Artificial intelligence is only beginning to reveal its potential, but experts warn that no strategy built around the technology can succeed without a solid data foundation. Matteo Verdari, Head of the Global AI Practice at SDG Group, explains why companies must bridge the gap between hype and operational reality.
This is what we call the “adoption gap” — the difference between what technology can do and what companies actually implement. Many studies estimate that around 60 percent of AI projects fail when they move into production. The reasons vary, but the common denominator is often a lack of tangible business value.
This usually comes down to misalignment among stakeholders regarding how organizations, processes, and systems need to evolve to integrate these solutions. There are also technical challenges, as many companies still rely on legacy systems that are not suited to advanced technologies. Finally, there is a cultural dimension, with people struggling to adopt new tools or change the way they work.
The fashion industry, like many others, is navigating a complex transition shaped by geopolitical and economic uncertainty. This often pushes companies to prioritize very short-term returns on investment, which becomes a major obstacle: if a use case doesn’t deliver immediate results, it is quickly abandoned.
The result is a continuous cycle of evaluations and proofs of concept that never translate into real business value. This is unfortunate, because there are already clear examples of how AI can generate value across the entire value chain. In product development, for instance, AI can accelerate the creation of design concepts. In inventory optimization, it can help tailor stock to specific locations and customer preferences. And in after-sales service, it can improve complaint management and reduce return times.
To build a house, you must start with the foundation – and in this case, the foundation is data. There can be no AI strategy without a solid data strategy. The starting point is having the right infrastructure in place to collect, organize, and manage the data that will feed AI models.
Once that foundation is established, the next step is to think strategically about use cases. At SDG Group, we created a dedicated unit, Orbitae, designed to support companies throughout the entire process – from design to engineering – while also managing models across their lifecycle and ensuring responsible adoption through training and organizational change.
Automation itself is not new. For years, companies have relied on technologies such as robotic process automation, which use predefined rules to carry out repetitive tasks. But these systems are relatively passive.
Today, we are entering a new phase with AI agents that, when properly trained, can take full responsibility for executing tasks end to end within a process. In procurement, for example, “negotiation agents” can manage orders by participating in tenders or analyzing bids. They can understand context – including inventory levels, pricing, or outstanding orders – and make decisions such as issuing purchase orders or triggering inbound logistics processes.
Successfully managing this shift requires two key elements: governance – clear protocols to ensure transparency in how these agents operate – and human adaptation, meaning a willingness to adjust work habits to collaborate effectively with these systems.
Talent and a deep understanding of the business will remain essential in every role. It is crucial to retain the ability to interpret what is happening beyond existing frameworks. Human intelligence can read between the lines, while large language models read within the lines.
This human capability should evolve to include AI as an additional voice. Think of it as a committee where, instead of being composed solely of people, an AI participates to offer different perspectives or scenarios that support better decision-making.
The key is to combine human responsibility in decision-making with the insights provided by machines – what we call a “human-in-the-loop” approach, where AI acts as a complementary support.