José Borja is the Director of the Tax IT Department of the AEAT, the architect of data governance for one of the most sensitive and critical organizations in the country. At the intersection of artificial intelligence, the modernization of legacy systems and cybersecurity, Borja manages the technological strategy that defines the relationship between the administration and millions of taxpayers.
The Tax Agency is in the midst of digital transformation, a pillar of its 2024-2027 Strategic Plan. This plan identifies AI as an essential transformative technology to improve efficiency. The agency’s challenge is twofold: on the one hand, to modernize its systems to process enormous volumes of data and, on the other, to do so while guaranteeing rigorous respect for the legal system and the protection of citizen rights.
In this interview, Borja explains his vision: the real challenge is not only technology, but ensuring easier and fairer tax compliance. It addresses the pillars of its AI strategy, based on unwavering human supervision, the tactical shift from inspection to prevention and the fundamental challenge of attracting talent.
(MCPro) What do you consider to be the main challenge in the transformation of the AEAT: the rapid evolution of AI, the modernization of legacy systems or the data governance itself?
(José Borja Tomé) The three challenges you mention are important, and that is why we work on all three intensely. However, the real challenge in the transformation of the AEAT is to take advantage of the available capacities and resources at all times to ensure easier and fairer tax compliance, and very particularly through the use of data.
(MCPro) In a scenario of accelerated digital transformation, where do you place the main challenge: in cybersecurity or in attracting specialized talent to manage these new tools?
(José Borja Tomé) These are two basic pillars for data management. The challenges of cybersecurity do not stop growing and have become one of the main concerns, but… without people we have nothing to do. Unfortunately, currently recruiting technological talent in public administrations is complex, and the Tax Agency is no stranger to this situation.
(MCPro) What are the pillars of the AEAT’s AI Strategy and how does its Strategic Plan 2024-2027 prioritize data governance?
(José Borja Tomé) The AEAT’s AI strategy focuses on the security and protection of citizen rights, rigorous respect for the current legal system, and the principles and values that govern the actions of the AEAT, with a special emphasis on human supervision. Automated administrative actions will in no case rely exclusively on the result obtained from an AI system.
An example of the priority that the AEAT gives to the safe use of AI is that the 2024-2027 Strategic Plan indicates that the Tax Agency is committed to the use of AI as an essential transformative technology for the evolution and improvement of effectiveness and efficiency in achieving objectives, especially in terms of information and assistance to the taxpayer and in the field of prevention, but also in the fight against tax and customs fraud, always respecting the commitments and principles before pointed out. In this context, data governance is essential to ensure that information is available where it is needed and that it is always used appropriately.
(MCPro) The AEAT operates on an on-premise Data Lakehouse architecture. How do you balance data sovereignty with agility? Are you considering a future evolution towards a Data Mesh model?
(José Borja Tomé) Our analytical system is built directly on a unified repository in which all the information is located, and there are no information silos. In this situation, in which we already have a monolithic information system, we do not perceive a real advantage in addressing a transition towards a Data Mesh philosophy.
On the other hand, our data architecture makes available to any user who needs it, in online self-service mode, all the information they need quickly and easily, without the intervention of technical personnel. Furthermore, the level of integration achieved between analytical and operational systems allows decision makers to act directly by basing their actions on data. More than 16,000 AEAT employees are active users of our analytical system.
(MCPro) In what areas of the fight against fraud is machine learning being applied with greater intensity, beyond “asset analysis”?
(José Borja Tomé) We are using AI with greater intensity in terms of information and assistance to taxpayers and in the field of prevention, but also in the fight against tax and customs fraud. For example, if a predictive model allows us to predict whether a taxpayer may be making an error when completing a return, we prefer to use this information to notify them at the time, rather than waiting until they have submitted their return to initiate a verification procedure, which is more costly and detrimental to everyone.
(MCPro) How has taxpayer assistance automation evolved from the Income Tax draft to the new cognitive virtual assistants?
