How Eni integrated AI search into its global website

Italian energy company Eni has long been leading the way when it comes to internal search tools. The company's new "energIA" on-site AI search tool was one of the first to appear on a corporate website. Jonathan Holt speaks with Massimo Guarnieri, Digital Experience and Identity Manager at Eni, about how they did it, gaining the approval of senior leadership, and what others can learn from their experiences.

Jonathan Holt: Long before AI search integrations, Eni had built a natural-language search engine that was much more advanced than the average search tool on corporate websites. Did you learn anything through that work that you were able to carry forward into energIA?

Massimo Guarnieri: It was 2016 when we implemented the natural-language search engine, Ask Now, on eni.com. A new trend was spreading on public search engines: formulating meaningful questions rather than simply entering keywords.

We began working on the concept of “intent”, mapping users' possible search objectives and defining the rules for constructing the best response, in line with communication rules. Adopting this solution allowed us to gain interesting new insights. We could analyse user behaviour, how information was searched for on the website and what the expectations of the various audiences were.

The analysis of the queries led us to enrich the list of intents initially identified. Finally, we realised that, in order to be more effective, it was necessary to integrate the system with a dedicated built-in dataset.

This experience was the basis for the energIA project. We recovered it in its entirety and enriched it with new technology.

JH: What were the main obstacles in terms of making energIA a reality?

MG: Being a first mover did not help, although this was clear from the very beginning. The native technical limitations of the solution we could adopt did not fully meet our communication needs. For this reason, we had to identify approaches that would allow us to remain aligned with the initial objectives.

It was not easy to convey how challenging it was going to be to implement this solution, given that most people were already familiar with public LLM solutions. However, it was not possible to take an external solution and apply it, as is, internally. The only feasible path was to use a raw component (RAG) and customise it with dedicated rules to ensure the reliability of the responses.

We therefore found ourselves facing very high internal expectations from non experts: the expectation of achieving a solution with functionalities even more advanced than those available on public tools.

JH: Why did you opt for a freestanding AI agent rather than one that is part of the main search facility or focussed on a specific user group?

MG: We have decided to separate the two functionalities for technical reasons. The development of this solution has been highly complex and has required a significant effort. To achieve the objective, it was essential to develop it in a way that avoided interdependence, thereby reducing potential criticalities.

The next step will be to integrate them into a single entry point that will not only support the formulation of questions but will also accommodate keyword- and document based searches.

We have chosen to make this solution available to all users, rather than restricting it to a single user group, to enhance the value of the product by equipping it with a knowledge base that covers all corporate topics addressed on the corporate website.

JH: What guardrails have you put in place?

MG: Artificial Intelligence is not immune to hallucinations. We have taken care to state in several places that energIA’s responses do not represent the company’s official position, and we encourage users to always verify the sources provided alongside each answer. Furthermore, for matters of particular relevance, we advise users to refer directly to the company’s official points of contact.

Despite these precautions, we have made every effort to minimise the risk of inaccurate responses by acting both on the selection of documents included in the knowledge base – avoiding redundancy across published content – and on the parameters governing the confidence level associated with each output. We also conducted extensive testing, providing confirmation of the quality and robustness of the work carried out.

JH: Was it difficult to get senior buy-in? How did you succeed?

MG: We have never been accustomed to following the mainstream; on the contrary, one of our characteristics is to experiment with innovative solutions in the field of digital communication. To date, our attempts have yielded positive results. Partly thanks to this prerogative of ours, but above all on the basis of the internal skills we have developed and demonstrated to date in both digital communication and technology, we have managed to obtain the necessary endorsement from top management to try to overcome this challenge as well.

JH: Is energIA proving popular with visitors to eni.com? What feedback are you getting?

MG: As soon as we published energIA, there was a high level of interest, especially among employees.

Our collective experience in consulting AI solutions has now become part of everyday life, and we are all accustomed to asking for information of any kind, including the most basic. As a result, users of eni.com had very broad and varied expectations, extending well beyond the intended scope of the solution.

Setting aside questions that clearly fall outside the remit of our corporate website, we nonetheless reflected on the opportunity to address queries that we had never previously considered covering, but which – upon careful reflection – are indeed worth reassessing. One example is: “How can I purchase Eni shares?”. Although a corporate website is not the place where shares are bought, such a question is entirely legitimate for a retail investor. And therefore, why should we not address it?

JH: What have you learned that you didn’t know on the day energIA launched?

MG: Reading some questions asked by users, I had further proof that reality exceeds fantasy!

JH: Do you have any advice for digital managers who are considering launching AI on their own sites?

MG: The first suggestion is to carefully assess whether it is truly worthwhile: to have absolute clarity regarding the risks and the objectives of the project.

The second is to remind oneself that there is no magic wand. Applying GenAI to your content will not, in itself, bring the work to a swift and effortless conclusion. Although the solutions available on the market today are undoubtedly easier to implement and more powerful than those that existed when we started this project, a significant effort, technical expertise, and – above all – common sense is still required. It remains essential to be oriented towards simplification and accustomed to thinking outside the conventional frameworks.

However, what underpins everything is a well balanced and cohesive project team which, in our case, performed exceptionally well. Finally, one must always be willing to question oneself, while avoiding the temptation to give up – even when it seems that no viable solution is within reach.

JH: What do you think the corporate website will look like five years from now?

MG: In my view, this has two components. The first is that the number of visitors to corporate websites will continue to decline. The second is that, in the near future, we will see a significant shift in the composition of the audiences visiting our sites. Categories such as investors and media – who are, by definition, interested in primary sources – will likely grow. Conversely, jobseekers may decrease, as they will find job postings already aggregated on external platforms, pre filtered on the basis of their CVs submitted via prompt.

Content architecture will need to be reassessed, and the website will largely consist of a database structured with precision, so that it can be easily processed by AI systems.

We will have to move away from traditional user journeys, as access paths will be very different and far more immediate – straight to the point. Companies will have less room to use their websites for corporate storytelling and will instead need to build a dynamic experience that adapts to user behaviour. Receiving a question from users will be a valuable starting point because, in addition to providing an answer, we will be able to propose a series of relevant follow up insights aligned with their identified intent.

In my view, companies will need to make a significant investment in centralised and certified data structuring. This data must be capable of feeding multiple channels: from the internet site to the intranet, from presentations to the generation of official documentation, and at the same time must be constructed in such a way as to support traffic generation via SEO.

JH: Is there anything else that you would like communicators to know about the work you’ve done on energIA?

MG: I would like to return to the importance of the team, the people and the way they work. The key to success was undoubtedly the team of professionals who worked on the project. We often found ourselves in situations where there were more problems than solutions. However, each member of the team was able to step outside their role and make a contribution that went far beyond their specific area of expertise. The project obviously started with a list of technical requirements, but it was also essential to share the communication objectives with the technical team in order to share the strategic vision of the project. At the same time, it was important for the communication team to understand the reasons behind the technical issues that arose. This meant that one person's problem became everyone's problem. This created a winning team spirit. It's no secret, but at the same time, it's not always a given.