Implementing AI in business processes is no longer just an option or an opportunity — it is increasingly a necessity for organizations that want to remain competitive. Today, artificial intelligence demands a radical shift in how productivity is optimized and costs and timelines are reduced. How to do it, however, is still far from clear.

Because while AI is on everyone’s lips, actual adoption and real understanding of how to use AI for industrial automation and organizational processes remains limited. In Italy, in 2024, 8.2% of companies with at least ten employees used at least one AI-based technology — an increase from 5% in 2023. Despite this significant growth, the country remains below the European average (Sources: AI 4 Italy – Istat; Eurostat).

The main challenges facing Italian organizations in implementing AI-based technologies are:

Limited AI investment. The AI market in Italy is valued at approximately €0.8 billion, with growth projections reaching €2.5 billion by 2027. Current investment levels, however, remain lower than in other European countries (Source: Global Technology Report 2024, Bain & Company).

Inadequate infrastructure. According to the Cisco AI Readiness Index, only 10% of Italian companies feel fully prepared from an infrastructure standpoint to support AI, with significant gaps in computing capacity and network performance.

Poor data quality. Many companies do not feel ready to manage data for AI projects. Only 13% of companies globally — and 10% in Italy — consider themselves ready in terms of data, pointing to challenges related to fragmentation, low quality, and the absence of metadata (Source: Cisco AI Readiness Index).

Lack of skills. The shortage of qualified personnel is probably the biggest obstacle. According to the Minsait-Luiss report “Artificial Intelligence in Italy — The Revolution Changing Business,” 19% of companies report a deficit in AI skills and specialized professionals.

Unclear application landscape. One consequence of the skills gap is an equally vague picture of what AI can actually be applied to.

---

Alternative Examples of AI in Business Processes

These figures align with our own experience of guiding organizations toward the practical application of new technologies. When we meet organizations interested in AI, ideas are usually quite confused, and it is rare for the conversation to move beyond the generative AI experience people have had with ChatGPT and similar chatbots.

It is only after going deeper into needs and solutions that involve alternative AI applications that we see the fog lift — visible in the way understanding visibly arrives for the people we are talking to. Because AI is much more than a text response. Here are some practical examples of artificial intelligence beyond chatbots, including a case study built here at Geckosoft.

Reduce costs with predictive maintenance. In manufacturing and industrial settings, equipment failures can mean significant costs and delays. Through predictive analytics, AI can monitor real-time data and anticipate when a machine will need maintenance, reducing downtime and optimizing resource management.

Personalize the customer experience — not just in chat. If chatbots answer questions, AI can do far more to improve the customer experience. Advanced algorithms analyze user behavior to deliver personalized experiences: suggesting products, optimizing marketing campaigns, and improving conversion rates.

Optimize the supply chain. AI can optimize the entire supply chain by forecasting demand, improving logistics, and minimizing waste. Advanced algorithms analyze historical data, market trends, and weather conditions to predict raw material availability and optimize inventory levels.

Raise the level of cybersecurity. Cyber threats are constantly evolving, but AI can provide a proactive defense. Machine learning algorithms can detect anomalies in network traffic, identify phishing attempts, and block attacks in real time.

Make better-informed decisions with advanced analytics. AI can help organizations make more informed decisions by analyzing large volumes of data to identify patterns, trends, and market opportunities. A retail company, for example, can use AI to analyze sales data and predict which products will perform best in upcoming seasons, optimizing purchasing decisions and maximizing margins.

Train AI on the specifics of your organization.

The Geckosoft case for Uluru. Uluru is a US-based startup that supports middle and high school students in their educational journey. It integrates with the LMS platforms teachers use to assign homework to students online, and helps organize study sessions based on a method developed by Uluru’s coaches. Geckosoft was brought in to build an AI trained on the Uluru method: an LLM capable of breaking homework assignments into smaller tasks, estimating the time needed to complete each one, and identifying the best moment in the day to do so based on the student’s availability calendar. No chatbot, then — but a technology capable of functioning as a real Uluru coach for every student, in a personalized way, seven days a week, twenty-four hours a day.

This was achieved by training the LLM on a large series of examples of assignments broken down and organized by duration and difficulty level. The examples were first generated through prompt engineering, followed by a fine-tuning process carried out with the help of Uluru’s coaches — ensuring precise optimization and high personalization of the AI model delivered to the client.

The Rational AI model is integrated directly into the Uluru app’s user experience: not a “classic” chat interface, but something that merges fully with the app’s UX, embodied in the Uluru mascot, helping each student plan their study activities.

---

What Is the Difference Between Prompt Engineering and Fine-Tuning?

Prompt engineering and fine-tuning are two approaches aimed at optimizing the use of an AI model.

Prompt engineering involves strategically framing the request — the prompt — given to an AI model in order to obtain the best possible response. The work happens on the input side, without modifying the model itself. Different formulations, structures, and levels of detail are tested to guide the AI’s behavior. Prompt engineering is fast, inexpensive, and works on any pre-trained model, such as GPT-4.

Fine-tuning is a process in which an AI model is retrained on a specific dataset to adapt it to a particular context or sector. New training data is fed to the model to update its weights and improve responses according to the organization’s needs. This makes the AI model more precise and optimized for specific tasks and domains — and provides protection against hallucinations and generic responses.

---

The examples shared here are just a subset of how AI-based technologies can help organizations improve productivity, reduce costs, offer services that would have been unthinkable a few years ago, and create new business opportunities.

Want to clear up the uncertainty surrounding your ideas about artificial intelligence? Let’s work through the essentials together.

---

TL;DR

AI adoption is now a competitive necessity, but many Italian companies are falling behind due to limited investment, insufficient infrastructure, poor data quality, a skills shortage, and limited clarity on applications. Only a minority use AI in any meaningful way, remaining below the European average. The most effective applications include predictive maintenance, advanced personalization, supply chain optimization, cybersecurity, and decision analytics. The Uluru case shows how models trained on specific data can go well beyond chatbots and integrate directly into processes. Prompt engineering and fine-tuning are complementary approaches for optimizing models. AI offers significant room for efficiency gains, cost reduction, and new services.

---

Originally published on Geckosoft website, translated with Sonnet 4.6 through Rational AI.