Every law affects lives. But before a law is passed, it should be clear who it will affect, how much it will cost, and whether it creates unnecessary burdens. Traditionally, understanding these effects—known as regulatory impact assessment—has been a slow, manual, and error-prone task, often constrained by limited human resources and document complexity. In Portugal, assessments could take days or even weeks and often lacked full consistency and transparency. This hindered timely decision-making, discouraged stakeholder participation, and introduced subjectivity into a process that should be evidence-based. The AI4IA project changed that.
Funded by the European Commission’s Technical Support Instrument and developed by NOVA IMS in collaboration with PlanAPP, Portugal’s policy and foresight unit, the AI4IA initiative created a pioneering AI-based proof of concept. This tool automates the extraction, classification, and analysis of legislative burdens. With cutting-edge Natural Language Processing and deep learning models, it scans legal documents to detect where and how new laws impose obligations on businesses and citizens.
The tool does more than flag issues—it quantifies economic impact, compares different versions of laws (e.g., EU directives vs. national transpositions), and supports human experts through continuous feedback loops. It’s a hybrid AI system: the machine learns from the legal experts, and the legal experts benefit from the machine’s speed and consistency.
The impact for science is significant: AI4IA demonstrated how advanced machine learning models, such as custom-trained BERT-based architectures, can be effectively fine-tuned for the legal domain. The project generated model pipelines and benchmarking methods now available for the broader research community. It also fostered interdisciplinary collaboration between legal scholars, data scientists, and policymakers, creating a research environment at the intersection of law, AI, and public administration.
But the real transformation is for citizens, businesses, and institutions. AI4IA has already shown that it can reduce the time of analysis from up to three workdays to under 15 minutes per document—with 92.1% recall and 100% precision when combining AI with human validation. In practical terms, this means legal experts can shift focus from mechanical document analysis to thoughtful evaluation and policy refinement, enhancing the overall quality and relevance of legislation.
For citizens, this translates into clearer, more proportionate, and more transparent laws. By systematically identifying and categorizing administrative obligations, AI4IA ensures that excessive or unnecessary burdens—especially on individuals or vulnerable groups—are flagged early. The result is a legislative process that feels more understandable and fairer, increasing public trust in institutions.
For businesses, particularly SMEs, the impact is equally profound. Legislative complexity and gold-plating often create unseen compliance costs that hamper competitiveness. AI4IA identifies these layers of administrative burden, enabling regulators to streamline obligations and remove unnecessary friction. In the words of the Portuguese Business Confederation, the tool directly supports their priority of regulatory simplification, making Portugal’s legislative environment more innovation- and investment-friendly.
For governments and policymakers, AI4IA delivers a data-driven decision-support system. The tool enables not just faster assessments but also deeper insights, allowing ministries to understand the cost implications of legislative proposals across different sectors and company sizes. This facilitates evidence-based governance, where policies are shaped by real-world data, not assumptions.
The impact is quantifiable and validated. Stakeholder confidence in the legislative process rose from 50% to 80% after exposure to the tool. User satisfaction was rated at 92%, and legal experts at PlanAPP confirmed that they could confidently use the system despite not having AI expertise—demonstrating the accessibility of the technology. Most importantly, PlanAPP highlighted that the tool enabled them to anticipate legal risks, make timely interventions, and better align laws with real-world outcomes.
Compared to the situation before AI4IA, where legislative assessment was fragmented, subjective, and slow, the new approach is systematic, transparent, and fast. What used to take days of reading, tagging, and manually comparing texts is now reduced to minutes, with AI surfacing the most relevant content and guiding expert judgment. The result is not merely improved productivity—it is structural transformation of how governments craft and assess laws.
Finally, AI4IA’s design is modular and scalable. While first implemented in Portugal, it is already being positioned as a blueprint for replication across EU Member States. Because the system relies on open-access EU legislation and follows interoperable standards, retraining the models with local data is straightforward. This makes AI4IA a promising European public innovation, showing how AI can serve not only science but democracy and the public good.
AI4IA shows that artificial intelligence can not only accelerate legislative analysis but also make it fairer, more transparent, and evidence-based, bringing technology closer to public service and democracy.
Mauro Castelli
