(AGENPARL) – Fri 21 November 2025 Leveraging technology: The Bank of Italy’s SupTech Journey
Giuseppe Siani
Head of Regulatory and Supervisory Directorate of Banca d’Italia
First Supervision Conference of the Central Bank of Argentina
‘Challenges of a Transforming Financial System
Buenos Aires, 13 November 2025
Introduction
I thank the organizers for inviting the Bank of Italy to this event where we can
discuss topics that are highly relevant for both supervisors and market stakeholders.
The financial sector is undergoing a profound transformation, led by the rapid and
strategic adoption of digital technologies. Digitalization has indeed become a strategic
driver for banks, by enabling them to enhance operational efficiency, strengthening
their risk analysis and management, improving service quality and deepening customer
relationships, thus fostering their business model sustainability in an increasingly complex
market.
Artificial intelligence (AI) plays a key role in explaining this market trend given that
it materially affects digital strategies. Based on our evidence, AI adoption is accelerating,
primarily driven by Generative AI (GenAI), which is expected to soon surpass more traditional
AI techniques, thus opening new frontiers in data analysis, customer interaction, and risk
management.
AI has also crossed a critical threshold: from specialized applications to a pervasive,
general-purpose technology, that thanks to language models engages directly with human
reasoning and fosters more efficient summarization, interpretation and decision support.
This marks a turning point for both financial institutions and supervisory authorities:
our challenge is to evolve alongside the financial sector, leveraging technology to simplify
supervisory processes, enhance analytical capabilities, improve risk detection and preserve
intrusive and forward-looking supervision.
Today, I would like to focus on three main topics: technology potential to enhance
supervisory activities; the relevant risks and challenges that supervisory authorities are
called upon to address; the Bank of Italy’s concrete experience with SupTech tools in
prudential supervision.
Technology potential – the Supervisory Perspective
Effective banking supervision requires complex and interdependent assessments to
identify traditional and emerging risks stemming from the digital transformation of the
banking and financial sectors. These include risks related to value chain fragmentation,
increased interconnections and operational dependencies across the financial ecosystem.
Supervisory activity should therefore follow an integrated approach and be carried
out through a sort of circular process where different tasks1 interact with each other to
develop the necessary technical analysis of different risk profiles and the appropriate
supervisory options. For example, experience gained through off- and on-site supervision
facilitates the identification of new risks and the deeper understanding of market dynamics,
which, in its turn, supports the necessary methodological reviews and influences both the
international and the domestic regulatory processes.
Indeed, supervisory activities rely heavily on robust data analysis, drawing from
long-standing data sources such as credit registers, regulatory reports, financial statements
and targeted surveys. In addition, supervisors might leverage new and unstructured data
sources, such as information drawn from the web, social media, and real-time market
data. Moreover, innovation could improve supervisors’ ability to massively extract insights
from PDF documents and images, enabling broader exploitation of valuable documents
(for example, board meeting minutes and other internal records), that often could remain
underutilized. I will come back to this issue later.
Last, but surely not least, technology enhances supervisory capabilities by fostering
the automation of routine tasks, unlocking deeper insights into existing analyses and,
ultimately, helping faster and more informed decision-making.
Technology-related risks and challenges
Innovation, however, is not without risks. As we integrate advanced technologies
into supervision, we face a range of challenges that must be addressed, such as data
quality and interoperability, legal and ethical considerations, particularly those related
to AI accountability and transparency, given the potential “black box” nature of more
advanced algorithms.
To this end, two fundamental principles should guide supervisors’ approach:
keeping “human in the loop” and preserving accountability. As for the former, innovation
must be embraced in a way that reinforces, rather than replaces, human judgment.
It is therefore more appropriate to use the term “AI‑assisted” rather than “AI‑driven”
supervision: AI supports informed, timely, and responsible decisions, strengthening
– rather than overshadowing – the human dimension that underpins trust and integrity
in financial oversight. Therefore, artificial intelligence, no matter how advanced, cannot
replace the needed expertise and personal sensitivity of skilled supervisors.
Specifically, rules and analytical methodologies developments, monitoring and control activities,
horizontal risk identification and supervision.
As for the latter, the ability to track and document data, resources, processes and
decisions supported by AI systems, as well as understand how they reach any specific
decision, are a key prerequisite to build both trust and accountability of our work. Lacking
that, we cannot be sure that our outcomes meet standards of fairness, compliance, and
appropriateness, potentially undermining our institutional mission.
