tax on future value," reads the research, which highlights the advantage of getting Ai solutions up and running faster than others. Those who start earlier accumulate data, skills, processes and trust that create a delay that is difficult to recover.

Therefore, companies must strike a balance between haste, which can produce fatal errors for business, and delay, which is equally damaging. "First of all, it has emerged that decision-makers must be competent in the field," says Ghisleri, "and furthermore, KPIs must be clearly defined to monitor the actual return on investment in artificial intelligence.

The Total Cost of Ownership

Despite the fact that many pilot projects have failed so far, the data collected by iKN does not indicate a decrease in investment in Ai: 'You should not be afraid of failure because failure can be the way to improvement,' Ghisleri continues. More awareness is beginning to emerge without this turning into disillusionment: "What we have seen through our research is the need to get down to the concrete, to stop testing and arrive at a defined plan to evaluate the results".

This is why the research speaks of the Total Cost of Ownership (Tco) of Ai, which also includes the costs of data preparation and management or organisational costs, including process redesign and worker retraining. But Tco also includes variables that are more difficult to estimate, such as ethical and legal risks or the increasingly relevant geopolitical risks, which require long-term assessments to avoid technological lock-in and risky dependencies.

Small and Medium Enterprises

In this context, the weight of the importance of resources is especially felt by small companies. But small companies also have characteristics that can turn into strategic advantages in the age of Ai: "Small and medium-sized enterprises have the ability to reorganise themselves much faster than large companies," comments Davide Casaleggio, president of Casaleggio Associati. "Artificial intelligence can develop new business models that make it possible to attack new markets in a way that was perhaps impossible just a few months before. In this, SMEs can have an edge, simply because they are more agile'.

The "shadow AI"

Another issue highlighted by the research is the danger of so-called 'shadow AI', i.e. the use of personal artificial intelligence tools by employees, which can create problems especially for security and data integrity. 'To avoid this, the first point is training,' Casaleggio comments, 'people need to know what the problem is of putting a confidential business plan inside ChatGPT for a translation. You don't necessarily need to be equipped with proprietary tools, not all companies can afford it, but you need to be very clear about how you can use Ai tools, in what contexts and for what purpose.

The study shows that the challenge is first and foremost cultural. There is often a lack of real understanding of risks and opportunities and the tendency is to focus attention only on the most obvious aspects of this technology. "The main problem is not governing time," Casaleggio concludes, "confusing speed with haste and waiting with paralysis.