How Italian CIOs produce value with gen AI

ChatGPT was a watershed moment in the evolution and adoption of AI. Some CIOs in Italy watched its boom with a hint of skepticism, though, wondering where the hype ended and real opportunities began. Others have seen gen AI as an opportunity to advance their AI projects with uniquely constructed technology.

This is the case of Alpitour World, one of Italy’s largest travel companies, where gen AI projects have merged with preexisting uses of AI. Case in point is its new conversational assistant copilot, AlpiGPT an internal search engine of corporate data that can personalize travel packages and quickly answer questions, says company CIO, Francesco Ciuccarelli. Employees are even calling it a trusted colleague.

According to Forrester’s 2024 predictions, 60% of skeptics will overcome their gen AI doubts by the end of the year and appreciate it for its uses in conversational assistants, and its ability to translate and synthesize content.

Another gen AI application winning over CIOs is its knack for coding, according to Alessio Maffei, ICT manager of Milan-based student and family-focused travel company Inter-studioviaggi.

“At first, I was wary of generative AI,” he says. “When ChatGPT came to market, and there were no other competitors, I had the impression it was hype. I tried to use it for text generation and information retrieval, but it seemed more suitable for a consumer environment than corporate reality. But driven by the need to code on new platforms and have fast responses, I started using gen AI for this goal and I found it very useful.”

Creating new business models

Gen AI is also unique in that it can generate useful business models. For example, in the telecommunications industry where operators have been struggling with shrinking margins for years, McKinsey estimates gen AI will help it recover quickly thanks to jobs in network operations, customer service, IT, marketing and sales, and support functions.

Some gen AI applications can already summarize customer voice and written interactions with the contact center, or, in marketing and sales, identify new sales leads from calls. One telco pilot project has achieved a conversion rate of more than 10%, for instance, and also creates personalized messages to address individual customers. Plus, it’s used to speed up procurement analysis and insights into negotiation strategies, and reduce hiring costs with resume screening and automated candidate profile recommendations.

Telcos in general are also experimenting with gen AI to analyze network data and streamline the entire software lifecycle, including generating and scanning code for vulnerabilities before launch.

Gen AI for coding

In Inter-studioviaggi’s case, gen AI applied to software development has acted as a battering ram to open up widespread use cases in corporate IT. Having overcome the initial perplexity about ChatGPT, Maffei tested gen AI in coding activity and found great benefits. This led to the start of more general gen AI projects for functions such as search in the company database, experimenting with ChatGPT, Microsoft Copilot and Google’s Gemini.

“The results of ChatGPT and Gemini were, in some ways, aligned, but with a diversity of thought similar to asking a question to two fellow developers,” he says. “The answer was similar, and correct, but as if it had been processed by two people with different skills and training. This was enriching to evaluate how to operate based on my way of writing code. It allowed me to program the way I’m used to.”

Maffei’s need was to create new apps using modern languages, similar to the previous ones, but with long-term support, and there were elements where he didn’t have time to train himself.

“But I was able to release a new platform on schedule, because generative AI provided me with quick solutions for writing code using the new language,” he says. “It was a great help on the timing of programming. In this project, gen AI halved the hours needed for my work.”

Keys to faster innovation

Maffei is now studying the inclusion of gen AI in some of Inter-studioviaggi’s services. For instance, there’s a project in the pipeline for a chatbot that responds to customers at times and on days when it’s more difficult to find live operators, or makes initial selections of requests to be passed on to operators.

“We have to respond to customers 365 days a year, 24 hours a day, and a chatbot that gives clear and certain answers, or shortens the conversation time with customer service advisors, is important,” says Maffei.

This project will start from the creation and training of the model on some specific Inter-studioviaggi programs where it’s easier to train AI, like short-term study abroad holidays, for which the company already has a large database.

“AI is the future for us,” says Maffei. “After this project, we’ll constantly introduce AI on other sectors and services like control of travel documentation.”

The evolution of the chatbot

For Alpitour World, the first experiments with traditional AI date back to 2018, conducted in an internal IT laboratory and in a protected research context. The following year, it created an AI & Automation function, and then in 2020, launched its first chatbot. Despite the progress, setbacks occurred.