In order to mitigate these risks, we have adopted a few concrete measures that try
to strengthen our internal governance, while embracing digital transformation: first of
all, we have established a clear top-down governance mechanism where a high-level
Steering Committee, composed by senior representatives from all involved departments,
acts as a hub to oversee AI initiatives, promotes and coordinates their adoption, and thus
ensures the necessary consistent and conscious innovation across our Institution.
This governance framework is complemented by the review of our internal policies
and guidelines to ensure the responsible use of AI in banking supervision, aimed at
preserving confidentiality, security, and transparency. It is also supported by internal
dedicated communication to inform users of AI systems about not only their strengths,
but also their limitations, and potential risks. I will elaborate on our internal approach
further later.
The Bank of Italy’s Suptech experience for prudential supervision
Financial supervisors all around the world are exploring how to leverage technology
to enhance their activities. For example, over the past years, the Single Supervisory
Mechanism (SSM) has increasingly taken advantage of innovative technology in its
supervisory work, which the Bank of Italy has also contributed to.
As far as our internal process is concerned, we have implemented a comprehensive
and coordinated supervisory approach to risk assessment: it involves integration of
findings and measures from both on-site and off-site activities carried out by ‘vertical’
supervisory units with insights from ‘horizontal’ market-wide assessments and
benchmarking. We also extend our responsibilities beyond banks and other financial
intermediaries (e.g. asset management companies, servicers and payment institutions)
to include third parties (for example IT outsourcers), thus promoting a broader holistic
risk detection and assessment.
To support this activity, the Supervision Department has embarked on a structured
“SupTech journey”, which has so far fostered integration of advanced technologies into
four selected prudential supervisory processes (three focused on off-site supervision and
one on on-site supervision).
The first tool focuses on automating fit and proper assessments (FAP); each year
we manage a few hundreds of candidate assessments that must be processed within a
pre-defined timeframe; in the past, they were carried out manually by accessing multiple
databases, thus making it highly time-consuming. Our tool allows for the automated
extraction of data from both internal and external databases and incorporates reputational
screening through AI-assisted analysis of press information.
Figure 1 highlights the improvements achieved following its implementation and
helps assess the concrete benefits in terms of time reduction for data extraction (i.e. not
in terms of overall time needed). Therefore, this tool mainly helps analysts on repetitive,
time-consuming, low value-added tasks and might improve qualitative assessment
through a sentiment analysis on appointee’s reputation based on Natural Language
Processing (NLP) techniques applied to global news sources2. This functionality draws
the analysts’ attention to news that could negatively impact candidate’s reputation, thus
enhancing the overall quality of the available information.
Figure 1
Time (minutes) needed for data extraction related to one FAP assessment
Source: Bank of Italy internal simulation.
The second tool relates to the assessment of ownership structures of supervised
entities, which helps streamline those cases characterized by lengthy and complex chains
of control. Thanks to Automated Reasoning (AR) technology applied to a Knowledge
Graph3, not only it integrates data drawn from multiple sources, but it can also infer the
hidden relationships that are not explicitly stated in the original datasets, thus providing
a deeper understanding of the underlying control structures. It also helps our supervisory
analysts to elaborate their assessments further through “what if” analyses4.
Factiva is a business information and research tool owned by Dow Jones & Company, which provides
organizations with search, alerting, dissemination, and other information management capabilities
(more than 32.000 sources such as newspapers, journals, magazines, television and radio transcripts,
photos, etc.), sourced from nearly every country in the world in 28 languages. World-Check is a risk
intelligence platform used by banks, insurance companies, and other institutions to support due
diligence processes, anti-money laundering (AML), know your customer (KYC), and sanctions screening
activities.
A Knowledge Graph (KG) is a data model particularly suited to deal with domains characterized by
the presence of very large networks of entities having complex interconnections. It consists of a data
source, typically a “graph database”, enhanced and augmented with new knowledge generated with AI
techniques (automated reasoning), by combining such sources and the available business experience.
A scenario simulation that allows to implement hypothetical changes by adding, removing and
modifying existing participations.
The tool is intuitive and interactive; it allows to browse, filter, expand nodes and
obtain detailed information, including the original data source. Figure 2 shows three
examples of graph representations that are generated by the tool. The first one is a
basic representation of direct participation in a supervised entity owned by different
stakeholders. The second graph highlights indirect participations and control relationships,
with the latter being identified by the orange lines. The last example shows a relation which
is not included in the original data source and has been calculated via automated reasoning.
This information is represented by the red dotted line in order to immediately alert analysts
that it has been derived by the AI module and therefore needs additional validation by the
competent analysts.
