From experimental AI to regulated AI: the value lies in the ability to transform technology
Artificial intelligence has now become part of the processes in many organisations, and everyone agrees that the real challenge is no longer to demonstrate its potential, but rather to put this technology into production and realise its value, to measure its effects and impacts, manage the data and security risks associated with it, and maintain human control over the company’s technological choices.
For artificial intelligence, the time has therefore come to put it to the test. Following a phase in which the priority was to understand what generative models were capable of, the focus has gradually shifted to more concrete ground: which processes can be transformed, which results can be measured, and what the actual value created is, net of the initial investment.
The figures compiled by leading international research and consultancy firms clearly illustrate this transition. According to Stanford’s AI Index 2025, in 2024, 78 per cent of the organisations surveyed reported using AI, compared with 55 per cent the previous year; those using Generative AI in at least one business function had risen from 33 per cent to 71 per cent. However, the same research highlighted that, in most cases, the reported economic impact remained modest.
The paradox between adoption levels and returns is even more evident when looking at the scalability aspect. In the global McKinsey survey published at the end of 2025, in fact, 88 per cent of respondents stated that their organisation used artificial intelligence in at least one business function, whilst only 7 per cent said that the technology had been rolled out at full scale across the organisation. The priority for CEOs and senior management must therefore be the ability to transform experiments and pilot projects into manageable tools that are integrated into business processes.
In other words, it is not enough to choose a powerful model, link it to a data repository and build an interface that makes it accessible to everyone; a truly effective AI system must address data quality and availability, security, computational costs, interoperability with existing architectures, staff skills and the ability to measure effects over time.
The risk factor and the issue of sovereignty
Added to this scenario is an element that has become central in recent years: risk. An AI model, however advanced, can still produce incorrect information, amplify biases present in the data, expose sensitive data or generate results that are difficult to verify. It is no coincidence that these risks have also been recognised by the US National Institute of Standards and Technology, which, in its profile on Gen AI, has identified other factors deserving the utmost attention – ranging from privacy and information security to environmental impacts – as key areas requiring monitoring.
The governance of artificial intelligence and agentic AI tools cannot therefore be regarded as an afterthought to be added at the end of a project, precisely because this technology must be an integral part of the design process. This means defining which data can be used, the limits of the machine’s autonomy, which models to choose and how to evaluate them, how to monitor responses, and how to intervene when the behaviour of the AI-co-piloted system is not consistent with strategic objectives.
Finally, there is the issue of technological autonomy. In the debate on digital sovereignty, the geographical location of the infrastructure or the nationality of the technology provider is not the central issue. The key to the ‘safe’ use of AI lies in the ability to retain decision-making capacity and control over data, models, architectures and technological dependencies. In Europe, the regulatory framework aims to make this requirement increasingly concrete, and since 2 August new provisions of the AI Act have come into force, including those on transparency for certain systems and enforcement powers; for certain categories of high-risk systems, however, implementation has been extended to December 2027, whilst for AI embedded in certain products, the deadline is August 2028. Setting aside the regulatory process, the transition from experimentation to production – when it comes to artificial intelligence – requires a method and, above all, demands that AI be integrated into business processes, where its value can be defined, measured and managed. This goes far beyond what is possible at the purely technological level.
Ultra, Almaviva’s framework for integrating AI into business processes
Driven precisely by the need to provide guidance on realising the value of AI within organisations’ most important and complex processes, Almaviva has developed Ultra, a framework that brings together the Group’s expertise in artificial intelligence and aims to support businesses and public administrations throughout the entire process, from strategy to implementation. Underpinning this framework is a firm conviction: to generate value, AI must be designed, implemented and managed through a model of ‘governed evolution’, capable of bringing together technology, data, security, people and business objectives.
According to Almaviva’s description, Ultra is structured around five foundations (Trust, Bridge, Transformation, Factory and Automation) and three vertical pillars. The foundations reflect the aim of bringing together sovereignty, security, compliance and sustainability (in the case of Trust); of connecting research, ecosystems and cutting-edge technologies (Bridge); of combining strategy, governance and adoption whilst harnessing people’s ability to drive change (Transformation), to take innovation from experimentation to full-scale operation (Factory) and, finally, to manage the lifecycle of software and processes through an AI-native engine (Automation).
The three pillars are built upon these foundations. The first is ‘Made’, which encompasses products and intellectual property developed by the Group, with the aim of making AI a governable and sovereign technology; the second is ‘Bespoke’, which concerns the development of solutions tailored to the client’s specific context; the third and final pillar is “Vertical”, which brings together vertical AI solutions tailored to specific sectors to meet the specific needs of each market sector more quickly.
Technological independence and measurable impact
In designing and developing Ultra, Almaviva took into account the specific characteristics of particularly complex and regulated environments, such as public administration, healthcare, the utilities and transport sectors, as well as the finance, industrial and defence sectors. The stated aim is to maintain a single architecture for data management, accessibility and protection, adapting it to different application contexts. Among the benefits highlighted by Almaviva are the freedom of technological choice – and consequently the absence of the risk of dependence on a single vendor – data governability and traceability and, above all, the ability to measure the impact on processes. In short, it is not a single model imposed on all use cases, but the ability to select ‘KPI-oriented’ architectures, technologies and solutions tailored to the specific problem, utilising a vendor-agnostic and multi-cloud approach. The proposal is also underpinned by a wealth of technological expertise in AI, comprising over 300 active projects, more than 30 experts and researchers specialising in artificial intelligence, over 900 AI-certified professionals, and a family of LLM (Large Language Models) designed from scratch on a proprietary architecture. At the same time, it offers the ability to deliver AI solutions both as off-the-shelf products and as bespoke solutions, tailored entirely to the client’s needs.
Finally, Ultra’s position is further strengthened by the recent launch of Almawave Labs, the Group’s new R&D subsidiary dedicated to sovereign artificial intelligence solutions. The company brings together around 100 AI specialists, including over 30 PhD holders, and focuses its research and development activities by leveraging proprietary assets such as the Velvet family of generative models and the AIWave platform. With the launch of Ultra, Almaviva is therefore reinforcing the principle that, in order to link innovation to tangible results, the discussion on artificial intelligence must shift from promise to execution. The new framework has been designed specifically to make this capability available to organisations and provide them with the tools to use AI as a governed, measurable and process-integrated component.
