How Finalyse can help
AI Fairness Assessment
Responsible AI is a serious concern for European regulators. AI drives innovation by enhancing decision-making and efficiency across all industries, finance makes no exception. However, challenges like data privacy and bias have led European regulators to ensure AI prioritizes fairness, human rights, and accountability, fostering responsible use while minimizing risks. In a nutshell, this is the spirit of the EU AI Act (enforced in August 2024) that wants to ensure a safe use of AI systems.
The adoption of the EU AI Act highlights the increasing regulatory focus on ensuring AI models are ethical, fair, and transparent. Financial institutions face mounting pressure to assess and demonstrate the fairness of their AI models, particularly in areas such as credit risk scoring, transaction monitoring, and customer interaction. Companies are expected to comply with EU AI Act by August 2026.
The European Commission will most likely expect all financial organisations to integrate in their risk management framework and ESG reports an AI Fairness assessment and support in AI conformity assessment.
How does Finalyse address your challenges?
Governance
Risk-level identification of AI risk system and integration of responsible AI report into the risk management framework and systems.
Perform AI conformity assessment
Ensure that the use of AI systems is lawful, ethical and technically robust.
Monitoring and maintenance
Performing periodical reviews and updates of the AI conformity assessment report.
How does it work in practice?
Finalyse offers a comprehensive AI model fairness assessment service aligned with the requirements of the EU AI Act.
Our service combines both quantitative and qualitative evaluations to assess key metrics such as fairness, bias, transparency, and accountability, while ensuring full regulatory compliance.
We support clients in various aspects of preparing for the EU AI Act conformity assessment, including:
- Identifying risk levels for AI systems
- Implementing AI governance frameworks
- Performing gap analyses and recommending improvements
Our methodology is aligned with the CapA conformity assessment procedure (as outlined in capAI.pdf), a structured tool designed to audit AI systems. Finalyse has previously applied similar internal review protocols for several clients.
The assessment process is structured into the following key stages:
Key Features
- Enhanced confidence in the fairness and compliance of AI models.
- Mitigation of reputational and regulatory risks associated with AI biases.
- Improved alignment with ethical AI principles, fostering trust among stakeholders.
- A competitive edge in meeting and exceeding EU AI Act requirements.
- Customised support from experts familiar with both regulatory frameworks and AI technologies.
Nemanja Djajić is a Managing Consultant with more than 12 years of experience in credit risk modeling and data science area. He gained his experience through multiple roles in banking industry including the position of data science department director in one of the biggest banks in Eastern Europe. Nemanja’s main area of expertise lies within the development and validation of risk and business related models, using traditional or machine learning methodologies.
Marino is a Senior Quantitative Risk & Machine Learning consultant at Finalyse. He is a qualified actuary with 7 years of experience. He has successfully delivered several projects for banks and insurance companies including a conversational chatbot, AI-powered price optimisation engine, Deep Learning time-series forecast models, challenger software for Credit Risk, AML, and PEP models for traditional and fully digital banking. His expertise extends to non-life insurance pricing, risk modelling, and regulatory frameworks such as IFRS and Solvency II. He is proficient in Python, Linux, Docker, Flask, GCP, Azure, AWS, and various other tech stacks.
Maciej is a Managing Consultant with 4 years of experience in credit risk model development and validation. Maciej has extensive knowledge of Pillar 1 and 2 related models gained on projects completed for clients across Europe and the Middle East. He was a vital part of projects focusing on Credit Risk Economic Capital model development, AIRB LGD model development, but also PD, LGD & CCF models validation under the IFRS 9 context.
Gergely Tréfa is a Senior Consultant at Finalyse with more than 5 years experience in credit risk model development, validation and data science. He has extensive experience in developing and validating PD, LGD, EAD models under IRB and IFRS frameworks. Gergely holds a Master's in Quantitative Finance and is proficient in R, Python, and SQL. In addition to regulatory modeling, he has developed non-regulatory machine learning models, such as collection and prepayment models, and has applied NLP techniques to automate risk reporting processes.