AI Explainability: CRIF and UNIMORE Transform AI Decisions into Transparent, Actionable Insights

As artificial intelligence becomes increasingly embedded in credit risk assessment, lending decisioning, and risk management processes, financial institutions face a growing challenge: ensuring that AI-driven decisions are not only accurate and efficient, but also transparent, explainable, and trustworthy.

To help address this challenge, CRIF has partnered with the Data and Text Analytics Lab of the Artificial Intelligence Research and Innovation Center (AIRI) at the University of Modena and Reggio Emilia (UNIMORE) on a research initiative focused on advancing AI Explainability and transforming complex model outputs into clear, actionable insights that can support real-world decision-making processes.

This collaboration demonstrates how research and industry can work together to make AI systems more transparent, strengthen governance frameworks, and support the responsible adoption of AI across the financial services sector. The outcomes of the project have already contributed to the development of CRIF ORCHESTRA, CRIF’s AI-powered decision intelligence and orchestration platform.

Enhancing Transparency in AI-Powered Credit Decisioning

AI enables financial institutions to assess risk more effectively, identify complex patterns in data, and support faster, more consistent lending decisions. However, as AI models become increasingly sophisticated, understanding how and why a model arrives at a specific outcome is becoming just as important as the outcome itself.

Today, transparency is a critical requirement for organizations seeking to strengthen AI governance, improve model risk management, meet evolving regulatory expectations, and build trust with both customers and stakeholders.

Within this context, the CRIF-UNIMORE project focuses on counterfactual explanations, one of the most advanced techniques within the field of explainable AI.

From Model Explanations to Actionable Guidance

Unlike traditional explainability approaches that describe the factors influencing a decision, counterfactual explanations identify realistic changes that could lead to a different outcome.

In a lending scenario, this means moving beyond explaining why a loan application was declined and providing practical insights into what could improve the likelihood of approval.

For example, a counterfactual explanation may suggest:

• Reducing the requested loan amount
• Adjusting the repayment period
• Lowering existing repayment commitments
• Modifying specific application parameters considered by the model

By highlighting realistic alternatives, explainable AI becomes more than a transparency tool. It evolves into a practical decision-support capability that helps institutions better understand model behavior while enabling more informed interactions with customers.

“As the Data and Text Analytics Lab, we have worked with CRIF on counterfactual explanations, which seek to clarify which changes in the input data could have led the model to a different decision. In credit, this means identifying realistic variations in the factors considered by the model that can change the outcome of an assessment. The challenge is not only to generate these explanations, but also to understand when they are genuinely useful by examining their plausibility, feasibility, stability, robustness, and consistency with the model’s behavior. Our collaboration with CRIF allows us to address these aspects through real-world use cases.”

Francesco Guerra, Professor at the Department of Engineering “Enzo Ferrari”, University of Modena and Reggio Emilia

Bringing Explainable AI into Operational Decisioning

The outcomes of this research are being translated into practical decisioning capabilities within CRIF ORCHESTRA that make AI-driven insights more transparent, understandable, and actionable. By embedding explainability directly into operational workflows, organizations can gain greater visibility into model behavior while strengthening governance, improving model lifecycle management, and supporting regulatory readiness.

This integration enables lenders to combine advanced AI capabilities with business rules, governance frameworks, and human oversight, ensuring that decision-making processes remain transparent, accountable, and aligned with organizational objectives. The initiative also demonstrates how explainability can be operationalized at scale, transforming advanced analytical models into trusted decision-support tools for business users.

“This project represents a concrete step in CRIF’s journey to move AI explainability from research into practical application within decision-making processes. The results of this project are already visible within CRIF ORCHESTRA, making model insights more understandable, actionable, and governable. For lenders, this means supporting faster and more consistent credit decisions, strengthening risk control and regulatory readiness, and enabling more transparent customer interactions that build trust.”

Marcello Nadalini, Executive Director and Head of Solutions and Digital Services at CRIF

Creating Business Value Across the Credit Lifecycle

The benefits of explainable AI extend far beyond individual lending decisions.

By providing greater transparency into how models operate, organizations can:

• Increase transparency and understanding of AI-driven credit decisions
• Support alternative lending scenario analysis and portfolio risk assessment
• Strengthen AI governance, model validation, and regulatory compliance
• Enable effective human oversight and more transparent customer interactions

For example, customer-facing teams can use explainable AI insights to discuss realistic lending alternatives with applicants, creating a more transparent and constructive customer experience. At the same time, risk management teams can identify common factors driving approval or rejection outcomes across customer segments, enabling data-driven portfolio optimization and stronger risk controls.

AI Explainability as a Foundation for Responsible AI

For CRIF, explainability is a key pillar of responsible AI.

While machine learning models provide powerful analytical capabilities and predictive insights, decision authority continues to be governed by business policies, governance frameworks, risk controls, and human oversight.

As part of this approach, explainability acts as a bridge between advanced analytics and human understanding, ensuring that AI-driven innovation remains transparent, accountable, and aligned with both business and regulatory requirements.

As financial institutions continue to accelerate the adoption of AI, the ability to understand, govern, and trust model outputs will become increasingly critical.

The collaboration between CRIF and UNIMORE demonstrates how combining academic research with industry expertise can enable organizations to move beyond traditional "black-box" AI models and build more transparent, explainable, and trustworthy decisioning systems.

Looking ahead, CRIF will continue investing in research initiatives that advance trustworthy AI and support responsible innovation. Through platforms such as CRIF ORCHESTRA, these innovations can be rapidly transformed into scalable solutions that help financial institutions and insurance companies improve decision intelligence, strengthen governance frameworks, accelerate digital transformation, and deliver a more transparent customer experience.