The European Artificial Intelligence Regulation (AI Act) is now applicable. For credit scoring players, this is a major change that imposes new transparency and explainability obligations.
What the regulation says
Credit scoring systems are classified as high-risk AI systems. This implies several obligations:
Decision transparency
Each score must be explainable. Which variables contributed to the decision? In what proportion? The evaluated company must be able to understand why it received a particular score.
Technical documentation
A complete technical file must be maintained, detailing the model architecture, training data, performance metrics, and bias control measures.
Human oversight
A credit decision cannot be fully automated. A human operator must be able to intervene and challenge the system's decisions.
Audit and traceability
Each decision must be traced and auditable. In case of dispute, the company must be able to reconstruct the system's reasoning.
How RocketFin complies
We have anticipated these obligations since 2024. Our platform provides:
- An explanation of the 5 main variables for each score - A complete evaluation history - An interface allowing operators to validate or challenge scores - Complete technical documentation accessible to regulators
Pitfalls to avoid
The black box
Models too complex to explain are no longer compliant. The right balance between predictive performance and interpretability must be found.
Lack of documented training data
If you cannot prove that your training data is representative and unbiased, you are in violation.
Over-automation
Automating 100% of credit decisions may have been efficient, but it's no longer legal. Human oversight is mandatory.
Conclusion
The AI Act is not a constraint, it's an opportunity. Companies that embrace transparency strengthen the trust of their customers and partners. Those who try to circumvent the rules face significant penalties.