Four ways to introduce AI in companies

The starting point of the research mentioned above is the tension characterizing the current phase of AI: on the one hand, keen interest and competitive pressure; on the other, limited clarity, technological ambiguity, and the need for substantial critical mass to effectively approach the subject. Many companies have data that are not yet fully reliable and systems that remain fragmented, yet feel the urgency to “be there” in the world of artificial intelligence.

So the authors asked themselves: How are Italian companies actually moving forward? And what organizational and technological responsibilities stem from these choices?

The study, involving 16 medium- and large-sized Italian companies, was conducted through case studies, manager interviews, and comparative analysis. What emerges is a varied picture: AI is being tested in the design of new products, supply chain management, marketing and customer service, but also in administrative processes, IT, and human resources. The hoped- for benefits are greater efficiency in processes; cost and error reduction; scalability and productivity; and improved customer experience.

The most interesting aspect, however, concerns the organizational models of implementation. In this vein, four adoption archetypes emerged:

• Infused AI, with modules already embedded in ERPs, where most responsibilities remain with the provider.
• AI on demand, where external tools are used without native integration, carrying potential governance risks but offering flexibility in the choice of market tools and the ability to evolve.
• Interfaced AI, based on vertical modules that interact natively with corporate systems.
• Hybrid architecture, where AI modules interface with an architecture composed of different management systems.

These archetypes are not equivalent alternatives. Instead what changes are the levels of responsibility and oversight required by the adopting company, and conversely, the degree of customization that can be achieved. The more complex and customized the solutions, the greater the governance burden on the company.

The lesson for managers is twofold. First, AI can’t be a band aid to fix poorly designed processes. Without clean data and coherent architectures, there’s a risk of triggering a vicious cycle. Second, adoption requires a gradual, congruous process:

• An incremental approach, starting with projects natively consistent with business priorities.
• Balanced governance, with IT safeguarding security and compliance and business functions free to experiment.
• Investments in culture and skills to support people in the critical use of tools.

Finally, the role of providers is far from marginal. Indeed, in a market where players change constantly, having a supplier capable of progressively integrating new solutions is what can help companies translate hype into lasting value.