(José Borja Tomé) Taxpayer assistance is one of the backbones of our Strategic Plan. The Tax Agency is making an enormous effort to improve the information and assistance it provides to citizens to facilitate voluntary compliance. In this effort, all technological tools come into play, from tax models and language simplification, to virtual assistants. We have also reorganized the way in which services are provided in person and remotely, extending hours and standardizing service levels and tools.
(MCPro) The AEAT has created new models (172, 721) for crypto assets. Are these regulations data engineering tools designed to acquire the “raw material” that fuels your Big Data systems?
(José Borja Tomé) The new models for cryptoassets do not have any particularities in this sense. Tax models are sources of information that the Tax Agency uses to fulfill its purposes. The data collected in the tax models is then used in accordance with the AEAT’s data and operational strategy.
(MCPro) The Ethical Commitment of the AEAT guarantees “human centrality”. Does this mean that every taxpayer selected by an AI for inspection is guaranteed a meaningful human review prior to notification?
(José Borja Tomé) The automated administrative actions dictated by the Tax Agency will in no case rest exclusively on the result obtained from an AI system. Human intervention is always guaranteed to supervise, validate or even veto the options proposed by the system. Ultimately, all decisions will be made by people.
Furthermore, human surveillance occurs throughout the life cycle and not only at the moment of inference. We apply this approach human in the loop in all phases: system design, data collection and preparation, model training and validation, and maintenance and continuous improvement.
(MCPro) How are your risk selection algorithms audited to prevent algorithmic bias and avoid ossifying historical inspection biases?
(José Borja Tomé) All our artificial intelligence projects are subject to a methodology that ensures compliance with our strategy, which prevents the risk of chronic biases. Special attention is paid to the quality, relevance and representativeness of the training data and the absence of biases that could lead to discriminatory performance is verified. In any case, we must not forget that inspection is not the priority area of AI development in the AEAT.
(MCPro) How is the conflict between the taxpayer’s “right to the algorithm” (transparency) and the need to protect the effectiveness of their anti-fraud tools managed?
(José Borja Tomé) Respecting in all cases article 95 of the General Tax Law, which establishes the confidential nature of data with tax significance. Furthermore, we guarantee that the dictated acts are adequately motivated, so that the taxpayer’s right to defense is ensured.
(MCPro) What is the relationship between the AEAT and the AESIA? What will the Agency’s “high-risk” AI models be monitored and audited?
(José Borja Tomé) At the moment we have not developed high-risk models in the AEAT, and the AESIA has not yet told us what the supervision and audit process will be like. In any case, our methodology allows us to ensure compliance with current regulations and be prepared for any audit.
(MCPro) The Comprehensive Digital Assistance (ADI) model has received criticism for the digital divide and staff overload. How do you balance digital efficiency with ensuring assistance to vulnerable taxpayers?
(José Borja Tomé) Providing quality services also in a human way, both in person and remotely, something in which the Tax Agency has been implementing constant improvements.
(MCPro) What is the AEAT’s strategy to attract and retain elite talent in data science and cybersecurity, competing with salaries in the private sector?
(José Borja) The selection processes for technology personnel are currently carried out by a body outside the AEAT. In any case, the organization and content of the work carried out at the AEAT make it possible to offer an attractive professional environment.
(MCPro) Being a Critical Operator, What are the most significant cybersecurity threats faced by the AEAT Cybersecurity Center?
(José Borja Tomé) In such a large and complex organization, and with so much data and external interactions, you can imagine that the number and variety of threats is enormous.
(MCPro) How is interoperability and secure data exchange guaranteed between the AEAT and other administrations (Social Security, CCAA) without compromising privacy?
(José Borja Tomé) Each data exchange must have legal support and is normally articulated through agreements in which the guarantees and responsibilities of the parties are defined.
(MCPro) What emerging technologies (generative AI, open source web scraping) are you actively exploring for future applications in monitoring or assistance?
(José Borja Tomé) Certainly generative AI can open new opportunities by offering the possibility of making people’s jobs easier. But as in all cases, we do our exploration ensuring the governance of the technology and respect for the guarantees that we have been discussing.