“Since 2021, we’ve begun to observe the limitations of this technology, namely the burden of training models to change the context or even just to answer new questions,” notes CIO Ciuccarelli. “But when we saw the launch and rapid spread of ChatGPT, we immediately grasped its potential to meet our needs and began experimenting on travel editor activities, for instance, to prepare to answer questions coming from sales channels, and for assistance on Alpitour World products. In this context, generative AI is a very useful support to create content.”

Another application developed through the AI & Automation function was the internal chatbot AlpiGPT.

“I’ve given colleagues the freedom to do research and experimentation together with our automation partner Mauden,” says Ciuccarelli. “We chose OpenAI’s GPT model acquired through Microsoft’s Azure, because GPT appeared to us to be the best at the time, and Azure was already being used in the company, allowing us fast and secure access regarding information segregation.”

This facilitated the transition through the data protection officer for compliance, which Ciuccarelli points out is an important part of an AI project. On the cost side, however, he didn’t have to ask to buy more GPUs, because it adopted AI as-a-service.

Choosing a model that fits

Ciuccarelli emphasizes the importance of a provider’s level of service, because it guarantees stability and speed of the product’s responses.

“For me as CIO, the constraint to scale production is the reliability of the system,” he says. “Alpitour World’s customer experience depends on the chatbot, and our provider guarantees us dedicated resources and constant response times. If you want to ensure performance, privacy, hallucination control, and other quality of service parameters, I think the closed model is better than a hyperscaler that simplifies some aspects compared to the open model. But it doesn’t mean this will be my answer in the future or for all types of applications. Research and development in the open world is moving at an impressive speed, and there’s the benefit of greater transparency.”

In fact, Alpitour World is also experimenting with “punctual” uses of gen AI with open systems, including small language models for coding activities, or specialized vertical applications, such as an assistant to support internal functions.

“It’s more flexible and less expensive,” Ciuccarelli adds. “But it’s not certain we’ll not also use these open systems in production and for external projects. In the short term, proprietary systems are a more prudent choice and give a fast time-to-market, but I don’t rule out open models. On the contrary, I imagine a multi-model future along the lines of multicloud.”

The MLOps paradigm

Equally important for Ciuccarelli is updating the gen AI model with MLOps and LLMOps, which help AI and algorithm governance. “In fact, machine learning models and generative AI, being based on neural networks, risk greater drifts and need exact prompts,” he says. “Such new phenomena aren’t always easy to understand and govern. New components, such as vector databases or orchestrators, have also entered the architecture.”

According to analysis by the World Economic Forum, governance is one of the four pillars of gen AI implementations, along with staff training, budgeting, and team alignment. The CIO will need to be supported by the CISO so concrete business cases for gen AI are accompanied by the study of risks and KPIs that guide targeted and defined solutions. Governance will also need to have the right amount of flexibility, in that it has to manage, not prohibit, the use of gen AI products by employees.

Bringing IT and business closer together

In current gen AI implementations and in those being studied for the future, Maffei has started from a ready-made model on which training is carried out with collaboration from expert external consultants.

McKinsey emphasizes the importance of gen AI to avoid doing everything in-house and instead, evaluate out-of-the-box options from start-ups or SaaS providers. Buying standardized products — those with which Ciuccarelli is now also involved — speeds up innovation, and suppliers can then be asked for custom features.

Another gen AI benefit in this regard is its ability to adapt and reuse implemented solutions for different use cases. An AI-powered chatbot developed to improve the productivity of call center agents, for example, can be repurposed with additional optimizations or data to answer new employee FAQs or provide IT support. Forrester also points out it brings IT and business closer than ever before. The success of generative AI projects — and the CIOs who lead them — is closely tied to the ability to understand what the business expects from IT, and deliver a product that answers these questions. These are applications that reduce cost and time, and allow you to do more with the same resources and co-create value.

For all projects involving AI at Alpitour World, IT, the internal research laboratory, and the Automation & AI function all act together by activating the various IT nodes distributed in the organization that connect technology with business.

“There’s always a link with the other functions,” says Ciuccarelli. “They express their needs, and IT takes care of the governance and design of the platforms. But gen AI makes the dialogue between IT and business more efficient because it helps build applications together with the feedback mechanism that makes the user the protagonist. Gen AI brings people and technology closer together, and IT has an extra edge to explain technologies and governance to the business.”