Figure 2
Examples of charts generated by the ownership structures tool
The third tool, recently developed, relates to the area of corporate governance; in
particular, it supports deeper and more extensive analysis of banks’ board minutes, which
is traditionally performed during many supervisory controls (for example, during on-site
inspections or thematic reviews). The tool indeed allows to identify specific Board members’
taking the floor, track both the frequency and the tone of relevant interventions, as well as
categorize the underlying discussion topics.
Therefore, it might enable supervisors to better assess boards’ factual functioning and
conduct thematic reviews on key issues such as internal controls, risk management and
governance practices. It also improves benchmarking on qualitative aspects of governance,
which promotes stronger interactions of our supervisors with corporate officers and,
ultimately, enhances the board members’ role.
Figures 3 and 4 present a hypothetical analysis designed to demonstrate the capabilities
of the tool. This example compares the board of a single bank with those of other institutions,
which act as a proxy for the broader domestic banking system; in particular, the first pie chart
shows a lower percentage of dissent/challenge interventions, complemented by a higher
percentage of acceptance interventions. The second pie chart shows a lower percentage
of non-independent non-executives’ interventions for the bank, compared to the system.
Sharing this insight with the bank may contribute to raise awareness of the concrete
dynamics within the board and support the adoption of appropriate actions.
Figure 3
Percentage of interventions by board members by tone and function
Figure 4
Proxy of banking system’s percentage of interventions by tone and function
The fourth tool in our SupTech suite is designed to support on-site inspectors
while drafting their findings and supporting the overall process of consistency checks
across different on-site reports. In particular, the tool includes two application services:
the first one provides AI-generated suggestions related to the findings being drafted
or reviewed, while the second one allows to classify, search and consult by topic the
applicable regulatory standards that must be mandatorily referred to in order to properly
support the finding themselves.
This tool therefore supports the drafting process and helps enhance the accuracy
and consistency of findings, while ensuring the secure and reliable handling of sensitive
information included in the on-site reports, and preserving the necessary confidentiality
throughout the process.
Despite their tangible benefits for the Bank of Italy’s supervision works, these
tools require our continuous risks monitoring and concrete mitigation efforts. Let me
share four examples.
First, to ensure accurate and reliable outcomes both during the test and the
adoption phases, we plan periodical reviews of our AI models even though they might
create inefficiencies given that such activity is resource and time consuming. Second, we
are implementing safeguard that helps mitigate operational risks, related for example
to internal and external unauthorized access to supervision data and systems.
Furthermore, every AI-generated output undergoes the necessary validation by
the competent supervisory analyst, whose professional judgment remains key to the
process. This approach helps ensure that supervisory decisions are grounded in human
reasoning and oversight, in line with the “human in the loop” principle, as stated
previously.
The fourth mitigant measure relates to the integration of different skill sets through
multidisciplinary teams: combining business expertise with specialized technical
knowledge – supervisors, data scientists, engineers, and legal experts – demands
strong coordination to ensure that every project delivers both regulatory compliance
and technological innovation.
In this regard, the Bank of Italy is acting on two levels: recruiting specialist profiles
and providing continuous technical and ethical training. In recent years, we have
invested significantly in experts in innovative areas, such as artificial intelligence and
blockchain, as well as cybersecurity and management engineering that are not always
available in today’s labor market. In addition, we are committed to strengthening the
digital literacy and the ethic profile of our staff, as the future of supervision depends
on human capabilities as much as on technology, and requires adaptability, curiosity,
as well as ongoing commitment to continuous learning.
While these measures significantly reduce risks, we remain aware that AI
technologies evolve rapidly and carry uncertainties that require ongoing oversight.
Conclusions
Integrating innovation into banking supervision cannot only imply upgrading ICT
systems but, rather, represents a strategic priority to keep pace with the rapidly evolving
financial and technological landscape.
The question is therefore no longer whether to use technology, but how to use it
effectively, in order to strengthen our internal processes by improving efficiency, accuracy
and gain additional qualitative insights into the risks faced by supervised entities.
This represents a continuous effort that requires addressing and monitoring the
evolving challenges and risks, particularly those stemming from the AI. To this end,
supervisory analysts must remain at the center of our strategies, not only by encouraging
people to apply critical thinking and verify results through independent sources, but also
fostering learning, and skills development and integration.
SupTech solutions also create opportunities for broader cooperation given that
supervisors worldwide perform similar tasks and face comparable challenges, including
resources constrains; it is therefore essential for supervisors to leverage synergies,
learning from each other by sharing concrete experiences and solutions. This collaborative
approach will help reduce duplication of efforts, accelerate implementation and ultimately
enhance our overall supervisory effectiveness.
Designed by the Printing and Publishing Division of the Bank of Italy